The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
E March/ A p r il 1993 V ol. 75, No. 2 W Dimensions of Monetary Policy Essays in Honor of Anatol B. Balbach Proceedings of the Seventeenth Annual Economic Policy Conference of the Federal Reserve Bank of St. Louis THE FEDERAL j reserve: l{.V\k <t > A r ST. I j IS OI i F e d e ra l R e s e rv e B a n k o f St. L o u is R e v ie w March/April 1993 In This Issue . . . iii Contributing Authors v President’s Message Thomas C. Melzer vii Editor’s Introduction Michael T. Belongia xiii T ed Balbach: A n Appreciation Armen A. Alchian 1 M onetary Aggregates, M onetary Policy and Economic Activity Robert H. Rasche 36 Com m entary Julio J. Rotemberg 43 V iew s on M onetary Policy W. Lee Hoskins 56 Com m entary G eo rg Rich 61 Financial Regulation and the Com petitiveness of the Large U.S. Corporation Harold Demsetz 68 Com m entary Charles I. Plosser 71 A ssessing A pplied Econometric Results Carl F. Christ 95 Com m entary David A. Dickey MARCH/APRIL 1993 101 Commentary David Laidler 103 Real Exchange Rates: Some Evidence from the P o stw a r Years Allan H. Meltzer 118 Com m entary Pedro Schwartz 133 The Gold Standard, Bretton W oods and O ther M onetary Regimes: A Historical A ppraisal Michael D. Bordo 192 Com m entary Manfred J. M. Neumann All non-confidential data and programs for the articles published in Review are now available to our readers. This information can be obtained from three sources: 1. FRED (F ed eral R eserv e E conom ic Data), an electron ic b u lle tin b o a rd service. You can access FRED by dialing 314-6211824 through your modem-equipped PC. Parameters should be set to: no parity, w ord length = 8 bits, 1 stop bit and the fastest baud rate your modem supports, up to 14,400 bps. Information will be in directory 11 under file name ST. LOUIS REVIEW DATA. For a free brochure on FRED, please call 314-444-8809. FEDERAL RESERVE BANK OF ST. LOUIS 2. T h e F ederal R e se rv e B an k o f St. Lo u is You can request data and programs on either disk or hard copy by writing to: Research and Public Information Division, Federal Reserve Bank of St. Louis, Post Office Box 442, St. Louis, MO 63166. Please include the author, title, issue date and page numbers with your request. 3. In t e r -u n iv e r s it y C o n s o r t iu m fo r P o lit ic a l a n d S o c ia l R e s e a r c h (IC P S R ). M em ber institutions can r e quest these data through the CDNet O rd er facility. Nonm em bers should w rite to: ICPSR, Institute fo r Social Research, P.O. Box 1248, Ann A rb or, MI 48106, o r call 313-763-5010. iii Contributing Authors A rm en A. Alchian M an fred J.M. N eum ann Department of Economics University of California Los Angeles, CA 90024 Universitat Bonn Lennestrasse 37 D-5300 Bonn 1 Germany M ichael D. B o rd o C harles I. P lo sse r Department of Economics Rutgers University New Brunswick, NJ 08903 Department of Economics University of Rochester Rochester, NY 14627 C arl F. Christ R obert H. Rasche Department of Economics The Johns Hopkins University Baltimore, MD 21218 Economics Department Michigan State University East Lansing, MI 48823 H a ro ld Dem setz G eo rg Rich Department of Economics University of California Los Angeles, CA 90024 Schweizerische National Bank Borsenstrasse 15 8022 Zurich Switzerland D avid A. Dickey Julio J. R otem berg Department of Statistics North Carolina State University Raleigh, NC 27695 Sloan School of Management Massachusetts Institute Technology Cambridge, MA 02139 W. Lee Hoskins Ped ro Schwartz The Huntington National Bank Huntington Center Columbus, OH 43287 Antonio Maura, 7 28014 Madrid Spain David Laidler Department of Economics University of Western Ontario London, Ontario N6A5C2 Allan H. Meltzer Carnegie-Mellon University Graduate School of Industrial Administration Pittsburg, PA 15213 MARCH/APRIL 1993 V President’s Message It is with great pleasure that we dedicate these proceedings in honor of Anatol B. (Ted) Balbach and his years of distinguished service at the Federal Reserve Bank of St. Louis. Ted first joined the Bank as a visiting scholar in August 1970. After returning to his position as Dean of the School o f Business at California State Uni versity at Northridge one year later, he rejoined the Bank in August 1972 as a Senior Economist; in 1975 he became Senior Vice President and Director of Research, the position he held until his retirement on Oct. 31, 1992. The period of his tenure offered a wide spec trum of research topics for monetary economists, and Ted led the research staff to put the Bank’s imprint on most. In the early days, the stress of 4 percent inflation precipitated an ill-advised ex periment with peacetime wage and price con trols and the closing of the gold window; a move to freely floating exchange rates soon fol lowed. Those days also saw a Republican presi dent declare “W e are all Keynesians now!” at the same time serious discontent with that dom inant paradigm was being articulated. Building on the foundation already established by Homer Jones and notable work such as the Andersen-Jordan equation, Ted led the staff to write rigorous, but still readable, articles on controversial topics: the links between money growth and both nominal spending and the inflation rate; the link between rapid money growth and higher, rather than lower, nominal interest rates; the merits of flexible exchange rates; and the adverse consequences of uncon strained discretionary actions in monetary pol icy. Although this research agenda ran in par allel to one thread of the academic literature, it contributed in important ways to making the controversial issues of the 1970s the common ground for discussion in the 1990s. The proceedings o f this conference represent the main areas of research Ted emphasized dur ing his tenure at St. Louis. These interests, in turn, reflect his ongoing concern that a research department in a Federal Reserve Bank has to be accountable for its expenditure of taxpayer dol lars. As such, the research topics must be rele vant to the concerns of a central bank, and the lesson of the work must be understandable to a broad audience. Whatever recognition has ac crued to the Bank over time can be attributed in large measure to an adherence to these sim ple principles. Many staff members achieved great success working for Ted and moved on to important responsibilities elsewhere. This occurred be cause Ted insisted on high productivity from the Research staff, while seeing to it that they received proper credit for their work. Clearly, he played a key role in the recognition that the Bank received during his tenure and in the professional success of many who worked for him. Nonetheless, Ted, like Homer Jones before him, was content to be anonymous to most people. These proceedings are a small effort to shed that anonymity and acknowledge that many people owe a great debt to Ted Balbach. Armen Alchian's warm opening comments give much of the flavor of sentiments expressed by speakers and guests alike. Indeed, the authors o f the papers in this volume produced papers of uni formly high quality as a sign of respect for Ted’s high standards: They knew he would have been offended by a routine "festschrift” that stressed style and forgot the substance. Stu dents of a central bank's responsibilities should find something of interest in each of these contributions. Thomas C. Melzer President and Chief Executive Officer Federal Reserve Bank of St. Louis MARCH/APRIL 1993 vii Editor's Introduction The Federal Reserve Bank of St. Louis, held its first Economic Policy Conference in 1976, in con junction with the Center for the Study of Ameri can Business at Washington University. Although this annual event has grown over time it has maintained a focus on many of the principles that Ted Balbach fostered during his tenure as direc tor of Research. An open exchange of ideas was high on that list and an annual gathering of prominent academics to discuss important con cerns of central banking was but another way in Ted’s mind to move intellectual debates forward. With his retirement in October 1992, it was fit ting, indeed essential, that the beginning of a new chapter in his life be marked with a tribute to the legacy he has left for all of us. According ly, the papers in these proceedings reflect eco nomic and policy issues that were never far from Ted’s attention. The conference’s first session dealt with two is sues that, more than any others, dominated Ted’s career at St. Louis: the effects of money on eco nomic activity and the commitment of the central bank to a goal of price stability. Robert H. Rasche, in "Monetary Aggregates, Monetary Policy and Economic Activity,” investigates the large and un anticipated shifts in money velocity during the 1980s that led to large errors in predictions of inflation and growing sentiment that the demandfor-money function is unstable. Rasche’s paper ap proaches this task from three perspectives: con troversies of the 1960s and 1970s that have been resolved, empirical failures of reduced-form spend ing equations in the 1980s and the short-run ef fects of changes in the growth rate of the nomi nal money supply on economic activity. In reviewing the historical controversies, Rasche concludes that current mainstream macroeconomic thought embodies the important elements of the original Andersen-Jordan equation: changes in nominal government spending do not produce a permanent change in nominal income (or velocity) unless accompanied by a change in the growth rate of the nominal money stock. He also concludes that shocks to the level of velocity are permanent. The first point was contradictory to the dominant Keynesian paradigm of 25 years ago and the latter anticipated the now commonplace care that is given to proper differencing of the data and the problem of spurious regression relationships. Just as the world of monetary policy began to take St. Louis-type arguments seriously, the 1980s produced a sharp break in trend velocity that dis credited the Andersen-Jordan equation in the minds of many. In the second section of his paper, Rasche explores whether this break more likely reflects a rejection of the underlying economic relationships or a specification error arising from a Lucas-type structural change. Rasche concludes in favor of the latter, arguing that a sharp break in inflation ary expectations explains the break in trend veloc ity. Rasche also discusses how, if this explanation is correct, simple money growth rate rules for policy will be dominated by rules with feedback of the sort described by Meltzer and McCallum. In the last section of his paper, Rasche inves tigates a current controversy: whether changes in nominal money growth affect real output. After evaluating several vector autoregression (VAR) models, Rasche concludes that there is evidence in support of both permanent real output shocks (of the real business cycle variety) and permanent money growth shocks on real output. Thus though the role of money in explaining fluctuations in real activity is not rejected, Rasche finds room for contributions from other sources as well. In his commentary, Julio J. Rotemberg focuses on Rasche’s claims of finding a stable money de mand function. After estimating velocity regres sions in the spirit of Rasche’s analysis, Rotemberg finds that the apparently stable long-run specifica tion coincides with an "incredibly unstable” moneydemand function at shorter frequencies. He also finds that the residuals of such relationships are highly correlated. MARCH/APRIL 1993 viii As reasons for these unsatisfying results, Rotemberg renews an earlier call from our 1989 conference to use Divisia monetary aggregates in place of the conventional simple-sum meas ures. Although he only approximates a crude Divisia measure of M l, the large break in observed M l velocity in the early 1980s is reduced substan tially when based on data from a weighted mone tary aggregate. Thus other explanations beyond those offered by Rasche may explain the veloci ty puzzle or there may be no puzzle to explain. Rotemberg also extends Rasche’s VAR analysis with additional support for recent studies that have shown asymmetric effects of monetary shocks on output—negative shocks reduce out put but positive shocks do not increase output. This asymmetry and other asymmetries affect ing interest rates are left as important issues to be investigated in further work. With this backdrop on how velocity has be haved over time and how monetary policy apparently affects economic activity, W. Lee Hoskins offered a philosophical overview of how a central bank should conduct itself. In his "Views o f Monetary Policy,” Hoskins drew on his previous experience as president o f the Fed eral Reserve Bank of Cleveland and voting mem ber of the Federal Open Market Committee to criticize central banks for their tepid commit ment to the goal of price stability, if not their demonstrable bias toward inflationary policies. These flaws in the charters o f most central banks can be overcome, in Hoskins' view, only by stat ing a specific mandate to achieve price stability, giving the central bank the necessary indepen dence to achieve that goal and holding it ac countable for any failure to do so. With much of the academic literature focusing on technical issues (for example, interest rate vs. money stock targets) or public choice argu ments to explain central bank failures to achieve price stability, Hoskins advances “a simple and less elegant explanation...that central bankers are suffering from a Keynesian hangover.” The point is that, as products of a generation that learned an economic model in which central bankers should attempt to manage fluctuations in aggregate output, as well as inflation, modern central bankers are merely employing the train ing they acquired 20 or more years ago. Thus when the economy is weak, their vintage of training indicates a need for monetary stimuluseven if it ultimately will cause higher inflation. This view, which does not incorporate more re FEDERAL RESERVE BANK OF ST. LOUIS cent evidence on the dubious effects of monetary stimulus on output, also fits nicely with the views of elected officials more concerned with near-term issues, such as employment, than with the long term issue of inflation. In this environment, an in flationary bias by central banks is not difficult to understand and the broad reforms Hoskins sug gests are tied to the argument’s main themes. In his commentary, Georg Rich largely agrees with Hoskins’ statement of principles but won ders whether the gap from theory to practice can be bridged, and if so, how. Rich first argues that a mandate to achieve price stability, such as embodied in legislation proposed by Rep. Stephen Neal (D-N.C.), is not sufficient to achieve zero inflation. Instead, Rich argues that an oper ational rule, perhaps of the form proposed by Meltzer or McCallum, is needed to keep an ac cumulation of short run operational decisions by the central bank from wandering too far from a long-run policy that will maintain price stability. At the same time, Rich recognizes that the need to react to shifts in money demand raises some doubts about the desirability of the slow, mechanical paths of adjustment pro scribed by these rules. Rich raises other practical issues as well. Should the central bank, for example, ignore any effects o f changes in the real exchange rate or respond to appreciations and depreciations in the real exchange rate with equal zeal? More over, should—or can—the central bank disregard any and all real costs associated with a monetary policy consistent with the pursuit of price sta bility? Overall, Rich’s comments suggest areas where the practical elements of Hoskins’ broad proposal need to be specified. The first day’s afternoon session took a detour from a central bank’s monetary focus to address related, but often overlooked, themes. The first is the functioning of competitive markets; price stability, after all, is not pursued from religious conviction but rather from the notion that the market mechanism will allocate resources more efficiently if economic agents can be reasonably certain about the future purchasing power of money. The second detour addresses the issue of good econometric practice; because Ted be lieved that economic understanding would ad vance only after theories were confronted by the data and refutable null hypotheses were tested, he viewed good econometric practice as essential to the work of a central bank. ix On the topic of the market mechanism, Harold Demsetz presented his thoughts in “Financial Regulation and the Competitiveness of the Large U.S. Corporation.” In particular, he addresses the effects of regulation of capital markets on shareholder control o f corporate management. When diffuse ownership impedes stockholders from controlling self-interested corporate man agement and capital market regulations inhibit greater concentration of ownership, corporate efficiency can be impaired. The story is not quite so simple, however. First, stockholders enter agreements with man agement voluntarily and in full knowledge of potential conflicts of interest. Second, even though corporate ownership is diffuse, it is not so diffuse that owners have no incentive or power to monitor management. Third, though greater concentration o f ownership might en hance control of management, this is achieved at a cost of increased firm-specific risk. Citing findings of other studies, Demsetz re ports that the top five stockholders of U.S. cor porations own about one-fourth o f voting stock, whereas this share is substantially higher abroad: 50 percent or more in South Africa, 33 percent in Japan and similarly higher figures for Germany and Sweden. Demsetz argues that the higher concentration of ownership abroad can be at tributed to restrictions in the United States that prevent or limit banks, insurance companies and others from taking equity positions in U.S. corporations. With most U.S. corporate equity then coming from individual investors, what ef fect might this have on corporate efficiency? Demsetz clarifies the issues surrounding the separation problem of ownership and control by noting the difference between closed-end and open-end mutual funds. In the former, investor funds are converted to assets owned by the fund', thus a dissatisfied investor can sell his shares but, because he cannot force the fund to be the purchaser, there is no threat that poor performance will threaten management’s control of the fund’s assets. In an open-end fund, however, investors who withdraw funds also diminish the fund’s asset base. Moreover, it is important to note that this is different from the sale o f stock in an individ ual company where, although the share price might decline, the assets controlled by manage ment are unaffected. Thus by adding this new twist to the conditions necessary for the separa tion problem to be important, Demsetz suggests an ownership structure that can be quite diffuse while still exercising effective control over management. In his commentary, Charles I. Plosser notes that this paper, like many others by Demsetz, raises an issue that (to his knowledge) has been overlooked by others. And, though he encour ages efforts to assemble empirical evidence on the association between corporate ownership and regulation on the one hand and corporate performance and control on the other, Plosser has doubts that the issues are likely to be eco nomically important. Plosser's first doubt arises from his belief in the market mechanism and the ingenuity of in dividuals. Based on evidence from studies of the efficacy of other regulations and the typical response o f individuals to the opportunity of large rewards for evading regulations, Plosser’s instinct is that the costs of regulations on cor porate ownership are small. He devotes the remainder of his commentary to the notion of comparative advantage in in vestments. Some funds, for example, specialize in risk sharing and as a consequence limit their stake in any one firm. Not only is there no rea son to expect that the managers of this type of fund have an advantage in corporate control, but there are also suggestions that some of these funds are largely uninterested in corporate control. Conversely, other funds specialize in corporate control by taking large ownership po sitions in single firms and by so doing do not diversify risk for their investors. Specialized funds of these types, in Plosser’s view, are but one market response to distortions created by regulation of corporate ownership. At the same time, new regulation, such as the provisions in the Financial Institution Reform Recovery and Enforcement Act that limit bank holdings of high-risk securities, may create new distortions that are important for the efficiency of the mar ket for corporate control. Carl Christ’s paper, "Assessing Applied Econo metric Results,” offers both philosophical com ments on the desirable properties of econometric models and practical suggestions for evaluating real models against the standards of an ideal model. The standards for accepting or rejecting a model and the quality o f forecasts are discussed MARCH/APRIL 1993 X in some detail. Christ also offers brief comments on the more popular methods that have been applied to macroeconomic time series in recent years. Most of Christ’s points are illustrated by a re-examination of what he calls ‘‘an old, plainvanilla equation that still works, roughly”: Latane’s (1954) inverse velocity equation. Noting that the specification of Ml/GNP = a + b (inverse of long-term bond rate) has some of the proper ties of a money-demand function—negative in terest elasticity and income elasticity restricted, by construction, to equal one—Christ wonders whether the original equation is stable when so many money demand equations have exhibited substantial instability over time. In a variety of experiments, no demonstrable instability is found. Christ notes, however, that this specification has a number of undesirable characteristics in cluding strong positive serial correlation. Em barking on a number o f approaches to this problem, Christ employs strategies from the simple addition of an autoregressive term to the use of an error correction representation with partial adjustment parameters. He finds equa tions that fit much better but are terribly unsta ble over time. In doing so, he highlights the need for considerable judgment in addressing the important question of interest while resist ing the temptation to find models that have bet ter in-sample descriptive statistics. Two discussants offered comments on Christ’s paper: David Dickey on the suggested approach to econometric modeling and David Laidler, speaking for the “stochastically challenged" among us, on the economics of Christ’s chosen exam ple. Dickey agrees with Christ’s general thrust and adds a few new examples of subtle relation ships that are often lost in mechanical transfor mations of data. That the error term is multi plicative in a log specification of the Quantity Equation, and hence implies heterogeneity of variance in the untransformed data, probably has not been considered by most economists who have estimated reduced-form relationships of this sort. Nor is it always recognized that ex act relationships hold on some scales but not on others. Although these might be considered sim ple examples by some, Dickey’s point reinforces Christ’s entire theme of taking care with the economic specification and the raw data used to estimate it. FEDERAL RESERVE BANK OF ST. LOUIS Dickey also comments on Christ's evaluation of a forecasting model’s performance: Should a good model see a quadrupling of the root mean squared error across a forecast horizon o f eight quarters? At first glance, one might think that this is a reasonable standard. Dickey shows, however, that the probability o f such a quad rupling is high even if the true model is known. This sobering result suggests a continuing reli ance on judgment to supplement the informa tion in econometric forecasts. Finally, on a related point, Dickey notes that the simple bivariate velocity regression in Christ's paper can be dressed-up in the adornments of cointegra tion. But at heart, the main ideas are similar to those in the original Latane study. David Laidler applauds Christ for his reitera tion of a point made at least 25 years earlier calling for a test of models against data that were not available when the model was formu lated. Indeed, Laidler sees a full research agen da for applied econometricians who might investigate how a number of the classic equa tions of the literature fare when confronted by more recent data. If other relationships were found to be as stable over time as the Latane equation, we might come closer to some consen sus on the enduring long-run relationships that govern the behavior o f aggregate data. This view stands in counterpoint to Laidler’s reading of the money-demand literature and the philosophy behind its voluminous work. Much of this work has argued that the demand-formoney function is unstable and has done so with evidence on some instability in its shortrun dynamics. But Laidler argues that no one has yet modeled these complex short-run inter actions and, as such, we never had any reason to believe that we should be able to find a sta ble short run money-demand function. Thus it should not be surprising that more sophisticated attempts at modeling autocorrelation and other problems have produced models that are less robust than simple specifications of the long-run relationships for which we have a theory. The conference’s last session addressed topics in international economics of interest to Ted: flexible exchange rates, and the gold standard as a monetary policy rule. On the first topic, Al lan Meltzer notes that the theoretical case for flexible exchange rates can go either way: they may be a relatively low cost way of reducing the variances of other variables or they may be xi a source of excess burden. Surprisingly, how ever, little empirical evidence has been produced that permits comparisons of the welfare implica tions of alternative exchange rate regimes. Meltzer’s paper is directed to this end. After reviewing the case for flexible exchange rates as put forth by Friedman in 1953, Meltzer offers empirical evidence on several o f the key issues in the flexible vs. fixed exchange rate de bate. The first is the possible excess volatility and welfare burden of flexible rates. Looking at data since 1973 for both the Bretton Woods and flexible exchange rate periods and for flexible rate countries and the members of the Exchange Rate Mechanism (ERM) Meltzer finds that the variability of relative prices does not vary sys tematically across exchange rate regimes. He also finds no evidence to support the proposition that output is more variable under a fixed ex change rate system. Moving on to policy issues, Meltzer reminds us that Friedman’s 1953 work attributed a large role to rearmament in exchange rate determina tion (because it affects relative prices and the balance of payments) and distinguished between permanent and transitory changes in exchange rates. Incorporating the first idea into an equa tion for the real exchange rate, Meltzer reports that “contemporaneous changes in money and in defense spending are the principal factors keeping the predicted changes in step with actu al changes.” He also presents evidence against the common finding that the exchange rate is nonstationary. Overall, Meltzer finds much in his empirical evidence to support the main propositions in Friedman’s 1953 essay. Pedro Schwartz agrees with the thrust of Meltzer’s analysis and applies it to current de bates over monetary union in Europe. As the evidence indicates that real and nominal ex change rates move closely together and that ex change rate variability does not spill over into the goods market, an exchange rate objective does not seem to be an important or proper ob jective for a central bank. Indeed, with no abili ty to influence real exchange rates and un burdened of worries about spillover effects, Schwartz interprets Meltzer’s evidence as more support for directing a central bank's attention to the attainment of price stability. With regard to European Monetary Union, Schwartz sees the potential gains associated with a single currency and the lower transac tions costs of trade. He also sees, however, the drawbacks of another political institution, the European Central Bank, subject to varying de mands to pursue goals apart from price stabili ty. Rather than move to a stronger government institution, Schwartz believes competition be tween issuers o f money may lead to better re sults for consumers o f monetary services. Michael Bordo picks up many of the themes raised by Meltzer and Schwartz: the welfare consequences of alternative exchange rate re gimes, the insulating properties of flexible rates, rules vs. discretion in the conduct o f monetary policy, international policy coordination and the case for international monetary reform. In a wide-ranging treatment of each issue, Bordo both reviews the existing literature and offers new empirical evidence to investigate why some exchange rate regimes have been more success ful than others. On the question of performance, Bordo finds the Bretton Woods convertible regime of 1959/1970 to dominate all others examined; only the recent floating rate period comes close to achieving its level of performance. He also notes that the classical gold standard performed well as a nominal anchor but poorly in terms of the stability of real variables. Moreover, he argues that the gold standard was more durable than Bretton Woods because it worked as a contin gent rule and, as such, allowed the flexibility for governments to adjust to shocks. Bordo goes further than Schwartz in conclud ing from his evidence that monetary arrange ments that surrender monetary policy autonomy will not work over time. Because countries will not surrender this autonomy to another authori ty whose commitment to price stability they cannot trust, the key advantage of a flexible rate regime—the ability to pursue an indepen dent monetary policy—is still valued highly. The stresses within the European Monetary System in September 1992 only reinforce Bordo’s con clusion. Manfred J.M. Neumann, who found much to agree with in Bordo’s paper, first tries to enhance our understanding of Bordo’s VAR evi dence by supplementing it with a basic theoreti cal model. Although this exercise is frustrating in the sense that it identifies many unknowns confronting economists and policymakers, it is highly instructive as to where future research might be most profitably directed. MARCH/APRIL 1993 xii Neumann then spends the remainder of his commentary on the reasons international mone tary arrangements tend to break down. His con clusion is that standards based on commitments to rules fail because the commitments are ulti mately not credible. Discussing the relative merits of two alternatives—precommitment by one central bank to price stability with all re maining countries precommitted to fixed ex change rates vs. precommitments by each na tion’s central bank to price stability—Neumann prefers the latter. His reason is that it will pro vide the ability to absorb idiosyncratic shocks (of many varieties) while still providing a credi ble nominal anchor for the price level. In sum, the papers in this proceedings issue reflect Ted Balbach’s world view: markets work, FEDERAL RESERVE BANK OF ST. LOUIS money matters, and empirical evidence is im portant. Add to these guideposts the principle that a policy institution supported by taxpayer dollars should direct attention to relevant, real world issues and you have the framework tha guided the St. Louis Fed’s research effort duri: his tenure. Although the Bank’s many clients and all o f those who worked for or with Ted during his 20 years of service can continue to enjoy his legacy, his presence at the Bank will be dearly missed. Michael T. Belongia St. Louis, Missouri April 23, 1993 xiii Ted Balbach: An Appreciation Anatol Berkman Balbach. Welcome to what is naively called the leisure class. And kiss your leisure good-bye: You had that by working for the Fed. Retirement? No, you’ve shed that and now you’re on your own time as you rush from here to there, no longer working under the 5:00 o'clock mentality. You and Rae have shed one taskmaster for another. But there is some good news. You are now old—a senior citizen or "senile” as I like to say. You’ll get senior dis counts at restaurants and hotels and theaters, but not from doctors or hospitals. You’ll get haircuts that are overpriced. By the way, I’ll offer you some advice. Because the capital value of us old people is very small— w e’ve got a short life expectancy—we lose little by breaking our contracts and commitments. You may therefore wonder why I am here even though I promised to come. Well, I’d really planned to call at the last minute and say, "I’m sorry I have to go to the hospital to have an an giogram and I can’t make it.” I’ve done that twice already at other places and it’s a good ex cuse. So why’d I come? Rae—she’s the one. She's smarter than you or me. She knew what I was going to do, so she phoned me about a month ago and said, "I got a golf date for you on Friday afternoon at Bellerive.” That's why I’m here. Enough about me and about the wisdom of maturity. You, Ted, were born in Konigsburg, Germany, on October 31, 1927, Halloween's day, a mother’s trick-or-treat. W e are glad to know now that it was a treat. As a youth I understand you were drafted into the labor camp and you shoveled stuff from here to there learning about comparative advantage. But then something remarkable happened I’m told. You, in another labor camp, met your mother quite by surprise. I think that’s right—a story we have to hear sometime. You came with her as displaced per sons to Los Angeles because of the prevalence in that area of some earlier relatives, immigrant relatives. Los Angeles still is a haven for refu gees and in many ways is better for it. Driving from the west side o f Los Angeles, where UCLA is, to the east side is like a quick trip through Israel, Japan, Korea, China, Vietnam, Mexico, Iran, Armenia and Africa. You came to UCLA, Ted, after attending high school in Los Angeles, where you thought you learned to speak English. Well, your pronunciation was a lot better than the graduate students we still have there. But the problem is: Why did you choose UCLA? You didn't know anything about it ex cept it was a low-tuition school and near your home. I presume you hadn’t yet learned about the principle about the travels of good and bad grapes. You weren’t shipped very far. To do you justice, though—and the principle, too—you really were shipped a long way from Konigs burg to Los Angeles. So that’s pretty powerful evidence that you really w ere a good grape. I can’t say you came to UCLA because I was there. I only taught statistics during your un dergraduate years, 1947-51. It might have been Karl Brunner, about whom I’ll say more later. You graduated in 1951 and, confronted with earning a real living, you entered graduate school. You set a record that will never be broken, earning your Ph.D. in record time: only 11 years. You must have learned an awful lot, though the truth is you w ere taken in the U.S. Army in 1955 until 1957. For that underpaid service to your new nation and assuming confi dently your army discharge, I thank you now on behalf of the people of the United States of America. After release from the army, you became an instructor at the University of California at the Santa Barbara campus and then a teaching assis tant at UCLA. You now had to earn a living— you were married. In ‘52, I believe, you started as my research assistant, or agent, in managing a study of common stock prices to test the an ticipation of inflation, done with Reuben Kessel. MARCH/APRIL 1993 xiv You foisted on to me a graduate student, Rachel Benveniste, you told me would help in the data collection. She did—she took over and managed the whole project. I now know that you and she had something going. I hope you realize your appointment was close to planned nepotism. But it worked well, and I soon received an invi tation to attend the marriage of Rae and Ted. I wished you both well and that wish has been granted. What you would have done without me I don’t know. Nobel aw ard-winning dissertation I was a director. But w e all know, o f course, that it was Harry M arkow itz, whose help I asked for, w ho really was a director. Incidentally, upon hearing about Bill Sharpe's receiving the award, I w ent and got his dissertation and reread it. It is an astonishing expository paper; it is beautifully written. I now tell all the students there, "Go read Bill Sharpe’s dis sertation. It is so w ell w ritten and it contains a lot o f good econom ics.” But for that matter, what of the many other students I have had who have married in my class, or at least married later? But only a word of gratitude. Upon reflection after 50 years ex perience as a successful though unintended marriage broker, my conviction is that college's main function in one in 10 cases is as a marriage market. So, when I heard about this honor to you, I went and got your dissertation. Would I be one of the signers as I was on Bill Sharpe’s? No. My God, Ted, if I had signed it, you would be get ting a Nobel Prize now. But it wasn’t a bad dis sertation; after all, it was signed by J. Fred Barron who suffered shortly thereafter a tragic, debilitating stroke. And by Bob Baldwin of inter national trade fame and William R. Allen, author of the best textbook in economics. And Robert Rutland and Robert Williams. As I recall your years at UCLA, I’m astounded at the superb quality of graduate and under graduate students we had at UCLA in the late 1950s and ’60s. Ted and Rae, you were prime examples. I used to contend seriously that the economics department at UCLA during those years had a ratio of student quality to faculty quality that was the highest in the nation. I like ambiguities, but it was true. You, Ted and Rae, were two of those who helped raise the stature of UCLA’s department to where Chicago became known as the "UCLA of the East.” Speaking o f Chicago, particularly "Chicago boys,” Rae was, I believe, the first of what real ly distinguished UCLA from Chicago. W e had the "UCLA girls” compared to the Chicago boys. Not many o f you know we had, without ques tion, the most spectacular female graduate stu dents: Rae, Anita Dance, Linda Kliger, Judith Mann, Susan Woodward, Vicki Carnahan to name only a few. You haven't heard o f most of them: after graduation they said to hell with it and they married rich people. Rae married a lit tle too early. The question this provokes in my mind is: "From whence came all these good students?” The likes o f Walter Oi, Tibor Fabian, Martin Bailey, Allan Meltzer, Steven Cheung, Mike Mussa, Lee Hoskins, Art Devaney, Jerry Jordan, Cliff Stone, and Robert Summers, who married Ken A rrow ’s sister and founded a dynasty. I apologize to others whom I could name. Of course I haven't forgotten Bill Sharpe of whose FEDERAL RESERVE BANK OF ST. LOUIS And last, the director, Karl Brunner, to whom you and I owe so much. No one enjoyed his stu dents more than Karl. He was very demanding. He treated them as his children, training, dis ciplining, scolding, abusing and teaching. I have a hard time speaking about him, such a dear friend. As a bright side of one who got him to UCLA, after Lloyd Metzler. He’s the one who said to me: "He’s the one you should get—the young man from Switzerland, on a Rockefeller Scholarship.” So I met with him in Chicago at the Palmer House hotel. And I was convinced. So I went back to UCLA and, after some hard work, convinced the older, senile citizens of the department to bring him there and they did and struck a load for all of us. During his early years at UCLA, Karl inter rupted his career to study statistics, philosophy of logic and the scientific method. As a result of that time taken out, his promotion in rank was scandalously delayed and denied because they said he hadn’t published enough. He suffered, but we gained by what he had learned. The gains to you are evident in the rigor o f your dissertation about the meaningfulness and the evidence pertinent to Clark Warburton’s inter pretation of the effects of money supply changes. You showed and told Warburton that he was ar guing essentially only that money supply changes predicted changes of the price level better than would have a purely random prediction. XV And it did, but it did not exceed a naive model. Your dissertation reported evidence consistent with the implication that the elasticity of the de mand for money was positive with respect to wealth and negative with respect to the rate of interest. You blasted a conception of a distinction between idle and total cash balances and ended by saying the data you collected were consistent with the proposition of money relevance. But that was way back in the ’50s. Most sig nificant, in my opinion, was your emphasis on the evidence of the usefulness o f the higher level hypothesis, the basic economic model from which it is derived, the existence and effects of money supply changes. Money is implied, yet, as you said in a footnote on page 34, we allow for information costs. W e now know they will also imply the real income effects that we attribute to money. You illustrated in that section that cheap information about recognition of com modities reduces those costs as compared to barter and is the key characteristic explaining the effects of money—what makes it money. Had Earl Thompson been there, he’d have explained why virtually costless recognizable receipts for prepayment of taxes serve as our money. As I reread your dissertation, signed September 20, 1962, almost 30 years to the day 1 was read ing it, I was reminded vividly of the hours we all spent with Karl Brunner trying to clarify and find answers to questions like “What is money?” “Why is it money?” "Why do changes in its supply af fect real income and employment, even if only transiently, whereas changes in the supply of shoes, automobiles and wheat do not?” "What presumptions or higher-level hypotheses had to be altered for that effect to be implied?” You had seen beyond the simple monetary patterns. By the way, I noticed you labeled your series M l and M2. Had that been done before or is it origi nal to you? Fortunately your early understanding of a prescription for research remained with you and is reflected in what you have been doing here ever since at the St. Louis Fed. Your career exem plifies and extends the American Promise: im migrant to honorable success and freedom in a capitalist economy. Ted and all of you here as sociated with Ted at the St. Louis Fed, from Homer Jones to Ted and most likely those who follow, you have our gratitude. Thank you and best wishes. Armen A. Alchian Professor Emeritus of Economics University of California-Los Angeles MARCH/APRIL 1993 1 R o b ert H. Rasche Robert H. Rasche is a professor of economics at Michigan State University. Monetary Aggregates, Monetary Policy and Econom ic Activity A LMOST A QUARTER century has passed since the publication of the (in)famous Andersen-Jordan (AJ) equation.1For a good portion of that time, Ted Balbach has been associated with the research department of the Federal Reserve Bank of St. Louis, and for a significant fraction of the period directed the research efforts o f that department.2 Throughout that period the Bank consistently advocated a monetarist approach to monetary analysis and monetary policy. It is appropriate at this point to look back and exam ine what lasting influence this perspective has contributed, both to analysis and to policymaking. ST. LOUIS ON THE ROLE OF MONEY This study has three parts. The first is a re examination of what monetarism and the St. Louis empirical representation thereof contributed. In particular, what controversies o f the late 1960s and 1970s now can be considered settled? The second examines the empirical failures o f the AJ equation in the 1980s and argues that these failures represent specification problems of the "Lucas variety” and not a rejection of the under lying theoretical framework. The implication of such a "Lucas effect” for prominent monetarist policy prescriptions is then analyzed. The third Tw o aspects of the AJ equation seemed partic ularly controversial in the late 1960s. First, the analysis focused on the relationship between nominal measures o f fiscal and monetary policy and nominal income. Second, the analysis focused on growth rates or first differences. Reduced to simplest terms, the analysis stated that the growth in velocity of narrow money, defined as the ratio of nominal GNP to a weighted moving average of M l, fluctuated around a positive deterministic trend and that some fraction o f these fluctuations were correlated with fluctuations in the growth part examines the monetarist proposition that has remained most controversial in recent years, namely the short-run impact of changes in nominal money growth on real economic activity. In particular, the analysis attempts to address the question raised by Cagan—why do vector autoregressions (VARs) produce infer ences about the impact of money on economic activity that contrasts dramatically with the con clusions o f historical analyses?3 ’ See Andersen and Jordan (1968). 2A precusor of the AJ equation can be found in Brunner and Balbach (1959). Michael Belongia is responsible for bringing this well known article to my attention. 3See Cagan (1989). MARCH/APRIL 1993 2 of nominal government spending.4This contrasts sharply with macroeconometric models that were developed contemporaneously. The implicit reduced forms of the latter models specified relationships between the level of nominal money balances and the level of real output. The models also endogenized the price level or inflation rate, but the typical reduced forms implied little if any price level response over the time periods in the AJ specifi cation. The lightning rod in the AJ equation was the conclusion that a maintained change in nominal government spending, unaccompanied by changes in the nominal money stock did not produce a permanent change in nominal income (or velocity) and that changes in high employment nominal tax receipts produced no statistically significant changes in nominal income (or velocity). These implications, which dramatically refuted the fixed-price Keynesian model, did not go un challenged. Numerous counter regressions were published which reported that the implied fiscal policy implications o f the AJ equation were arti facts of measurement error and/or sample specific.5 The point that seems to get lost in the back ground of these challenges is the robustness of the long-run response of nominal income growth to monetary growth shocks: the conclusion that monetary shocks, in the absence of fiscal shocks, have only transitory impacts on velocity growth held its ground in the face of repeated "regres sion attacks”.6 In retrospect it appears that in two significant respects the macroeconomics profession has largely surrendered and accepted the perspective of the AJ equation. First, velocity has been rehabilitated as a useful theoretical device across a broad range o f macroeconomic thought. Monetarists have steadfastly maintained the usefulness o f this concept. Tw o of Greg Mankiw’s (1991) "dubious Keynesian propositions” speak directly to the points raised in the AJ equation: Point No. 2— 4This interpretation of the AJ equation was not widely recognized at the time of publication, I suspect in part because the original specification was published in first differences rather that log differences and also because the specification was never was presented as a hypothesis about velocity. The original presentation was intended as a sequel to the Friedman-Meiselman debate. See Jordan (1986). 5See, for example, deLeeuw and Kalchbrenner (1969), Cor rigan (1970) and Davis (1969). 6For example, Benjamin Friedman (1977) argued that the original Andersen-Jordan conclusion with respect to fiscal policy was sample specific. However, the permanent effect of money growth on velocity is robust to his changes in sample periods. FEDERAL RESERVE BANK OF ST. LOUIS “[T]he lessons of classical economics are not help ful in understanding how the world works”; and Point No. 4—[Fliscal policy is a powerful tool for economic stabilization and monetary policy is not very important.” Mankiw further asserts “for purposes o f analyzing economic policy, a stu dent would be better equipped with the quantity theory o f money (together with the expectations augmented Phillips curve) than with the Keynesian cross.” Some new Keynesians may repudiate Mankiw, since this statement could be paraphrased that a student would be better equipped with the AJ equation (together with the St. Louis model) than with the Keynesian cross.7 Never theless, a statement such as this (original or par aphrase) was heresy 25 years ago, and it can only be said of the St. Louis view o f monetary analysis and monetary policy “you've come a long way baby.”8 Most of the attention that real-business cycle theorists give to money has focused on the rela tionship between money and real output in the short run. Proponents of this approach generally dismiss any causal effect from money to real output, arguing that correlations between changes in money and changes in real output reflect feed backs from real output onto an endogenous money stock. This is not a denial o f all signifi cant parts o f the St. Louis position. Plosser (1991), for example, argues that "money, without ques tion, plays the dominant role in determining the rate of inflation.” Presumably money then also has important impacts on the path of nominal income, though real shocks are also important from this perspective. Real-business cycle spec ifications have recently expanded to include in flation and nominal variables. At least some of these expanded specifications incorporate a traditional demand-for-real-balances function, with point estimates o f long-run income elastici ties that are fairly close to unity.9 Thus these models do not reject the usefulness of velocity 7See Andersen and Carlson (1970) for a discussion of the St. Louis model. eThat the St. Louis view is still contested in discussions of public policy is evidenced by the report of March 31, 1992, that 100 economists, including six Nobel Memorial Prize laureates, sent in an open letter to President Bush, Chairman Greenspan and members of Congress calling for additional government spending, lower interest rates and tax credits for business investment to stimulate economic growth (“ Top Economists Urge Officials to Boost Federal Spending to Stimulate Growth” , Wall Street Journal, March 31, 1992, p. A2). 9See King, Plosser, Stock and Watson (1991). 3 as a long-run concept relating money to nominal income. The second aspect of the evolution of macroeconomic thought toward the AJ equation involves the modeling of shocks to velocity. The AJ equa tion was consistently estimated in differenced form, and thus the implicit assumption of the specification is that shocks to the level of veloc ity are permanent. At the time this analysis was constructed, the discussion of the role of per manent and transitory shocks that is so prominent in recent analyses was unforeseen. Nevertheless, there is vindication for the St. Louis modeling approach in the now conventional wisdom that many macroeconomic time series (including velocity) appear to be "difference stationary" and that there are serious problems o f "spuri ous regressions” in estimations involving levels of such data series.1 0 The conclusion from this discussion is that from current theoretical and econometric perspectives there are important ways in which the original St. Louis analyses "got things right.” Nevertheless, the AJ equation has disappeared from contem porary discussions of monetary policy.1 Why 1 then the demise of the AJ equation? THE DEMISE OF THE AJ EQUATION: ANALYSIS AND SOME IMPLICA TIONS FOR MONETARIST POLICY PRESCRIPTIONS The demise of the AJ equation is well illus trated in figure 1. Tw o different measures of velocity are plotted there. The first is the con ventional ratio of nominal GNP to M l. The sec10See Nelson and Plosser (1982) and Granger and Newbold (1974). "R elatively few attempts to reestimate the St. Louis equation have occurred in recent years. Batten and Thornton (1983) extend the sample period through third quarter 1982. Belongia and Chalfant (1989) estimate regressions using M1A, M1 and divisia variants of both those aggregates (including variables for relative energy prices and strike dummies, but excluding fiscal policy variables) over a first quarter 1976-third quarter 1987 sample. With the exception of Divisia M1A they find money growth elasticities that are significantly less than 1.0. They conclude the following: “ Results indicate that the M1 and broad aggregates are all associated with significant structural change in the money-income relationship around 1981. Gavin and Dewald (1989) estimated St. Louis equations (again omitting fiscal policy variables) over second quarter 1961-th ird quarter 1980 and first quarter 1971-fourth quarter 1982 samples. They conclude from out-of-sample forecasting experiments that “ M1 has done so poorly in the 1980s ond is the ratio o f nominal GNP to a geometric moving average of M l, where the weights in the moving average approximate the weights in the lag polynomial o f the log of differences in money in the AJ equation.1 It is clear that the 2 velocity measure implicit in the AJ equation replicates the behavior o f the traditional M l velocity quite closely, both before and after 1980. Both measures have a strong positive deterministic trend that ends in the early 1980s. This trend was captured in the AJ equation by a significant positive intercept on the order of 2.5 percent to 3.0 percent per year. With the break in the trend in velocity in the 1980s, it is clear that the AJ equation falls apart. In Rasche (1987) I showed that essentially all narrowly defined monetary aggregate velocities in the United States exhibit similar breaks in their deterministic treads in the early 1980s but that once these breaks are considered, the time series properties of the various velocities are not substantially different in the 1980s com pared with the earlier period (see figure 2).1 3 Thus to understand the demise o f the AJ equa tion, it is crucial to understand the origins of the trend in velocity. A considerable number and variety o f expla nations have been advanced for the change in velocity behavior observed in the 1980s, but most of these are not consistent with the pat terns observed in the data.1 Monetarism in gen 4 eral, and the AJ equation in particular, is based on the proposition that a stable long-run demand function for money exists; that is, the demand for real balances depends on relatively few vari ables, including real income, or wealth, and various rates of return on nonmoney assets. that it does worse on average than the monetary base over the entire postwar period, even though it performed better than the base for the 30 years before 1980.” 12The weights are taken from Appendix Table 2 in Carlson (1982) as .40, .40 and .20 on InM,, lnMt _., and lnM,_ 2 respectively. 13These conclusions are not altered by updated data. Over the sample period first quarter 1948-fo u rth quarter 1981 the mean change in velocity (St. Louis velocity) is 3.45 (3.46) percent per year, and the standard deviation is 4.74 (4.74) percent per year. The mean for the first quarter 1982-th ird quarter 1990 is - .7 2 (-.72) percent per year and the standard deviation is 6.04 (5.75) per cent per year. The mean change in the second sample is not significantly different from zero [p = .49 (.46)). In Rasche (1990) I con cluded that the velocities of the broadly defined monetary aggregates M2 and M3 showed little if any changes in trends at this time. 14See Rasche (1987). MARCH/APRIL 1993 4 Figure 1 Velocity Measures First Quarter 1948 through Third Quarter 1990 V e lo city Figure 2 Growth Rate of St. Louis Equation Velocity Measures First Quarter 1948 through Third Quarter 1990 Percent FEDERAL RESERVE BANK OF ST. LOUIS 5 The theory relates the level o f real balances de manded to the level of specific variables. How ever, the AJ equation, proposed as a reduced form of a model containing such a moneydemand specification, is estimated in difference form. Such statistical methodology is correct in that it properly adjusts for the apparent nonstationarities of the observed data series. Unfor tunately, differencing data series maintains only short-run relationships among the various series and overlooks any long-run relationships that may exist simultaneously. In the last decade, particularly in the past five years, innovations in econometric technique allow for the simultaneous treatment of nonstationary data and estimation of long-run relationships among the levels of variables.1 These techniques, 5 namely cointegration analyses, maintain the spirit of the reduced form approach in differ ences of the data, but permit the analysis to incorporate the specification of long-run rela tionships among the levels of the variables, if such relationships exist. If identifying restric tions are satisfied, such a relationship can be interpreted as the long-run money demand func tion that is fundamental to the AJ analysis.1 6 Some studies have documented the existence of such a cointegrating relationship among real balances, real income and nominal interest rates.1 7 The implied long-run income elasticity o f money demand in such estimated equations is not sig nificantly different from unity; hence there is a long-run stationary relationship between the level of velocity and the level o f nominal interest rates. What then of the changes in the mean growth rate of velocity in the 1980s relative to the mean growth rate in previous decades? If a stable longrun money demand equation that relates the level of velocity to the level of nominal interest rates exists and if the deterministic trend (drift) in nominal interest rates changes, then the drift in velocity must change correspondingly to ac commodate the stable money demand specifica tion. Hence a reduced form in differences of 15See Granger (1981); Engle and Granger (1987); Johansen (1988 and 1991); and Phillips (1991). ,6See Hoffman and Rasche (1991b). 17See Hoffman and Rasche (1991a, 1991c and 1992). 18Survey data and inflation forecasts for the United States are consistent with such an interpretation of the outcome of the 1981-82 recession. 19The break in velocity drift as a result of a break in expected inflation is the hypothesis advanced by Milton Friedman, though to the best of my knowledge he did not velocity such as the AJ equation, given a stable money demand function, implies an unchanged constant only as long as there are no significant changes in interest rate trends. Since during the 1980s there is a complete break from the upward trend of nominal rates of the previous two decades, the break in velocity drift is completely consis tent with stability o f the money demand function. Although the velocity break of the 1980s does not invalidate the theoretical propositions on which the AJ equation is based, it suggests that some rethinking o f traditional monetarist policy pre scriptions is in order. What forces are likely to generate breaks in interest rate trends? A plau sible candidate, and the one o f most concern for monetary policy prescriptions, is inflation expec tations. Assume that there is an established initial regime in which expected inflation has a positive trend. Assume that the monetary authorities take successful actions to stabilize the inflation rate and that this regime change is reflected in the expectation of future inflation at some con stant rate.1 The likely outcome o f such a policy 8 shift is that the drift in nominal interest rates will disappear as will the drift in velocity.1 9 This suggests that the time series properties o f velocity and the constants in reduced form equations specified in differences are dependent on specific monetary policy regimes through expected inflation trends specific to the policy regimes. If true, this stands as one of the few clear-cut examples o f a "Lucas effect” beyond the original Phillips curve example.2 0 One o f the consequences o f such a "Lucas effect” is that straightforward application of no feedback monetary growth rules for narrowly defined monetary aggregates can lead to out comes different from those predicted or desired.2 1 A monetary authority that desires to stabilize an inflation that has been drifting upward might be inclined to set a monetary growth objective equal to a projected growth rate for natural output plus a desired stabilized inflation rate, minus the historically observed drift in the elaborate the mechanism described here. 20See Lucas (1976). 21ln Milton Friedman’s defense it must be noted that he originally proposed a no feedback rule in terms of a more broadly defined aggregate, old M2. An aggregate such as new M2, in a regime without interest rate ceilings, is unlikely to suffer from the problem discussed here. For some evi dence on the stationarity of new M2 velocity over the postAccord period, see Hallman, Porter and Small (1991). MARCH/APRIL 1993 6 velocity o f a narrowly defined monetary aggre gate. If the authority maintains this money growth rate after expected inflation has stabi lized, under the above "Lucas effect” the drift in velocity will have disappeared and the actual steady rate of inflation will prove to be lower than the planned inflation rate. During the tran sition period to the steady inflation regime, the drift in velocity will be slowing and hence the growth o f nominal income will drop below the planned inflation rate plus the projected growth rate of natural output.2 If the aim of the mone 2 tary authorities is to reduce, as well as to stabi lize, inflation and if actual and expected inflation adjust to the change in monetary policy slowly so that pt > p* while the drift in velocity is in transition, then real output growth will fall below q* for some time during the transition period.2 3 Meltzer [1987] and McCallum [1988] propose alternatives to a fixed money (base) growth rule that allow feedback from velocity to the planned growth in money (base). The rules are designed to account for permanent shocks to velocity, but not to respond to transitory velocity shocks. The rules set the growth rate of the monetary base equal to a desired growth of nominal income (p* + q* in the above notation) less a moving average of the drift in base velocity.2 4 The rules establish base growth consistent with the planned stable inflation once stabilization is achieved, and the rules also adjust base growth to compensate for the declining velocity drift during the transition period to the stabilized inflation rate. Thus on the surface it appears that these feedback rules immunize monetary policy from the adverse consequences o f the “Lucas effect” on velocity drift. However, this conclusion depends critically on the credibility of the monetary authority. As long as private agents believe that the monetary au 22Set mt = p* + q * - v, where mt is the maintained growth rate of the nominal money stock, p* is the planned steady inflation rate, q* is the projected growth rate of natural output and v is the historically observed drift of velocity. Then during a transition period (pt + qt) = (mt + vt) = (p* + q*) + (vt - v). When the drift in velocity starts to react to the change in expected inflation, (v, - v) < 0 so (Pt + qJ < (p * + q*)23(q, - q*) = (p* - pt) + (vt - v ) < 0 . 24McCallum’s rule provides an additional adjustment to base growth as nominal output is observed to deviate from nominal natural output. 25See Holbrook (1972). Consider, for example, a feedback rule of the form: bt = 0(L)LVt + LX, + £t, where b, is the FEDERAL RESERVE BANK OF ST. LOUIS thority is following the feedback rule consistently, inflation expectations should adjust either in antic ipation of or with the observation over time o f fall ing inflation. The feedback mechanism will adjust base growth as desired. Both the Meltzer and McCal lum rules are deterministic. In practice, stochastic fluctuations around such deterministic rules will be observed which may make direct verification of the rule difficult. If the monetary authority lacks credi bility, feedback rules such as these could prove unstable. Suppose the rule is implemented by the monetary authority and inflation and inflation expectations begin to stabilize. This lowers the drift in velocity, and the feedback rule calls fo r base growth to be adjusted upward (see figure 3). The McCallum rule, which ultimately restores nominal income to the specified path o f nominal potential income, requires that base growth and nominal income growth overshoot equilibrium base growth during the transition period (see figures 3 and 4). If private agents do not under stand the rule well, or if the increase in base and nominal income growth is interpreted by such agents as an abandonment o f the rule, then inflation expectations could start adjusting upward. This would change the drift o f veloc ity, and the rule would then call for reductions in base growth. It is not difficult to conceive of a situation where the monetary authority lacks credibility, in which the Meltzer-McCallum rules suffer from instrument instability (Holbrook [1972]) if the observed behavior of the monetary base affects inflation expectations, and through this the drift in base velocity.2 5 The conclusions from these observations on the reduced form behavior of velocity is that con stant growth rules applied to narrowly defined monetary aggregates are unlikely to be success ful in stabilizing a nonzero inflation trend. The success of feedback rules that depend on observed velocity behavior can depend critically on the credibility o f the monetary authority. In the growth rate of base velocity, X, is other factors to which the feedback rule responds, and t, are random fluctuations generated by fluctuations of sources of monetary base out side the control of the monetary authorities and that can not be perfectly forecasted. Let inflation expectations respond to observed base growth p f+1 = 8(L)bt. Finally, let velocity growth respond to trends in inflation expecta tions: v, = co(L)pf+1. Substituting the latter two equations into the first equation gives [1 - 0(L)5(L)co(L)L] bt = X, + £t. Invertibility of the polynomial [1 — 0(L)5(L)co(L)L], and hence the absence of instrument instability depends upon the expectation formation mechanism, 5(L). 7 Figure 3 McCallum Rule: Nominal Income Growth Percent Figure 4 McCallum Rule: Base and Base Velocity Growth Percent MARCH/APRIL 1993 8 absence of credibility, the adjustments to the growth of the aggregate required by the feed back rule can provoke adjustments to inflation expectations that introduce instrument instabil ity into the feedback rule. CAN THE TRANSITORY RESPONSE OF REAL OUTPUT TO MAINTAINED CHANGES IN MONEY GROWTH BE INFERRED FROM REDUCEDFORM MODELS? The Role o f Identifying Restrictions The focus of much of the recent discussion of the role o f money and monetary policy is not on the response of nominal income, but rather on the response o f real output. Cagan (1989) summarizes a large body of recent empirical research and reaches the conclusion that "lately ... monetary research has turned again ... and new studies claim that money has little or no effect on output and other real variables.” VARs figure prominently in recent research and are the source of much of the evidence from which the negative conclusions about the impact of nominal money changes on real output are drawn. Cagan faults the VAR approach as follows: "The VAR seems ... to be hopelessly unreliable and low in power to detect monetary effects of the kind that we are looking for and believe, from other kinds of evidence, to exist.” I will argue here that Cagan's skepticism about the conclusions of VAR analysis is justified, but for reasons beyond those he enumerated. The most important aspect o f VAR analysis is the one most frequently slighted in drawing con clusions about policy shocks from such analyses. VARs are reduced forms of some unspecified economic model; as such they have common roots with the AJ equation. Reduced forms, in themselves, provide no information about the impact of nominal money shocks, or any other policy shocks o f interest to economists. To pro vide such information, VARs must be supplemented with sufficient identifying restrictions, derived from some economic model, to uniquely extract information about the impact of monetary shocks on real output within the economic structure defined by the identifying restrictions. Sims (1986) clearly explains the critical role of 26See Basmann (1963) and Learner (1985). FEDERAL RESERVE BANK OF ST. LOUIS identifying restrictions in VAR analysis. Sims defines the economic model as follows: m m (1) E A(s)Y(t-s) = E B(s)e(t —s); Var(e(t)) = Q s=0 s=0 and the corresponding VAR (reduced-form) model for Y as follows: m (2) Y(t) = E C(s)Y(t — + u(t); Var(u(t)) = E s) s= 1 Sims notes the following: The most straightforward example of iden tifying restrictions on A(0), B(0) and Q is the Wold causal chain. According to this idea, Q should be diagonal, B(0) = I and A(0) should be triangular and normalized to have ones down the main diagonal when the variables are ordered according to causal priority. Using the fact that with B(0) = I, E = A(0) Q A(0)', the triangularity o f A(0) implies that, once we have put the variables in proper order, we can recover A(0) and Q from E as E ’s unique LDL decomposition. That is, it is known that there is a unique way to express a positive definite matrix E in the form LDL’, where L is lower triangular with ones down its diagonal and D is diagonal. Applying the standard LDL algorithm to E gives us A(0) as L and Q as D. This triangular orthogonalization has become a standard practice as part o f the interpretation o f econometric models (emphasis added) (p. 10). Though this set of identifying restrictions has become so common in VAR analysis that only rarely is it acknowledged explicitly, it is neither unique nor uncontroversial. Criticisms of and arguments against both the appropriateness and necessity of the causal-chain (triangular) specifi cation are longstanding.2 A simple example of 6 the nonuniqueness of this approach is given by the three separate sets of identifying restrictions that Sims applies to his six-variable VAR. All of these identification schemes maintain the assump tion that Q is diagonal, but they impose differ ent exclusion restrictions on A(0), including restrictions that do not impose a triangular structure on A(0). Recently, attention has turned to identification by restrictions on the steady-state coefficient 9 Table 1 Estimates of Four Variable VAR: 11/1955-IV /1981 Dep. Variable: •n(Q,_ -i) ln(Q,_2) ln (Q ,-3) ln(Q, _4) 400Aln(Pt _ 1 ) 400Aln(Pt _2) 400Aln(Pt _3) 400Aln(P,_4) ln(Mt _ i) ln(Mt _ 2) ln(Mt _3) ln(Mt _4) RTB,_, RTB,_ 2 r t b ,_ 3 r t b ,_ 4 CONSTANT D67 D79 R2 SEE ln(Q,) 0.8790 0.0171 -0 .1 5 8 0 0.2778 -0 .0 0 0 5 -0 .0 0 0 7 -0 .0 0 0 6 0.0007 0.3883 0.1956 -0 .4 6 2 6 -0 .1 0 9 5 -0 .0 0 1 5 -0 .0 0 1 9 0.0026 -0 .0 0 2 9 -0 .1 5 3 0 -0 .0 1 0 4 0.0041 0.9900 0.0089 t 7.74 0.11 -1 .0 8 2.51 - 0.86 -1 .1 3 -0 .9 8 1.17 1.99 0.69 - 1 .6 3 - 0 .5 3 -1 .0 8 - 0.88 1.12 -1 .5 7 -1 .3 6 - 2.21 0.54 400Aln(Pt) -14 .3 77 8 -140.4030 30.2668 -3 .9 9 8 8 0.1616 0.0708 -0 .0 0 9 5 -0 .0 2 6 4 -2 .1 3 6 3 33.6234 -10 .9 30 9 -1 6 .6 63 3 0.6513 -0 .8 3 2 7 0.5965 -0 .1 5 6 5 0.4184 1.2788 -1 .5 5 1 5 0.7000 1.6500 matrix, A = S A(s), rather than by restrictions £0 = on A(0).2 This latter approach seems more 7 promising because there appears to be consider able agreement over a broad range of macroeconomic theories on identifying restrictions that apply to a steady-state macroeconomic model.2 In contrast, economic theory provides 8 little if any information about identifying re strictions on the dynamic structure of macroeconomic specifications. In particular, during the past 10 years researchers have broadly debated the identification o f a short-run moneydemand function, to the extent that alleged short-run money demand functions are at best problematic and at worst fall into a class of "incredible” identifying restrictions.2 9 If identification of a short-run money demand function is “incredible,” then any “ shocks” extracted from VARs under these restrictions will at best represent linear combinations of money-demand and money-supply shocks. 27See Bernanke (1986); Blanchard and Quah (1989); and King, Plosser, Stock and Watson (1991). 28See Hoffman and Rasche [1991c] for an illustration of how the restrictions on the KPSW (1991) common trends model are consistent with the identifying restrictions for the steadystate of a standard textbook macroeconomic model. t - 0.68 -0 .5 2 1.11 -0 .1 9 1.43 0.64 -0 .0 8 -0 .2 3 -0 .0 6 0.64 - 0.21 -0 .4 3 2.51 -2 .0 3 1.39 -0 .4 6 0.02 1.47 - 1.11 ln(Mt) 0.0274 0.0091 -0 .0 9 3 0 0.0780 - 0.0002 -0 .0 0 0 3 -0 .0 0 0 6 - 0.0002 1.1862 -0 .1 5 5 5 -0.0131 - 0.0100 -0 .0 0 4 4 0.0039 - 0.0010 0.0007 -0 .1 8 7 9 0.0016 0.0018 0.9900 0.0056 t 0.38 0.10 - 1.00 1.12 -0 .5 4 -0 .6 7 -1 .4 9 -0 .5 3 9.62 -0 .8 7 -0 .7 3 -0 .0 8 -5 .0 3 2.83 -0 .7 0 0.63 -2 .6 4 0.55 0.37 RTB, -3 .7 5 6 7 4.2816 -1 0 .7 8 8 7 11.3726 0.0770 0.0487 0.0339 -0.0451 22.0609 5.9233 0.8224 -2 9 .0 4 1 7 1.0494 -0 .7 9 9 3 0.7571 -0.3761 -6 .5 2 3 3 -0 .4 7 7 5 1.6174 0.9400 0.7260 t -0 .4 0 0.35 -0 .8 9 1.25 1.55 1.00 0.67 -0 .8 9 1.38 0.25 0.04 -1 .7 0 9.15 -4 .4 2 4.01 -2 .4 9 -0 .7 1 -1 .2 4 2.61 Under these conditions it is impossible to sepa rate the impact effects of money on output from the reaction o f money to output through whatever reaction function characterizes the behavior of the monetary authorities. The Importance o f Specification and Identifying Assumptions The questions discussed previously are partic ularly important in the discussion of the effect of nominal money shocks on real output. To illustrate this, consider a four-variable VAR, that includes real output, inflation, nominal money (M l) and a short-term nominal interest rate (Treasury bill rate).3 The general conclusion 0 that has emerged from the study of such VARs is that “ most of the dynamic interactions among the key variables can best be explained as aris ing from an economic structure in which mone tary phenomena do not affect real variables. Thus ... monetary instability has not played an 29See Laidler (1982 and 1985); Cooley and LeRoy (1981); Carr and Darby (1981); Judd and Scadding (1981); and Gordon (1984). See also Sims (1980). 30These VARs are in the form of Sims (1980) and Litterman and Weiss (1985). MARCH/APRIL 1993 10 Table 2 Estimates of Four Variable VAR: 11/1955-111/1990 Dep. Variable: ln(Q,_,) ln(Qt 2) ln (Q ,-3) ln(Q,_4) 400Aln(Pt _ 1) 400Aln(P,_z) 400Aln(P,_3) 400Aln(Pt _4) ln(Mt _,) ln(Mt _2) ln(Mt _3) ln(M,_4) R T B ,„, r t b ,_ 2 r t b ,_ 3 r t b ,_ 4 constant D67 D79 D82 R2 SEE ln(Qt) 1.1398 -0 .0 4 2 8 -0 .2 0 7 4 0.1281 - 0.0002 -0 .0 0 0 4 -0 .0 0 0 3 0.0007 0.0455 0.2742 -0 .4 0 8 2 0.0837 -0 .0 0 0 4 -0.0031 0.0034 -0 .0 0 1 8 -0.0951 -0.0031 0.0027 - 0.0010 0.9900 0.0090 t 11.85 -0 .3 0 -1 .5 1 1.30 -0 .3 0 -0 .7 1 -0 .4 2 1.15 0.34 1.19 -1 .7 8 0.64 -0 .3 1 -1 .8 3 1.84 -1 .2 4 -1 .0 3 -0 .7 4 0.39 -0 .1 4 400Aln(Pt) -8 .6 5 4 -4 .8 9 9 8 15.1429 -3 .7 3 2 7 0.1557 0.0775 -0 .0 0 4 0 0.0101 13.1115 -1 7 .6 69 6 15.3508 -7 .5 2 3 6 0.4783 -0.3581 0.1302 0.0592 0.5120 1.4683 -1 .4 6 1 0 -3 .8 3 4 9 0.6900 1.4900 important role in generating fluctuations.”3 1 Estimates of this four-variable VAR are shown in table 1 and table 2, for sample periods that begin in second quarter 1955 and end in fourth quarter 1981 and third quarter 1990, respec tively. The starting point for both samples is chosen to avoid the pre-Accord data. The first sample ends before the apparent break in the trend of M l velocity discussed previously. The second sample includes the 1980s. The VAR is supplemented with three dummy variables cho sen to define roughly four inflation regimes with different trends.3 2 The implications of these VARs for the response of real output to "money shocks” identified by the Wold causal chain structure with variables ordered as real output, inflation, money and interest rates are quite sensitive to the choice of the sample period (figure 5). Closer examination reveals that this is associated with dramatically 3 See Litterman and Weiss (1985). 1 32The dummy variables are as follows: D67 = 1.0 for 67:4 and subsequent observations; D79 = 1.0 for 79:3 and subsequent observations; and D82 = 1.0 for first quarter 1982 and subsequent observations. 33lt is also difficult to reconcile the “ interest rate shock” FEDERAL RESERVE BANK OF ST. LOUIS t -0 .5 4 - 0.21 0.66 -0 .2 3 1.64 0.82 -0 .0 4 0.10 0.58 -0 .4 6 0.40 -0 .3 0 2.51 -1 .2 7 0.42 0.25 0.03 2.06 -1 .2 5 -3 .1 8 ln(Mt) 0.0216 0.0509 -0 .1 0 2 6 0.0369 -0 .0 0 0 3 -0 .0 0 0 5 -0 .0 0 0 5 - 0.0002 1.4116 -0 .2 3 8 9 -0 .1 3 6 8 -0 .0 3 1 6 -0 .0 0 5 6 0.0057 -0 .0 0 0 5 0.0015 -0 .0 6 9 5 0.0019 - 0.0024 - 0.0111 0.9900 0.0062 t 0.33 0.52 -1 .0 9 0.54 -0 .7 5 - 1.22 -1 .2 7 -0 .4 3 15.13 -1 .5 1 -0 .8 7 -0 .3 5 - 7 .1 9 4.88 -0 .3 7 1.47 - 1.10 0.65 -0 .5 0 -2 .2 4 RTB, 11.6235 7.5080 -18 .6 52 3 1.7967 0.1181 0.1022 0.0287 -0 .0 2 2 8 7.3153 -1 6 .9 36 2 17.7711 -9 .6 5 3 0 0.9899 -0 .6 2 7 5 0.5228 -0 .1 9 5 2 -9.3111 -0 .1 5 7 3 1.8046 -0 .1 6 8 3 0.9300 0.7440 t 1.45 0.64 -1 .6 4 0.22 2.50 2.18 0.58 -0 .4 8 0.65 -0 .8 9 0.93 -0 .8 9 10.45 -4 .4 7 3.42 -1 .6 3 - 1.22 -0 .4 4 3.10 -0 .2 8 different long-run responses of the nominal money stock to the "money shock” (figure 6). Both samples show the real output response to the "money shock” rises to a peak and then trails off. However, the nominal interest rate exhibits a transitory positive response to the "money shock” in both samples which is difficult to reconcile with the identification of the "money shock” as a monetary policy action (figure 7).3 3 Two other variables of interest are implicit in the VAR menu: real money balances and veloc ity. The impulse response function for velocity to a "money shock” is shown in figure 8. The implicit velocity response is almost uniformly negative in both sample periods, and in the third quarter 1990 sample has the peculiar characteristic of having a response below -1.0 even after 40 periods. The realization that the four-variable VAR identified by the Wold causal chain specification with a monetary policy action because although the immediate impact of such a shock on interest rates is positive, the permanent effects of this shock on nominal rates, inflation, money and real output are all negative in both sample periods. 11 Figure 5 Real Output irf to Nominal Money Shock Figure 6 Nominal Money irf to Nominal Money Shock MARCH/APRIL 1993 12 Figure 7 Inflation irf to Nominal Money Shock Figure 8 Velocity irf to Money Growth Shock FEDERAL RESERVE BANK OF ST. LOUIS 13 Table 3 Johansen Maximum Likelihood Estimation of Four Variable VECM ln(M/P), 400*AlnP, InQ, Rtb Sample Normalized Cointegrating Vectors Max A Test Trace Test 11/1955— IV/1974 r=0 39.8 r«1 16.9 r«2 .29 r=0 22.9 r«1 16.6 r«2 .29 11/1955— IV/1975 52.1 21.7 2.2 30.4 19.5 11/1955— IV/1976 52.7 20.3 2.4 32.4 11/1955— 111/1979 54.7 20.5 .31 11/1955— IV/1981 67.1 17.2 11/1955— 111/1990 64.8 22.6 P'c 1.0 0.0 0.0 1.0 - 1.0 0.0 .1224 -.9 1 9 8 2.2 1.0 0.0 0.0 1.0 - 1.0 0.0 .1200 -.731 1 18.0 2.4 1.0 0.0 0.0 1.0 - 1.0 0.0 .1185 - .7296 34.1 20.2 .31 1.0 0.0 0.0 1.0 - 1.0 0.0 .1109 -.7 3 1 5 .77 49.9 16.4 .77 1.0 0.0 0.0 1.0 - 1.0 0.0 .0965 - .7498 4.6 41.3 18.9 4.6 1.0 0.0 0.0 1.0 - 1.0 0.0 .0987 -.8 1 0 6 18.9 21.1 25.8 12.9 14.9 19.2 6.5 8.2 11.7 Critical values from Osterwald-Lenum (1990) 10 percent 5 percent 1 percent 28.7 31.5 37.2 15.7 18.0 23.5 6.5 8.2 11.7 defines additional interesting economic meas ures as linear combinations of the menu entries raises the following question: Are the results invariant to the explicit choice of menu entries? Clearly if the degree of differencing of the vari ables in the VAR were the same, the OLS esti mates would produce the same results regardless of the particular linear combinations explicitly chosen. However, the degree of differencing varies among the variables in the typical VAR study as log levels of real output and nominal money appear along with log differences of the price level (inflation). An alternative menu is to enter real balances along with either inflation or nominal money growth. The advantage of these choices is that the three variables that are tradi tionally included in money-demand specifications— real balances, real output and nominal interest rates—now explicitly appear in the VAR.3 4 In table 3, some results are reported from the 34Such a VAR is an expanded version of the VAR used by Hoffman and Rasche [1992] to investigate long-run money demand. 35See Johansen (1988 and 1991). This restriction was im posed because it was never rejected in the three variable estimation of a VAR with real output, inflation, real money balances and the Treasury bill rate. These results indicate the tests for stationary linear combinations (cointegrating vectors) among the four variables using the Johansen maximum likelihood estimator under the restric tion that the log o f real balances and the log of real output enter any such cointegrating vectors with equal and opposite signs.3 Both of the like 5 lihood ratio tests—the trace test and the maximum eigenvalue test—typically reject the hypothesis of one or few er cointegrating vectors at the 5 percent level, and in some samples at the 1 per cent level. In every case the tests fail to reject the hypothesis that two or few er cointegrating vectors exist. Thus we conclude that among these four variables there are two permanent and two transitory shocks. To obtain a unique (to a scalar multiple) eco nomic interpretation o f the two cointegrating menus investigated by Hoffman and Rasche (1992) and because in that study the unrestricted long-run income and interest elasticities were found to be quite imprecise and sensitive to the choice of the sample period. MARCH/APRIL 1993 14 Table 4 Johansen Maximum Likelihood Estimation of Four Variable VECM ln(M/P), 400*AlnM, InQ, Rtb (Real Balances and Real Income Coefficients Constrained) Sample 11/1955— IV/1981 Trace Test r=0 65.2 r«1 20.3 Normalized Cointegrating Vectors Max X Test . r< 2 1.5 r=0 44.9 r<1 18.8 r«2 1.5 P'c 1.0 0.0 0.0 1.0 - 1.0 0.0 .1068(02) - ,8067(.22) 1.0 0.0 0.0 1.0 - 1.0 0.0 .1088(.03) - .8867(22) 0.0 1.0 - 1.0 0.0 0.1217 - 1.0000 0.0 1.0 - 1.0 0.0 0.1202 - 1.0000 Wald Test of Overidentifymg Restriction /3 'c(2i4) = - 1 . 0 x2(i) = 1.54 p = .21 11/1955— 111/1990 75.8 26.1 5.3 49.7 20.8 5.3 Wald Test of Overidentifying Restriction /3c(2.4) = - 1 0 x2(i) = -27 p = .60 11/1955— IV/1981 Estim ates of Restricted C ointegration V ectors 1.0 0.0 H/1955— IV/1990 vectors present among these four variables, identifying restrictions must be imposed on the estimated matrix o f cointegration vectors.3 In 6 this case the exclusion o f one variable from each cointegrating vector is sufficient to achieve identification. The exclusion restrictions intro duced here eliminate the inflation rate from one cointegrating vector and real balances from the other. The resulting identified cointegrating vec tors, n o r m a liz e d f o r r e a l balances and inflation respectively, are reported as j)c in table 3. The remaining unconstrained coefficients in these matrices are quite stable across sample periods. The estimated interest rate coefficient in the cointegrating vector with real balances is close to the estimate that Hoffman and Rasche ob tained for the long-run interest semielasticity of money demand in the United States.3 The esti 7 mated interest rate coefficient in the cointegrat ing vector with the inflation rate ranges from -0.9 to -0.7 and is not significantly different from -1.0 consistent with a long-run Fisher effect, which implies a stationary real interest rate.3 8 36See Hoffman and Rasche (1991c). 37See Hoffman and Rasche (1992). 38Significance is examined using Wald tests developed by Johansen (1991). Stock and Watson (1991) also report evi dence for a stationary real interest rate. 39See Johansen (1991). FEDERAL RESERVE BANK OF ST. LOUIS 1.0 0.0 The difficulty in interpreting results from this specification o f the VAR is that nominal money or its growth rate does not appear explicitly among the variables in the VAR. An alternative specification is to replace the inflation rate with the growth rate o f nominal money and allow the inflation rate to be determined implicitly by the identity relating nominal money growth and real balances to inflation. Some results from the estimation of this VAR are presented in table 4 using the same sample periods as in table 1 and table 2. These results are basically the same as those in table 3. The Johansen likelihood ratio tests again reject the hypotheses that one or few er cointegrating vectors exist. When the identifying exclusion restrictions and normaliza tion are applied to the two estimated cointegrat ing vectors (/?'), the interest semielasticity in the velocity vector is approximately 0.11 and the interest coefficient in the vector error with the money growth rate is between -0.8 and -0.9. The latter estimates are not significantly differ ent from -1.0 on the basis of a Wald test.3 9 15 Table 5 Estimates of Four Variable Restricted VECM: 11/1955-111/1990 Dep. Variable: Aln(M/P)t _ 1 Aln(M/P)t _2 Aln(M/P)t _3 400A2lnMt _ 1 400A2lnM ,_2 400AzlnM,_3 A ln Q ,., AlnQ,_ 2 AlnQt _ 3 A R T B ,., ARTBt _2 a r t b ,_ 3 CONSTANT D67 D79 D82 CIV1t _4 CIV2,_ 4 R2 SEE Aln(M/P)t) 0.3418 0.3287 0.2738 0.0000 -0 .0 0 0 4 - 0.0011 0.1312 0.0903 -0 .0 4 7 3 -0 .0 0 6 6 -0 .0 0 0 5 - 0.0010 -0 .0 1 1 8 0.0004 0.0020 -0 .0 0 3 9 - 0.0121 - 0.0012 0.6100 0.0068 t 2.07 1.98 1.62 0.008 -0 .6 3 -1 .6 1 0.18 1.22 -0 .6 4 -7 .5 5 -0 .4 9 -0 .8 7 -1 .9 7 0.26 0.60 -1 .0 4 -1 .9 6 -1 .8 1 400A2ln(Mt) 44.87 81.18 81.06 -0 .7 5 -0 .7 9 -0 .9 9 -3 .7 3 19.67 -1 6 .4 9 -2 .1 3 -0 .1 5 - 0.20 -8 .9 9 1.06 0.0116 -3 .6 2 -9 .0 3 -1 .0 5 0.54 2.38 t 0.78 1.40 1.37 - 4 .5 6 -3 .8 5 -4 .3 2 -0 .1 4 0.76 -0 .6 4 -6 .9 3 -0 .4 3 -0 .5 0 -4 .3 2 1.91 0.01 -2 .7 7 -4 .1 8 -4 .6 2 Aln(Qt) 0.1106 0.1793 0.1302 -0 .0 0 0 4 - 0.0001 -0 .0 0 0 7 0.0926 0.0598 -0 .1 4 5 2 - 0.0001 -0 .0 0 3 7 -0 .0 0 0 3 -0 .0 1 5 4 -0 .0 0 1 4 0.0022 -0 .0 0 5 5 -0 .0 2 5 8 -0 .0 0 0 8 0.24 0.0087 t 0.52 0.84 0.60 -0 .6 0 -0 .1 7 -0 .8 1 0.98 0.63 -1 .5 4 - 0.11 -2 .8 1 -0 .1 9 - 2.10 -0 .6 7 0.51 -1 .1 4 -3 .2 7 -0 .9 3 ARTB, -4 1 .3 2 -2 8 .4 7 -6 .9 8 0.1166 0.2007 0.2795 8.22 22.31 0.2366 0.0837 -0 .6 9 7 3 0.1120 -0 .5 9 8 0 -0 .1 4 2 9 1.1099 -0 .3 7 7 7 -0 .5 5 0 4 0.1792 0.4100 0.6914 t -2 .4 6 - 1.68 -0 .4 1 2.44 3.35 4.18 1.09 2.97 0.32 0.94 -6 .7 3 0.94 -0 .9 9 - 0.88 3.26 - 1.00 - 0.88 2.71 NOTE: CIV1 and CIV2 are the two stationary linear combinations of the four dependent variables. The vector error correction model (VECM) in table 4 can be reestimated with the overident ifying restriction (i'cll 4 = -1.0 imposed. The con | strained estimates of / are obtained using the ?' two-step estimator in Rothenberg and the asymptotic covariance matrix for / derived by ?' Johansen.4 The restricted estimates o f /?' are 0 given at the bottom of table 4. These estimates are used to construct two linear combinations of the levels of the four different variables to obtain estimates of the remaining parameters of the restricted VECM. The estimated coefficients o f the restricted VECM are shown in table 5 for the 11/1955— III/1990 sample.4 1 The interesting question that these results raise is: Can the two permanent shocks among these four variables be associated with individ 40Rothenberg (1973) proves that his two-step estimator is a restricted maximum likelihood estimator when the unres tricted estimator is asymptoticly normal and coverges at rate T - 1 . Johansen (1991) shows that his estimator of /3' /2 is asymptotic normal, but converges at rate T ~ 1. The max imum likelihood properties of the restricted estimator have not been established for this case. 41The dummy variables are not important for the estimation of the cointegrating vector (CIV) involving real balances. The separation of the shift in the constant of the VECM into components representing shifts in the deterministic trend and shifts in the mean of this cointegrating vector ual variables? Or in the terminology of King, Plosser, Stock and Watson (KPSW) (1991): Can we derive a structural model from the reducedform model with steady-state character istics suggested by economic theory? The interest ing hypotheses to test are as follows: • One permanent shock corresponds to a real-output (productivity) shock as sug gested by real-business cycle theories; and • The second permanent shock corresponds to a money growth-inflation-nominal inter est rate shock consistent with a broad spec trum of macroeconomic theories. The common-trends modeling approach of KPSW identifies the permanent components of each time series by restricting them to be ran dom walks. A common-trends model exists if indicates that the mean of the CIV is little changed in the 80s compared with the previous 25 years. (See Yoshida and Rasche [1990].) The dummy variables are important for the estimation of the second (real interest rate) coin tegrating vector. They suggest a large increase in the mean real interest rate during fourth quarter 1979-fourth quarter 1981 followed by a substantial, though not fully off setting reduction in the mean real rate after 1981. This is consistent with the work of Clarida and Friedman (1984), Huizinga and Mishkin (1986) and Roley (1986) all of whom found shifts in the relationship of nominal rates and infla tion in 1979 and 1982. MARCH/APRIL 1993 16 the permanent components of each time series are equal to linear combinations of the orthogonal permanent shocks that are suggested by eco nomic theory. In the case under consideration here, the existence of the hypothesized commontrends model requires that the permanent com ponents o f real output and money growth are equal to the two permanent shocks and hence are orthogonal. These correlations are 0.047 and -0.065 for the samples ending in fourth quarter 1981 and third quarter 1990, respec tively. The extent that the permanent compo nents of real output and money growth violate the necessary conditions for the existence of a common-trends model can be judged by the size of the off-diagonal element of the n matrix as defined by KPSW.4 In the sample ending 2 fourth quarter 1981 the estimated restricted VECM implies that n.M = 0.107 and in the sam ple ending third quarter 1990 the estimated res tricted VECM implies that rT2 = -0.007 under 1 the identifying restrictions that the permanent components are random walks. Because the absolute values of these estimates are both close to zero, we conclude that the data are consis tent with a common-trends representation with independent, permanent real-output and perma nent money-growth shocks. KPSW (1991) show how impulse response functions are constructed for permanent shocks in such a common-trends model. Graphs of these impulse response functions are shown in figures 5-18. The long-run properties of these impulse response functions are completely determined by the cointegrating vectors and the near orthogonality of the permanent compo nents of real output and money growth. The long-run responses of velocity, inflation, money growth and nominal interest rates (figures 14, 16, 17 and 18) to a permanent shock to real output are all identically equal to zero. This fol lows from the orthogonalization of the common trends when real output is ordered before money growth. The long-run responses of real output to a permanent money growth shock are not identically zero (figure 10), reflecting the small correlations between the permanent com ponents o f real output and money growth. The long-run responses of inflation and nominal interest rates to a permanent money-growth 42See King, Plosser, Stock and Watson (1991). 43See King, Plosser, Stock and Watson (1991). 44See Plosser (1991). 45See Litterman and Weiss (1985). FEDERAL RESERVE BANK OF ST. LOUIS shock (figures 11 and 13) are identically equal to 1.0 as determined by the values o f the esti mated coefficients in the cointegrating vectors. In the long run, the level o f velocity is increased slightly by the permanent increase in money growth in response to the permanently higher value of nominal rates (figure 8). The long-run responses are consistent with the steady-state properties of most macroeconomic models, but this is not “news” once the elements of the coin tegrating vectors have been estimated. Additional interesting information can be found in these figures. Estimates for both samples sug gest that the transitory responses to either per manent shock die out after two to three years. These implied lags in the adjustment to the steady state seem quite short relative to much of the conventional wisdom, though the length of the transitory reaction of velocity to a perma nent money-growth shock is surprisingly similar to that in the AJ equation. The reactions to a real-output shock are not exactly those implied by a pure real-business cycle model because output effects from this type o f shock build only gradually (figure 15), during which period there are highly serially correlated negative impacts on the inflation rate (figure 16). The real output response here is quite similar to the output response to a "balanced-growth” shock obtained by KPSW in their six-variable restricted VAR model (figure 6).4 There is a transitory money-growth 3 response (figure 17) associated with the output shock, but because the money measure here, M l, includes inside money, this response is con sistent with the picture drawn by some realbusiness cycle theorists.4 4 At first glance, it appears that the variance decomposition of real output in this model is consistent with the conclusion that “monetary instability has not played an important role in generating fluctuations.’’4 The variance decom 5 position of real output from the fourth quarter 1981 sample indicates that the permanent “money-growth” shock accounts for about 23 percent of the variance of real output at all fore cast horizons. In contrast, the permanent “realoutput” shock accounts for only 7 percent of the variance of real output at a one-period hori- 17 Figure 9 T-bill Rate irf to Nominal Money Shock Figure 10 Real Output irf to Money Growth Shock MARCH/APRIL 1993 FEDERAL RESERVE BANK OF ST. LOUIS 18 19 Figure 13 T-bill Rate irf to Money Growth Shock Figure 14 Velocity irf to Real Output Shock MARCH/APRIL 1993 20 Figure 15 Real Output irf to Real Output Shock Figure 16 Inflation irf to Real Output Shock FEDERAL RESERVE BANK OF ST. LOUIS 21 MARCH/APRIL 1993 22 zon but increases to 66 percent o f the variance at a 12-period horizon. When the sample is extended through third quarter 1990, the per manent "money-growth” shock accounts for only 7 percent of the forecast variance at a onequarter horizon, and this declines steadily to one percent o f the forecast variance at a 12-quarter horizon. In this sample the perma nent ‘‘real output” shock accounts for 31 per cent of the forecast variance o f real output at a one-quarter horizon, and this increases steadily to 87 percent o f the variance at a 12-quarter horizon. From this information and the impulse response functions in figure 10, it is tempting to conclude that monetary shocks have little shortrun effect on real output. The variance decomposition o f each cointegrat ing vector can also be computed. At a onequarter horizon 6 (32) percent of the variance of real balances around equilibrium real bal ances is attributable to the permanent "moneygrowth” shock, 27 (24) percent is attributable to the permanent "real-output” shock, and 67 (44) percent is attributable to the two transitory shocks. At a 12-quarter horizon the correspond ing decomposition is 4 (21) percent and 71 (48) percent. At a one-quarter horizon the cor responding decomposition of the variance of the real interest rate around the equilibrium real interest rate is 41(3) percent, 1 (1) percent and 58 (96) percent. At a 12-quarter horizon the decomposition is 24 (10) percent, 20 (27) percent and 56 (63) percent. These decompositions are based on the third quarter 1990 (fourth quarter 1981) sample estimates. It is also possible to allocate the deviation of actual real balances from equilibrium real bal ances (or the actual real rate from the equilibrium real rate) at any point in the sample period to the history of the permanent and real shocks. Following KPSW (1991) write Xt= /it +ATt+ r*(L))'7 t where r){ is a vector of “structural” disturbances.4 6 Let ft'c be the matrix o f cointegrating vectors. Then » c X, measures the deviations o f actual real /}' t balances from equilibrium real balances and the actual real rate from the equilibrium real rate. But /J'X, = P'jit + p'cAx, + P'T*(U r}t = /?T*(L)rjt because / is orthogonal to iu and A. ?' The fallacy o f concluding that monetary insta bility is not important for economic fluctuations from this system under this class of restrictions 46See Appendix B. 47See Friedman (1974) and Andersen and Carlson (1970). FEDERAL RESERVE BANK OF ST. LOUIS involves the interpretation o f the "money-growth” shock (figure 12). Ultimately this shock becomes a maintained change in the growth o f nominal money. However this is not the case initially. For the first two to three years, the money growth response to the permanent "moneygrowth” shock contains a large transitory com ponent and the net effect is frequently of the opposite sign to the permanent effect. This response pattern certainly does not conform to the traditional monetarist policy experiment. In the latter case, the policy intervention involves a shift from one maintained growth rate of money (or the monetary base) to a different maintained monetary growth rate. Under these conditions the traditional monetarist hypothesis is that the initial impact o f the policy intervention will largely affect real output, but that over time this effect will disappear as the inflation rate approaches its new steady-state rate.4 7 The only identifying characteristic o f a mone tary shock in this analysis is the steady-state restriction that the impulse response of money growth to such a shock is one. However, this restriction does not define a unique monetary shock, but rather a whole class of such shocks. This is clear from the impulse responses of money growth to the two "transitory shocks” that are plotted in figures 19 and 20. By con struction, in both samples the steady-state response of money growth (and all other varia bles defined by the VAR) is zero. Thus it is pos sible to define the class of monetary shocks equal to the permanent "monetary shock” plus any weighted sum of the two transitory shocks and satisfy the identifying restriction for a monetary shock. Within this class o f monetary shocks it is impossible to determine the shortrun impact of monetary policy on real output. For example, consider defining the response of real output as the sum of the responses to the permanent “money-growth” shock and the two transitory shocks. Such a composite shock has the identical steady-state response as the perma nent "money-growth” shock and^so satisfies the identifying restrictions for a permanent mone tary intervention imposed by our model. Yet on a one-quarter forecasting horizon such a com posite shock accounts for 69 (93) percent of the variance in real output for the sample period ending third quarter 1990 (fourth quarter 1981). On a 12-quarter horizon the fraction of the forecast 23 MARCH/APRIL 1993 24 variance in real output attributable to such a composite shock decreases to 13 (34) percent for the sample period ending in third quarter 1990 (fourth quarter 1981). The weighted-sum impulse response functions for money growth are shown in figures 21 and 22 for the two sample periods. Large transitory deviations from unity remain in both cases. The fraction o f the variance o f deviations of real balances from equilibrium real balances attributable to this composite shock is 73 (76) percent at a one-quarter horizon and 75 (66) percent at a 12-quarter horizon for the sample period ending third quarter 1990 (fourth quar ter 1981). The fraction of the variance o f devia tions of the real interest rate attributable to this composite shock is 99 (99) percent at a one-quarter horizon and 80 (73) percent at a 12-quarter hori zon for the sample periods ending third quarter 1990 (fourth quarter 1981). The lack of identification of the short-run real output response in the absence of a specification of the monetary rule, or monetary policy reaction function, that prevails during the sample period can be shown easily using a simple macroeconomic model that satisfies all of the steady-state iden tifying restrictions imposed on the VECM. Con sider the following: In contrast, the monetary intervention of traditional monetarist analysis is not contained in the general class o f monetary shocks defined as the permanent "monetary shock” plus a weighted sum o f the transitory shocks. Consider a regression o f the following form: (IMPMP, - 1.0) = /JjIMPMTlj + /?,IMPMT2i + £s where IMPMPi is the impulse response of money growth to the permanent "money shock” and IM PM TL and I\lPMT2i are the impulse responses o f money growth to the transitory shocks. The traditional monetarist policy experiment is defined in the class of identified monetary shocks if there are fas that produce an esti mated impulse response pattern that replicates the deviations o f the impulse response function to the permanent "money shock” from unity. This result does not hold for either sample period. For the sample ending fourth quarter 1981, (IMPMP - 1.0) = -ll.S g U M P M T ^ )(-10.56) 6.19(IMPMT2i) + l . (-9.28) (1) (2) (3) (4) lnP( = t_,lnP"+ y[lnQ, - lnQ't + £n '] InMR, = InQ, - 0i, + t 2 t it = r, + tlnP*+ - InP, 1 lnQt = k + lnAt - art + £3 t where equation (1) is an expectations-augmented Phillips curve (Lucas supply function) that relates deviations o f real output (Q,) from natural out put (Q't to inflation expectation errors (lnPt) t ,lnP^. Equation (2) is a money-demand function that relates real money balances (MRt; InMR, = InM, - lnPt) to real output and nominal interest rates (it) with a unitary income elasticity of money demand. Equation (3) defines nominal interest rates as the sum o f the real rate (rt) and the expected future rate o f inflation (tlnP^+ 1 lnPt). Equation (4) defines the demand for real output in terms of the real interest rate and autonomous planned expenditures (A,). This model is closed by two additional specifications. First, we assume that expectations are gener ated by adaptive expectations o f inflation:4 8 < ,-iP f = .-aPf-i + A<P,-i - i - t P t - J 5> where Pt = lnP(-lnP,_1 and t_lPf_itl = J n P ^ . - ln P , . , R2 = 0.81 SEE = 1.15 while for the sample period ending third quar ter 1990, (IMPMP, - 1.0) = - 9.06(IMPMTL) (-4.31) 0.94(IMPMT2j) + £ . ( - 1 .0 ) R2 = 0.23 SEE = 2.47 “^Adaptive expectations in the inflation rate are chosen as an algebraicly convenient way of generating a model that potentially has transitory real output responses to perma nent nominal money growth shocks and has the steadystate characteristics of the estimated VECM. This is only for illustration of the identification problem. In particular, the type of inflation expectation shift discussed in the sec FEDERAL RESERVE BANK OF ST. LOUIS Second, a stochastic monetary rule (policy reaction function) is specified as follows: (6) AlnM, = fx + ^Ai, + ^,(AlnM, , - /i) + £4 | This rule allows for contemporaneous interest rate smoothing (^,>0) and for offsetting o f past deviations from the steady-state money growth tion beginning on p. 3 as the root cause of the shift in velocity drift is not consistent with an adaptive expecta tions mechanism. 25 Figure 21 Weighted Sum of Permanent and Transitory irf Figure 22 Weighted Sum of Permanent and Transitory irf MARCH/APRIL 1993 FEDERAL RESERVE Table 6 Adjoint Matrix (A*) - [1 + < 2B]a[1 - (1 - A)B - /JA(1 - B)] + a<D,A[1-B] D - [1 + < 2B]a[1 - B]2 t> [0(1 +<D2B) + <t>i][1 - B]2 °[1 “ B] BANK O ST. LOUIS F - [ 1 + < 2B][(a + /})(1 -( 1 -A)B)] + « 0 ,A [1 -B ] D - [1 + < 2B]ay[1 - (1 - A)B] + 0,[1 -« y A -B ][1 - B ] t> - [ 1 + < 2B]/iA[1 -(1 -A)B] - 0>,[1 - B]2 D [(a + P K 1 -B ) - aprX) [1 +tf>2B][(1 +oA)(1 - B ) + AB) [1 + <D 2B][1 - a y A - B][1 - B] [ 1 + 0 2B][Y(1 -(1 -A )B ) + (1 - B)]2 - [1 - a A - B ] <J>,[(1+oA)(1-B) + AB][1 - B] 4>,[1 — cryA — B][1 - B ] 2 <t»i[Y-Y(1 -A)B + (1 - B)2][1 - B] [a y (1 -(1 -A )B ) + (« + » (1 -B )2 cpyH 1 - B ) ] det = [1 +<t>2B][oy(1 -(1 -A)B) + (» + /i)(1 -B )2 - a/JyA(1 - B)] + <D,[1 -a y A -B ][1 - B ] ro o> 27 path (^2 0). Thus with appropriate parameter > values this specification can accommodate a range o f central bank behavior from nominalinterest rate smoothing to a stochastic no-feedback money growth regime. This model can be reduced to a four-variable VAR in lnQt, lnMRt, i( and AlnMt, driven by the exogenous variables InQ", lnAt and the shocks £jt. With some tedious algebra, the moving average representation of the model can be expressed as follows: InQ, (7) lnMRt -ylnQ , + £„ 1.0 £ 2, det k + InA, + £3 | i. . AlnM. _ - ^ ( l + f ) + £ where the polynomial matrix A* and the poly nomial det are given in table 6. In the deter ministic steady state, the impulse response functions are independent of the parameters of the monetary rule (^,, <2 and real output f) > responds only to changes in InQn, (1.0). Simi larly in the deterministic steady-state AlnMt responds only to (i (1.0). However, the transitory responses o f real output to money-growth shocks are not zero. In particular, the greater is the interest-rate smoothing (^,), the smaller are the transitory responses of real output to monetary shocks. Thus estimation of VARs in this type of model will produce different impulse response functions based on different behaviors of the monetary authorities, and it is not possible to infer from those impulse response functions the short-run impact o f a change in money growth under a no-feedback rule, without prior knowl edge o f the form and parameter values of the sample period monetary rule(s). A recent analysis by Strongin (1991) is an attempt at defining a monetary policy disturbance. His identifying restriction is that monetary policy shocks have exactly offsetting impacts on nonborrowed reserves and borrowed reserves and hence have no effect on total reserves. In con trast he assumes that "reserve-demand” shocks in principle affect all three aggregates. Much of Strongin’s discussion of historical Federal Reserve operating procedures focuses on the likely dis 49The exclusion of monetary policy disturbances from total reserve shocks is analogous to identifying the supply function by the exclusion of income in a classical demand/supply model. tribution of reserve-demand shocks (his < para f > meter) between nonborrowed and borrowed reserves. The size of this parameter is not rele vant to his identification problem, though it is important for estimation if the parameter value differs across subsamples.4 The identifying res 9 triction allows him to construct a measure of monetary policy shocks but does not address the structure o f the monetary rule or policy reaction function. Strongin implicitly assumes that there is no contemporaneous feedback from interest rates onto his monetary-policy shock because both total and nonborrowed reserves precede the fed funds rate in his Wold causal chain. Thus his identifying restriction does ' not address all o f the problems raised here. Unfortunately, inference about monetary regimes (policy reaction functions) using regression tech niques has proved illusory. Khoury (1990) reviews 42 attempts at the estimation o f reaction func tions for the Fed over various sample periods. He concludes that "the results were in disarray” and "the specification search showed that very few variables were robust in a reaction function ... consistent with the lack o f robustness in the literature.” The additional attempts at developing reaction functions that are included in Mayer do not overturn this conclusion.5 Thus it appears 0 appropriate to conclude that at present we lack adequate information to make inferences from time series analyses on the vexing question of the short-run impact of nominal shocks on real output. CONCLUSIONS Significant elements of the St. Louis research agenda are now widely accepted, at least in U.S. academic circles and to some extent within the Federal Reserve System. Nevertheless, issues of short-run impacts o f monetary policy remain unresolved. Among these are the following two critical topics: 1) changes in the drift o f velocity and the extent to which such changes are generated by changes in inflation expectations and 2) the short-run impacts o f nominal money shocks on real output. The first o f these questions is critical to the design of monetary rules and/or operating pro cedures that will retain credibility during the 50See Mayer (1990). MARCH/APRIL 1993 28 transition to an alternative inflation regime. The second question has long been debated and appears to be re-emerging as a focus of time series anal ysis. The analysis presented here suggests that the information necessary to pursue this agenda successfully is not yet available. One critical precondition to such analysis is a reasonable specification of the monetary regime(s) during the sample period. In this respect, Cagan’s (1989) appeal for more "historical” research warrants careful consideration. A potential application of such a historical analysis is a test o f Strongin’s (1991) identifying restriction for monetary policy shocks. His res triction provides an estimated time series for such shocks. W e can infer from published Records o f Policy Actions of the FOMC the tim ing and to some extent the magnitude of policy interventions to change the fed funds rate and/or borrowed reserves targets.5 If the identifying 1 restriction is valid, time series estimates of the monetary policy shocks should correlate well with the data extracted from these historical records. REFERENCES Andersen, Leonall C., and Keith M. Carlson. “ A Monetarist Model for Economic Stabilization,” this Review (April 1970), pp. 7-25. Andersen, Leonall C., and Jerry L. Jordan. “ Monetary and Fiscal Actions: A Test of their Relative Importance in Economic Stabilization,” this Review (November 1968), pp. 11-24. Basmann, R.L. “ The Causal Interpretation of Non-Triangular Systems of Economic Relations,” Econometrica (July 1963), pp. 439-48. Batten, Dallas S., and Daniel L. Thornton. “ Polynomial Distributed Lags and the Estimation of the St. Louis Equation,” this Review (April 1983), pp. 13-25. Belongia, Michael T., and James A. Chalfant. “ The Changing Empirical Definition of Money: Some Estimates From a Model of the Demand for Money Substitutes,” Journal of Political Economy (April 1989), pp. 387-97. Cagan, Phillip. “ Money-lncome Causality— A Critical Review of the Literature Since A Monetary History,” in Michael D. Bordo, ed., Money, History, and International Finance: Essays in Honor of Anna J. Schwartz (Chicago: University of Chicago Press for the NBER, 1989). Carlson, Keith M. “ A Monetary Analysis of the Administration’s Budget and Economic Projections,” this Review (May 1982), pp. 3-14. Carr, Jack, and Michael R. Darby. “ The Role of Money Supply Shocks in the Short-Run Demand for Money,” Journal of Monetary Economics (September 1981), pp. 183-99. Clarida, Richard H., and Benjamin M. Friedman. “ The Behavior of U.S. Short-Term Interest Rates Since October, 1979,” Journal of Finance (July 1984), pp. 671-82. Cooley, Thomas F., and Stephen F. LeRoy. “ Identification and Estimation of Money Demand,” American Economic Review (December 1981), pp. 825-44. Corrigan, E. Gerald. “ The Measurement and Importance of Fiscal Policy Changes,” Federal Reserve Bank of New York Monthly Review (June 1970), pp. 133-45. Davis, Richard G. “ How Much Does Money Matter: A Look at Some Recent Evidence,” Federal Reserve Bank of New York Monthly Review (June 1969), pp. 119-31. DeLeeuw, Frank, and John Kalchbrenner. “ Monetary and Fiscal Actions: A Test of Their Relative Importance in Eco nomic Stabilization— Comment,” this Review (April 1969), pp. 6-11. Doan, Thomas A. User’s Manual RATS Version 3.10 (Evanston, IL: VAR Econometrics, 1990). Engle, Robert F., and Clive W.J. Granger. “ Cointegration and Error Correction: Representation, Estimation and Testing,” Econometrica (March 1987), pp. 257-76. Friedman, Benjamin M. “ Even the St. Louis Model Now Believes in Fiscal Policy,” Journal of Money, Credit and Banking (May 1977), pp. 365-67. Friedman, Milton. “ A Theoretical Framework for Monetary Analysis,” in Robert J. Gordon, ed., Milton Friedman’s Monetary Framework: A Debate With His Critics (University of Chicago Press, 1974). Gavin, William T., and William G. Dewald. “ The Effect of Disinflationary Policies on Monetary Velocity,” The Cato Journal (Spring/Summer 1989), pp. 149-64. Gordon, Robert J. “ The Short-Run Demand for Money: A Reconsideration,” Journal of Money, Credit and Banking, Part 1 (November 1984), pp. 403-34. Granger, Clive W.J. “ Some Properties of Time Series Data and Their Use in Econometric Model Specification,” Journal of Econometrics (May 1981), pp. 121-30. Bernanke, Ben S. “ Alternative Explanations of the MoneyIncome Correlation,” Carnegie-Rochester Conference Series on Public Policy (Autumn 1986), pp. 49-99. Granger, Clive W.J., and P. Newbold. “ Spurious Regression in Econometrics,” Journal of Econometrics (July 1974), pp. 111-20. Blanchard, Olivier Jean, and Danny Quah. “ The Dynamic Effect of Aggregate Demand and Supply Disturbances,” American Economic Review (September 1989), pp. 655-73. Hallman, Jeffrey J., Richard D. Porter, and David H. Small. “ Is the Price Level Tied to the M2 Monetary Aggregate in the Long Run?” American Economic Review (September 1991), pp. 841-58. Brunner, Karl, and Anatol B. Balbach. “ An Evaluation of Two Types of Monetary Theories,” Proceedings of the Thirty-Fourth Annual Conference of the Western Economic Association, September 2-4, 1959 (Santa Barbara, California), pp. 78-84. Brunner, Karl, and Meltzer, Allan H. The Federal Reserve's Attach ment to the Free Reserve Concept. Subcommittee on Domestic Finance, U.S. House of Representatives, Com mittee on Banking and Currency, May 7, 1964. 51See Brunner and Meltzer (1964) and Rasche (1987). FEDERAL RESERVE BANK OF ST. LOUIS Hoffman, Dennis L., and Robert H. Rasche. “ Long-Run Income and Interest Elasticities of Money Demand in the United States,” Review of Economics and Statistics (November, 1991a), pp. 665-74. ________“ The Demand for Money in the U.S. During the Great Depression and Post War Period: Identifying the Source of Shifts in Velocity,” mimeo (1991b). 29 ________“ Identification and Inference in Cointegrated Systems: A Synthesis of Recent Developments,” mimeo (1991c). ________“ Money Demand in the U.S. and Japan: Analysis of Stability and the Importance of Transitory and Permanent Shocks,” mimeo (1992). Holbrook, Robert S. “ Optimal Economic Policy and the Prob lem of Instrument Instability,” American Economic Review (March 1972), pp. 56-65. Huizinga, John, and Frederic S. Mishkin. “ Monetary Policy Regime Shifts and the Unusual Behavior of Real Interest Rates,” Carnegie-Rochester Conference Series on Public Policy (Spring 1986), pp. 231-74. Business Cycle Interpretation,” in Michael T. Belongia, ed., Monetary Policy on the 75th Anniversary of the Federal Reserve System (Boston: Kluwer Academic Publishers, 1991), pp. 245-74. Rasche, Robert H. “ M1-Velocity and Money-Demand Functions: Do Stable Relationships Exist?” Carnegie-Rochester Con ference Series on Public Policy (Autumn 1987), pp. 9-88. ________ “ Demand Functions for Measures of U.S. Money and Debt,” in Peter Hooper et. al., eds., Financial Sectors in Open Economies: Empirical Analysis and Policy Issues (Washington: Board of Governors of the Federal Reserve System, 1990). Johansen, Soren. “ Statistical Analysis of Cointegration Vectors,” Journal of Economic Dynamics and Control (June-September 1988), pp. 231-54. Roley, V. Vance. “ The Response of Interest Rates to Money Announcements Under Alternative Operating Procedures and Reserve Requirement Systems,” NBER Working Paper 1812 (January 1986). ________“ Estimation and Hypothesis Testing of Cointegration Vectors in Gaussian Vector Autoregressive Models,” Econometrica (November 1991), pp. 1551-80. Rothenberg, Thomas J. Efficient Estimation With Apriori Infor mation (New Haven: Yale University Press, 1974). Jordan, Jerry L. “ The Andersen-Jordan Approach After Nearly 20 Years,” this Review (October 1986), pp. 5-8. Judd, John P., and John L. Scadding. “ Liability Management, Bank Loans and Deposit ’Market’ Disequilibrium,” Federal Reserve Bank of San Francisco Economic Review (Summer 1981), pp. 21-44. King, Robert, Charles I. Plosser, James H. Stock, and Mark W. Watson. “ Stochastic Trends and Economic Fluctuations,” American Economic Review (September 1991), pp. 819-40. Khoury, Salwa S. “ The Federal Reserve Reaction Function: A Specification Search,” in Thomas A. Meyer, ed., The Political Economy of American Monetary Policy (New York: Cambridge University Press, 1990). Laidler, David E.W. The Demand for Money: Theories, Evidence and Problems (New York: Harper and Row, 1985). Sims, Christopher A. “ Macroeconomics and Reality,” Econometrica (January 1980), pp. 1-48. ________“ Are Forecasting Models Usable for Policy Analysis?” Federal Reserve Bank of Minneapolis Quarterly Review (Winter 1986), pp. 2-16. Stock, James H., and Mark W. Watson. “ A Simple Estimator of Cointegrating Vectors in Hiqher Order Inteqrated Systems,” mimeo (1991). Strongin, Steven. “ The Identification of Monetary Policy Dis turbances: Explaining the Liquidity Puzzle,” Federal Reserve Bank of Chicago Working Paper W P-91-24, (December 1991). Yoshida, Tomoo, and Robert H. Rasche. “ The M2 Demand in Japan: Shifted and Unstable?” Bank of Japan Monetary and Economic Studies (September 1990), pp. 9-30. _______ . Monetarist Perspectives (Cambridge, Massachusetts: Harvard University Press, 1982). Learner, Edward E. “ Vector Autoregressions for Causal Infer ence?” Carnegie-Rochester Conference Series on Public Policy (Spring 1985), pp. 255-303. Litterman, Robert B., and Laurence M. Weiss. “ Money, Real Interest Rates, and Output: A Reinterpretation of Postwar U.S. Data,” Econometrica (January 1985), pp. 129-56. Lucas, Robert E. Jr. “ Econometric Policy Evaluation: A Critique,” Journal of Monetary Economics (Supplementary Series, Volume 1, 1976) pp. 19-46. Mankiw, N.G. “ The Reincarnation of Keynesian Economics,” NBER Working Paper 3885 (October 1991). McCallum, Bennett T. “ Robustness Properties of a Rule for Monetary Policy,” Carnegie-Rochester Conference Series on Public Policy (Autumn 1988), pp. 173-203. Meltzer, Allan H. “ Limits of Short-Run Stabilization Policy,” Economic Inquiry (January 1987), pp. 1-14. Meyer, Thomas A. The Political Economy of American Monetary Policy (New York: Cambridge University Press, 1990). Nelson, Charles R., and Charles I. Plosser. “ Trends and Random Walks in Macroeconomic Time Series: Some Evidence and Implications,” Journal of Monetary Economics (September 1982), pp. 139-62. Osterwald-Lenum, Michael. “ Recalculated and Extended Tables of the Asymptotic Distribution of Some Important Maximum Likelihood Cointegration Test Statistics,” Institute of Eco nomics, University of Copenhagen, mimeo, (January 1990). Phillips, P.C.B. “ Optimal Inference in Cointegrated Systems,” Econometrica (March 1991), pp. 283-306. Plosser, Charles I. “ Money and Business Cycles: A Real Appendix A Technical Description of the Assum ptions in the Simulation of the McCallum Rule The initial regime (periods 1-19) in figures 3 and 4 before the implementation of the McCallum rule are base velocity growth at tv = .0075 per period. The monetary base is assumed to grow at a rate that increases at tr = .0001563 per period. Thus nominal income growth is increas ing at a rate of t. = .0001563 per period. The rate of increase in nominal income growth is assumed to reflect the trend in inflation, which in turn is assumed to reflect the trend in nomi nal interest rates. The trend in nominal interest rates and the trend in velocity must satisfy the restriction tv - e rtr = 0 where er is the interest semielasticity of velocity if base velocity and interest rates are cointegrated. er is assumed to be 48, using the estimated semielasticity of M l MARCH/APRIL 1993 30 velocity from Hoffman and Rasche [1992]. Nominal income and nominal potential income are assumed equal throughout this period. The regime switch to the McCallum rule is an nounced and implemented in period 20. The desired growth rate o f nominal income in the new regime is .0075 per period. It is assumed that the announcement of the new policy results in an immediate elimination of base velocity growth. It should be noted that the path o f nominal income growth once the McCallum regime is implemented is independent of the assumptions about growth in the prior regime. Nominal income growth in the McCallum regime is totally deter mined by the assumed growth of velocity start ing 16 periods prior to the implementation of the rule, and the reaction of velocity growth to the institution o f the new regime. The particu lar initial conditions for base growth used here are chosen strictly for consistency with the assumed initial growth rate o f base velocity. replications are plotted in Figures A1-A10, together with confidence bands of ± 1.96* (standard deviations of the impulse responses across replications). The graphs suggest that the short-run responses of real output and velocity with respect to both permanent shocks are measured with considerable precision, in particular that the real output response to a permanent "money growth” shock is initially significantly positive and that the real output response of a perma nent ” real output” shock is significantly less than 1.0 for about 10 quarters. In contrast, the measurement of the short-run responses of inflation, money growth, and interest rates to both permanent shocks is highly imprecise. Appendix C Sources of Data All data series were extracted from Citibase. The primary sources are as follows: Treasury Bill Rate: Three month secondary market rate. Federal Reserve Bulletin. Table 1.35, line 15. Appendix B Confidence Intervals for Im pulse Response Functions Estimates of the precision of the impulse response functions from the sample ending in 90:3 w ere constructed from a Monte Carlo integration. The estimated coefficients and covariance matrix o f residuals from a VAR aug mented by two error correction variables were shocked using the algorithm described in Doan [1990], example 10.1. The elements of the coin tegrating vectors were held constant at their estimated values, since Johansen [1991], The orem 5.5, proves that the estimated asymptotic covariance matrix of I I = a p' depends only on the estimated asymptotic covariances of a and the estimated p and not on the estimated asymp totic covariance of p. 1000 replications on the parameter values were constructed and the parameters of the KPSW common trends model were recomputed for each replication. The mean value of KPSW’s critical I I , , parameter across all replications is -.0057 with a standard deviation of .0371. These parameters were used to derive impulse response functions. The means of various impulse responses across the 1000 FEDERAL RESERVE BANK OF ST. LOUIS M l: Seasonally adjusted monthly data. January 1947-December 1958 con structed following Rasche (1987), Appendix A. January 1959-December 1989 Board of Governors of the Federal Reserve System, Money Stock Revisions, March 1990, Table 1. January 1990-March 1990 Federal Reserve Bulletin, July 1990, Table 1.21, line 1. April 1990-June 1990 Federal Reserve Bulletin, October 1990, Table 1.21, line 1. July 1990 Federal Reserve Bulletin, January 1991, Table 1.21, line 1. August 1990 Federal Reserve Bulletin, February 1991, Table 1.21, line 1. GNP: Seasonally adjusted quarterly data. January 1947-April 1987 Survey of Current Business, July 1990 January 1988-March 1990 Economic Report of the President, February 1991, Table B-l. Real GNP: Seasonally adjusted quarterly data. January 1947-April 1987 Survey of Current Business, July 1990 January 1988-March 1990 Economic Report of the President, February 1991, Table B-2. 31 Appendix Figure 1 Velocity irf to Money Growth Shock Appendix Figure 2 Real Output irf to Money Growth Shock MARCH/APRIL 1993 32 Appendix Figure 3 Inflation irf to Money Growth Shock Appendix Figure 4 Money Growth irf to Money Growth Shock FEDERAL RESERVE BANK OF ST. LOUIS 33 Appendix Figure 5 T-bill Rate irf to Money Growth Shock Appendix Figure 6 Velocity irf to Real Output Shock MARCH/APRIL 1993 34 Appendix Figure 7 Real Output irf to Real Output Shock Appendix Figure 8 Inflation irf to Real Output Shock FEDERAL RESERVE BANK OF ST. LOUIS 35 Appendix Figure 9 Money Growth irf to Real Output Shock Appendix Figure 10 T-Bill Rate irf to Real Output Shock MARCH/APRIL 1993 Julio J. R otem berg Julio J. Rotemberg is a professor of economics at the Sloan School of Management at the Massachusetts Institute of Technology. He thanks the National Science Foundation for research support. Commentary R OBERT RASCHE’S PAPER is divided into three parts. In the first part Rasche argues that many monetarist propositions have become widely accepted by macroeconomic theorists. With this I agree completely. The work of the so-called New Keynesians should really be called New Monetarist. Many o f the papers in this tradition focus exclusively on the effect of monetary shocks. Moreover, the key building block for these papers is a quantity equation where velocity is treated as constant. This is consistent with a basic monetarist tenet that money demand is pretty stable and changes in money represent mainly autonomous changes in money supply. The second part of Rasche's paper is concerned with bolstering the view that money demand, at least in some sense, has been stable over time. This is the part of the paper with which I disagree most. The third part of the paper concerns vector autoregressions and the ambiguous role they give to monetary disturbances. Rasche argues that it is hard to identify the economic meaning of the residuals in these vector autoregressions. It is particularly hard to determine which residual or which combination of these residuals represents shocks to the money supply. Therefore it is diffi cult to use these particular statistical techniques for evaluating monetary policy. I am in basic agree 'Luca s’ data actually run from 1900, so he displays even more stability than reported here. FEDERAL RESERVE BANK OF ST. LOUIS ment with this part of the paper, so when I dis cuss it, I will mainly elaborate on themes that Rasche develops. Let me start with the issues raised by the second part of the paper. The Andersen-Jordan equation relates the change in nominal GNP to the change in money (measured by the monetary base) and changes in several measures o f fiscal stance. Regressions of output on money are a basic sta ple of empirical macroeconomics today, and Andersen and Jordan deserve credit for pioneer ing regressions o f this form . As Rasche emphasizes, such regressions really make sense only if you can think o f monetary growth as representing an exogenous impulse. This requires among other things that money demand be stable. The instability o f money demand has been researched at length. One apparent instability on which Rasche focuses is the change in velocity’s trend around the early 1980s. He attributes this change to a change in interest rate trends. The implicit suggestion is that money demand is stable after all. Rasche's view, which echoes Lucas (1988), is that there exists a stable, long-run money demand equation that can be estimated using levels of velocity and interest rates.1 This cointegrating regression explains the trend in velocity with the trend of interest rates in the pre-1980 37 Figure 1 Actual and Fitted Velocity Velocity period. I have rerun a similar equation using CITIBASE data from 1959:1 to 1989:2, and the results are as follows: GNP (1) log(_i w F ) = ‘ R2 = 0.649 1286 (0.029) D.W. + 0 0 6 2 , ’ t (0.004) = 0.18, where i is the interest rate on Treasury bills. My estimated interest elasticity is a bit smaller than Rasche’s, but the cointegration result is the same. To understand what is going on, it is helpful to look at figure 1, which shows the fitted values of this regression together with the actual values for the logarithm o f velocity. W e see that the fitted values match the trend in velocity rather well through the early 1980s. The ability o f interest rates to account for the long-run movements in velocity is remarkable. I have to confess that the ability to track the trends in velocity is so amazing that I wanted to check whether the inter est rate’s regression coefficient was unchanged from the pre-1979 to the post-1979 period. I thus ran the following regression: GNP (2) log ( ------) = 1.346 + 0.047/ + O.Olld, * / Ml ' ' 1 (0.038) (0.008) (0.005) R2 = 0.665 D.W. = 0.16, where dt represents a dummy variable that equals zero before 1979:1 and one afterwards. The coefficient on the interest rate is thus sig nificantly higher in the second part o f the sam ple. Therefore all stability problems should not be considered solved. Nonetheless, it is impressive that a single interest rate coefficient can track the trends in velocity. The question at this point is whether Rasche has found the money demand equation. An alternative view is that the ability o f interest rates to explain long-run trends in money is a coincidence and that economists should really search for a money demand equation that explains deviations around trends. In this view trends in money are caused by secular changes in regulation and technol ogy, which have nothing to do with interest rates. Thus trends in interest rates and velocity are unreliable sources of information about the semi-elasticity of money demand with respect to interest rates. On a priori grounds, one should prefer Rasche’s interpretation because it doesn’t rely on anything outside the model, such as regulatory changes and technical progress. And yet I must admit that I resist Rasche’s view that he has found the true underlying money-demand function. I resist because I am bothered by the huge residuals in MARCH/APRIL 1993 38 this equation. The fitted values are often 20 percent or more away from the actual level of velocity. Moreover, these huge residuals aren’t just random; they are strongly correlated with high-frequency movements in interest rates. Thus around 1975 when interest rates were relatively low, velocity was also predicted to be low. There must have been a huge reduction in money demand around this period to explain the actual behavior of velocity. Put differently, the medium- and high-frequency movements in interest rates, including the spike of 1979, must be attributed to large temporary changes in money demand. I find this hard to believe, how ever, and tend to trust the conventional wisdom that attributes many of these changes in inter est rates to monetary policy. But to believe this conventional wisdom, you have to believe that the short-run interest elasticity o f money demand is different from that estimated by the coin tegrating regression. In other words, you have to believe that the cointegrating regression does not deliver the stable money demand curve that can be used for short-run policy analysis. To see what difference this makes, I have rerun the regression explaining the log of velocity with interest rates through the end o f 1981 but adding a trend. In other words, in this regression the trend is due to technical progress in credit cards and other advances that allow individuals to conserve on money balances. In figure 1 the fitted value with the trend is closer to the actual value than is the fitted value without the trend. Figure 1 supports Rasche’s view that money demand is stable after all because it explains long-run swings but this stability is purchased at a heavy price. It must be that money demand is incredibly unstable at short and medium frequencies. This comment just puts the shoe on the other foot because it raises the question of why money demand rose greatly (and velocity declined) in the early 1980s. Although I am far from having a com plete explanation of this phenomenon, I want to return briefly to the theme I presented when I was here three years ago. I said then that sim ply adding monetary assets that pay different interest rates makes no sense and that proce dures such as the Divisia method advocated by Barnett (1980) should be used instead. This seems particularly germane to the question of why velocity fell in the 1980s. The reason is that cur rency and non-interest-paying demand deposits 2See Rotemberg, Driscoll and Poterba (1992) for a more detailed analysis. FEDERAL RESERVE BANK OF ST. LOUIS did not rise unprecedentedly in this period. Figure 2 shows the velocity o f the aggregate that includes only these non-interest-bearing assets and it continues to trend upward. What did increase dramatically in this period is the holding of other checkable deposits that pay interest. But adding these to the rest is simply misleading. Other checkable deposits are much more attractive as savings instruments than the other components of M l, so they should be regarded as less monetary. Exactly how much less monetary than other checkable deposits is perhaps a matter o f debate. For current purposes, let me propose that they are about one-third as monetary as the other ingredients in M l, so a proper aggregate can be constructed by adding one-third o f other check able deposits to the other two monetary compo nents.2 The logic behind this is as follows. In the late 1980s the Treasury bill rate used by Rasche averaged about 7 percent, and interest rates on other checkable deposits w ere between 4 percent and 5 percent. Thus the gap between these two interest rates is about one-third of the gap between the interest rate on currency and that on Treas ury bills. Figure 2 displays the result o f treating other checkable deposits as being one-third as monetary as the other assets. Little change in trend can now be detected. The exercise I just finished is hopelessly crude, and I am not really trying to push the idea that the velocity trend is a natural constant. Rather I am trying to say that the remarkable fit of the Rasche regression explaining velocity should be taken with a grain of salt. There are other plausible reasons for velocity to have fallen in the 1980s. In conclusion, even if Rasche’s regression gives us the true money-demand relation, the volatility of velocity is substantial. This makes a rule where money grows at a constant rate unattractive relative to a rule where monetary changes track changes in velocity. This raises the following two related questions. How do you ascertain that the Fed has done a good job of accommodating changes in velocity, and how do you measure actual monetary impulses over and above those needed to satisfy changes in money demand? I agree with Rasche’s basic thrust that a purely statistical approach cannot disentangle the endogenous and exogenous parts o f changes in 39 Figure 2 The Velocity of Different Aggregates V e lo c ity 2 .25 - 1.75 ■ M1 Log Velocity * Log V elocity of C urrency and Demand Deposits - Log Velocity w eighing OCDs by one-third I 1959 I I 61 I I 63 I I 65 I I 67 I I 69 I I 71 money. It is hard to know exactly what is being captured by innovations in money within typical VARs. I am particularly bothered by the incon sistency between regressions of various varia bles on money and money innovations on the one hand and regressions that use variables that reflect changes in Federal Reserve intentions on the other. In the case of regressions that use money and money innovations on the right-hand side, one generally finds that money raises output and interest rates. Over the years, several authors have constructed variables on the basis of the FOMC minutes that are supposed to reflect Federal Reserve intentions. Boschen and Mills (1992) show that all these proxies have similar correlations with subsequent levels of GNP and interest rates. In particular, after the proxies indicate that the Fed wishes to tighten, output falls while interest rates rise. These opposite reactions o f output and interest rates are quite consistent with textbook models and it is incon sistent with the simultaneous increase in output and interest rates that tends to follow increases in money. So how can one reconcile regressions I I 73 I I 75 I I 77 I I 79 I I I 81 I 83 I I 85 I 87 I I 1989 on proxies of Fed intentions on the one hand and regressions on money growth on the other? One feature o f the U.S. postwar period is that many of the well-known episodes of changes in Fed policy involve deliberate tightening to slow inflation.3 Thus the historical evidence appears consistent with the asymmetries found by Cover (1992) and De Long and Summers (1988). These authors find that output is much more strongly correlated with negative monetary innovations than with positive ones. The latter have a small positive effect on output that is statistically insignificant. De Long and Summers construct their innova tions by running the following regression: (3) Alog M l, = 0.005 + 0.457 Alog M1m (0.002) (0.091) 0.145 Alog GNPt l + 6e-5Trend (0.083) (2e-5) R2 = 0.321 D.W. = 1.93 The residuals o f this regression can be used to construct DM", which equals the smaller of the residual and zero. Thus this variable captures 3See Romer and Romer (1989). MARCH/APRIL 1993 negative innovations. As a further analysis of asymmetries, I considered a regression of the Treasury bill rate on changes in money and on DM" over the period 1960:3-1989:2. The results are as follows: (4) it = -0.04 + 1.31itl - 0.68it2 + 0.72i13- 0.5i(4 (0.198) (0.088) (0.151) (0.166) (0.106) 51.4DM (15) + 75.0DM" (25) - 10.0DM,"s + 8.2DM"3 - 52.7DM" (27) (25) (22) R2 = 0.944 4.9DM (18) * + 3 4 D M t-3 (17) - 12.8DMt (14) D.W. = 2.13, where DM represents the change in the logarithm of M l. W e see here that, as in typical VARs, lagged changes in money tend to increase interest rates. But on the other hand, the effect of lagged nega tive monetary innovations is negative. This means that negative monetary innovations actually raise interest rates; they have the same correla tion with interest rates as the proxies that indi cate that the Fed wishes to tighten.4 Combining these results with those of Cover and those of De Long and Summers, I conclude that negative monetary innovations affect the economy as money supply shocks should, whereas positive shocks do not. How should you interpret the positive innova tions in money in light o f their correlation with GNP and the Treasury bill rate? One cannot say that they represent simply monetary accommo dation to increases in the stochastic component of the demand for money. Given that these innovations lead to rises in interest rates, the accommodation can be only partial. But partial accommodation of money-demand disturbances should lead to declines rather than small increases in output. I am thus inclined to believe that these positive innovations in money represent in part accommodation by the Fed of other shocks whose effect is to increase future output. Thus the Fed is accommodating increases in money demand that are due to increases in output rather than mere 4I also checked whether I could detect asymmetric effects on inflation (to see whether the asymmetric effects on interest FEDERAL RESERVE BANK OF ST. LOUIS money-demand disturbances. If this is true, it suggests that the Fed is sometimes farsighted. These results can be used to discuss an alter native explanation of the asymmetries found by Cover and by De Long and Summers. This alter native view holds that all monetary innovations represent exogenous increases in the money supply but that the effects of changes in the money supply are intrinsically asymmetric. The traditional analogy is that monetary policy oper ates like a string and that strings are useful only for pulling the economy down, not for pushing it up. Several theories of such asymmetries have been proposed. One cause o f this structural asym metry could be that reductions in reserves force banks to cut loans, whereas banks can react to an increase in reserves by raising their holding of securities. If investment depends on the sup ply of loans and not on the interest rate and the supply of loans is affected only by monetary contractions, then only money-supply reductions have a powerful effect on the economy. Another possible cause o f this asymmetry is discussed by Caballero and Engel (1992) who show that asym metry is a natural consequence o f the steadystate distribution of prices in an economy with costs o f price adjustment. Both o f the preceding hypotheses may well be able to explain the asymmetry found by Cover and by De Long and Summers. They cannot, how ever, explain the asymmetric effect on interest rates as easily. The asymmetric response of interest rates casts doubt on the hypotheses by showing that the effects are asymmetric even in securities markets, not just in markets subject to frictions (for example, the market in which banks inter mediate loans to businesses and the markets in which firms set prices that are relatively rigid). As Robert Rasche emphasizes, there is still much to be done to understand the precise role of mone tary innovations in statistical models. To understand this role, we have to connect money innovations with other historical and institutional data. Only then can we ascertain the empirical importance of money-supply disturbances in the economy. REFERENCES Barnett, William A. “ Economic Monetary Aggregates: An Application of Index Numbers and Aggregation Theory,” Journal of Econometrics (Summer 1980), pp. 11-48. rates represent asymmetric effects on real returns as well). Inflation does not appear to respond asymmetrically. 41 Boschen, John F., and Leonard O. Mills. “ The Effects of Countercyclical Policy on Money and Interest Rates: An Evaluation of FOMC Documents,” mimeo, 1992. Lucas, Robert E. Jr. “ Money Demand in the United States: A Quantitative Review,” Carnegie Rochester Conference Series on Public Policy (Autumn 1988), pp. 137-68. Caballero, Ricardo and Eduardo Engel. ‘‘Price Rigidities, Asymmetries and Output Fluctuations,” NBER Working Paper 4091 (June 1992). Romer, Christina D., and David H. Romer. “ Does Monetary Policy Matter: A New Test in the Spirit of Friedman and Schwartz.” in Olivier, Jean Blanchard and Stanley Fischer, eds., NBER Macroeconomics Annual 1989 (MIT Press, 1989), pp. 121-70. Cover, James P. ‘‘Asymmetric Effects of Positive and Negative Money Supply Shocks,” Quarterly Journal of Economics (November 1992), pp. 1261-82. De Long, J. Bradford, and Lawrence H. Summers. “ How does Macroeconomic Policy Affect Output?” Brookings Papers on Economic Activity (1988:2), pp. 433-94. Rotemberg, Julio, John Driscoll and James Poterba. “ Money, Output and Prices: Evidence from a New Monetary Aggre gate,” mimeo 1992. MARCH/APRIL 1993 43 W. Lee Hoskins IV. Lee Hoskins is president and chief executive officer of The Huntington National Bank in Columbus, Ohio. This paper is given in honor of Ted Balbach and his service to the Federal Reserve Bank of St. Louis. His resolute pursuit of sound economics as the bedrock of monetary policymaking and his indomitable spirit, even when the policy process ran amok, has served us all well. I thank John Davis, Sandra Pianatto and members of the Research Department of the Federal Reserve Bank of Cleveland for helping to shape and advance my views on monetary policy during my four years with them. Views on Monetary Policy T „ HE IDEAL MONETARY POLICY requires a credible and predictable commitment to main tain the long-term purchasing power of a currency. The performance o f central banks, which have traditionally been entrusted with monetary policymaking, is far from this ideal simply because a clear mandate for price-level stability—zero inflation—is absent. In practice, central banks serve as instruments that govern ments use to pursue multiple objectives that they believe serve their interests. Therefore central banks pursue monetary policies that at best have only a fragile commitment to price stability. Governments are currently pursuing policy coordination or monetary union strate gies that are little more than attempts to implement a regime of monetary protectionism, in the global economy. The future o f monetary policy rests on the continuing struggle between politi cians seeking policies that serve their short-term agendas and global financial markets that limit the actions o f an individual central bank. In my remarks I discuss why central banks have been established, their bias toward infla tion and the importance o f independence and accountability to their effectiveness. I also argue that zero inflation should be the dominant objective of a central bank and that current efforts to coordinate monetary policies are likely to conflict with that objective. WHY CENTRAL BANKS? What is the justification for a central bank? Can some configuration of private institutions in a so-called free-banking environment perform the functions o f a government-sponsored mone tary authority? Are central banks necessary? In his 1959 Millar Lectures at Fordham University, Milton Friedman provided a classic statement o f the economic rationale for central banks.1 Friedman’s argument appealed funda mentally to the costs inherent in a pure commoditystandard system, for example, a gold-standard system. These costs arise both from pure resource costs and perhaps more significantly from sub stantial short-run price variability resulting from inertia in the adjustment of commodity-money supply to changes in demand. The inefficiencies these costs represent are a significant disadvan tage o f commodity-money exchange systems. As a consequence there is a natural tendency, borne out by history, for pure commodity standards to be superseded by fiat money. But particular aspects of fiat money systems—such as fraudulent ’ These lectures were subsequently published as A Program for Monetary Stability. MARCH/APRIL 1993 44 banking practices, natural monopoly characteristics and tendencies for localized banking failures to spread to the financial system as a whole—resulted in the active participation of government. W e have come to know this active participation as cen tral banking. Rationales for establishing central banks have not gone unchallenged, not even by Friedman.2 Disruptions in payments can be costly, but so are the instabilities and inefficiencies caused by the lack of an effective anchor for the price level in fiat money systems. Moreover, theoretical discoveries in finance and monetary economics, closer attention to the lessons of historical bank ing arrangements and advances in information and financial technologies have contributed to a healthy skepticism about the superiority of central banks and government regulation to alternative market arrangements. For example, some o f the financial-backstop functions performed by central banks and banking regulators may have weakened private market incentives to control and protect against risk.3 Still, those who argue for alternative monetary structures must at least recognize that their case rests on untested propositions. Yes, it would be wrong to accept unthinkingly our current central banking system as the best alternative for performing the monetary functions of advanced economies, but it would also be wrong to claim that the current central banking system does not reflect society’s choice of an institutional arrangement to perform those functions. It is not sufficient to argue that m arketoriented alternatives to our current central banking systems functioned better in other times and places, for example, in 18th-century Scotland.4 This begs the question of why such a system did not prove to be sustainable. Nor is it sufficient to argue that this system would have prevailed if not for government intervention and interference. This line o f debate fails to consider whether a political equilibrium that would support a market-oriented system in an advanced economy exists anywhere. It is premature to claim that some hypothetical monetary system can or should dominate institu tional arrangements that have already evolved from extended political and economic experience. I believe that the prudent first course is to con 2See Friedman and Schwartz (1986) 3See Goodhart (1988). FEDERAL RESERVE BANK OF ST. LOUIS sider the advantages of improving the perform ance of central banks. The benefits o f a properly managed fiat currency are considerable, and the issue is or should be how to provide the central bank with a proper charter to ensure policy action that generates price-level stability in the long term. If such efforts fail, market alterna tives should be sought. Because I am most familiar with the Federal Reserve, let me use it as an example. Before the creation of the Federal Reserve in 1913, the country prospered without a central bank. Broadly speaking, the impetus for creating the Federal Reserve was a series of banking panics that led to contractions in money and credit that in turn caused serious disruptions in eco nomic activity. The nation sought to improve its banking system by establishing a means for providing an elastic money in the context of a monetary standard based on full convertibility to gold. The gold link was severely weakened by the Gold Reserve Act of 1934. The Federal Reserve was the result of a com promise between those who would have kept the banking system entirely private and those who wanted government to assume a prominent role in a rapidly growing economy. Other nations have grappled with the same problems and created similar institutions. Today many republics of the former Soviet Union and several eastern Euro pean nations are facing these same issues. We now have a world monetary system in which governments, through central banks, monopolize the supply and management o f inconvertible fiat monies. The displacement o f the commodity standard that prevailed at the time the Federal Reserve was founded has exposed problems not other wise envisioned in 1913. For example, the price level has no anchor except for that provided by the resolve of Federal Reserve policymakers. The quadrupling of prices since 1950 dramati cally demonstrates the failure of Federal Reserve policymakers to provide such an anchor for the monetary exchange system. Fed policymakers’ commitment to price stability is neither as explicit nor as strong as necessary for the successful management of a fiat currency. The gradual demise of our convertible monetary standard has brought us to a point that requires a basic 4For a discussion of the free banking era in Great Britain, see White (1984). 45 change to the framework within which the Federal Reserve functions if the benefits of a fiat currency are to be achieved without large offsetting costs. The evolution o f the global monetary system reflects a common, though unstated, acknowledg ment that the benefits of a fiat monetary standard are substantial. Wise administration of that standard requires a central bank in some capacity. In this context, the essential issue is this: How can nations achieve the benefits o f a fiat money standard and simultaneously constrain the exercise of that power to the service of the public good? Put another way: How can a nation prevent its central bank from debasing the monetary standard it is charged to protect? INFLATIONARY BIAS OF CENTRAL BANKS The answer to these questions seems to elude us. Witness the universal debasement of currencies by central banks since the loss of a commodity standard as a price-level anchor. To find the answer, we must review central bank charters and the incentives provided to those who con trol monetary printing presses. Public-choice economists have focused on this issue and developed a rich literature; however, I feel they fail to provide a satisfactory explanation of the secular bias toward inflation among central banks (with different charters and varying degrees of independence from political influence). Moreover, this approach fails to explain why in earlier periods governments did not consistently exploit the opportunities to inflate by realigning their currencies against gold or dropping their convertibility. Another explanation for persistent inflation that has some appeal is policy mistakes, or inap propriate targets or operating procedures of central banks. This explanation also leaves some unanswered questions. Why aren’t policy mistakes symmetrical? That is, why don’t they cause defla tions as well as inflations, leaving the average price level unchanged over time? Perhaps policy mistakes are biased toward inflation because of the operating procedures employed, such as interest rate targeting. Yet the Bundesbank, which uses monetary aggregate targets, produces a rising price level. The Bank of Japan uses interest rate targets and has generated a similar increase in its price level during the past two decades. If a central bank is dedicated to price-level stability over time, the choice o f targets or operating procedures probably only influences the variabil ity o f inflation rates around a zero mean. In short, a central bank that truly wants to achieve pricelevel stability can do it with any number of operating techniques, as long as they control money growth over time. Perhaps a simple, and less elegant, explanation for persistent inflation is that central bankers are suffering from a Keynesian hangover. Central bankers, politicians and the public are merely reflecting the prevailing economic dogma that government has the responsibility and ability to manage aggregate output and employment, as well as inflation. I have argued and continue to believe that a major source o f price-level insta bility comes from multiple objectives assigned to central banks—economic growth, employment, price stability and exchange rates. It is true that politicians pressure central banks to achieve dif ferent objectives at different times. Such politi cal pressure can produce inappropriate policy actions; however, the responsibility for assign ing multiple objectives to central banks rests as much with the economics professions as it does with politicians. For the last 50 years, many economists have supported various theories of business-cycle management, which required that central banks shift from one objective to another. Today businessmen, politicians and most economists continue to believe that if the economy is weak, the central bank should respond regardless of the cause of the weakness. And so it does. Some o f the current discussions about mone tary policy and the Federal Reserve suggest that the lessons of the 1970s may be fading from our memories. Calls for lower interest rates or more rapid money growth are not at all unusual. More often than not, those suggestions seem impelled by desires for growth or desires to o ff set the problems o f particular sectors of the economy. They seem based on the notion that there is a trade-off between inflation and output or between inflation and employment that can be exploited by the central bank. Some o f us learned from the experience of the 1970s that such a trade-off does not occur over time. Instead, higher inflation only added to uncer tainty, distorted resource allocation and reduced economic performance below the maximum sus tainable level with price stability. Members of a central bank policy committee such as the Federal Open Market Committee (FOMC) reflect what is believed by the mainstream. MARCH/APRIL 1993 46 In January 1990 the National Association of Business Economists surveyed its members and asked the following question: "Is reducing the inflation rate to zero over the next five years the appropriate objective of monetary policy?”5 More than 80 percent of the respondents answered no. Their responses indicate that they believe the FOMC should trade o ff inflation for some other objective, presumably economic growth. At about the same time, the House Sub committee on Domestic Monetary Policy sur veyed 500 members o f the American Economics Association who list monetary economics as either their first or second specialty. The unpublished survey shows that only a slight majority o f those who responded favored zero inflation over the next five years. I believe that much of the inflationary bias of central banks over the past 50 years reflects the prevailing view that output and employment fluctuations can be smoothed with monetary policy. Currently, before each FOMC meeting, members of the Committee are presented with the policy views of several prominent economists. Either explicitly or implicitly, these views invari ably present the policy choice in terms of a Phillips curve trade-off. Staff projections at the FOMC meeting also imply such a trade-off, as do the statements by some FOMC members. Moreover, policy actions, such as a reduction in the federal funds rate, often follow the release o f employment or output statistics, further rein forcing the notion that the Federal Reserve can manage real variables. To the extent that this explanation o f central bank behavior is valid, inflationary bias will not be eliminated until there is agreement within the profession on price-level stability as the dominant objective for central banks. INDEPENDENCE AND ACCOUNTABILITY The problems that emanate from multiple, and often incompatible, objectives are well known. To contribute to maximum economic growth over time, central banks must achieve price-level stability. Achieving this goal requires that central banks be free from political expediencies—that is, that they have independ ence within government. Substantial evidence indicates a link between central-bank independ 5See NABE Policy Survey (1990). FEDERAL RESERVE BANK OF ST. LOUIS ence and the ability to achieve price stability. Recent studies show that countries that grant their central banks the greatest degree of inde pendence have had the lowest rates o f inflation.6 Even taking into account other sociopolitical factors that might cause inflationary pressures, the degree o f central-bank independence appears to have an important effect on a country’s infla tion rate. However, with independence must come accountability. Even the clearest objectives will prove elusive without accountability; independence without direct accountability is a dangerous brew for those who drink it. Great harm has come from well-intentioned, independent cen tral bankers with little or no accountability— witness the United States in the 1930s. Many mechanisms exist today to bring accountability to central banking; for example, the employ ment contract of the governor o f the central bank of New Zealand contains a price-stability requirement. The objectives, degree of independence, and accountability o f the central bank are substan tially determined by its legal structure. For example, a clear legislative directive to achieve price-stability goals above all others and the freedom to pursue price-stability initiatives would all but eliminate potential conflict with other objectives. The vexing question o f what extent, if any, a central bank should compromise the price-stability objective to pursue auxiliary goals, such as smoothing real output fluctua tions or stabilizing exchange rates, should be resolved and dictated in the legislative charter. True independence and strict accountability can be attained only legislatively. Compared with the central banks of other countries, the Federal Reserve System has a better structure to execute monetary policy effectively; however, the Fed is not as well posi tioned as other central banks. The Federal Reserve is charged with multiple objectives that are often incompatible but that at least include price stability. It is functionally independent within government, but it faces intermittent challenges to its autonomy. Its independence comes from both its charter and its practice. Independence is essentially a delineation between the responsibilities o f Congress and the executive branch on one side and the monetary 6See Alessina (1988) and Banaian (1983). 47 authority on the other to limit the motive and means to debase the value of the nation's money. The source of tension between monetary and fiscal authorities is the central bank’s ability to create money. Because the creation of fiat money imposes an implicit tax on money balances, the monetary authority is one source of government revenues. For the most part, the long-run viability o f the government's fiscal operations requires that its real current debt burden plus the present value of its expenditures equal the present value of revenues. Thus if the path of debt plus expendi tures diverges from the path of explicit tax rev enues, fiscal viability requires that the discrepancy be satisfied by seigniorage from monetary growth. This scenario is typically referred to as fiscal dominance over the monetary authority. The original Federal Reserve charter left many doors open for the executive branch to influence monetary policy. These were partially closed when the Banking Act of 1935 removed the Secretary o f the Treasury and the Comp troller of Currency from the Board of Gover nors of the Federal Reserve System. In addition, the law established the FOMC, with the seven governors and five o f the Federal Reserve Bank presidents as voting members, ensuring that power within the Federal Reserve would be shared between political appointees and regional bank presidents. Thus the fire wall that made the Federal Reserve, and not the executive branch, responsible for monetary policy objectives was reinforced. It was strengthened further by the Treasury-Federal Reserve Accord o f 1951, which served as a clear statement that the Fed would not be coerced into solving the federal govern ment's debt-management problems. The institu tional structure was designed to ensure enough Federal Reserve independence within the govern ment to carry out this mandate without interference. This independence in principle has held up in practice. The dramatic increases in federal deficits in the early- and mid-1980s prompted fiscal dominance believers to predict that it would be impossible to achieve and maintain inflation rates below the disastrous levels of the decade’s start. So far, this prediction has not come to pass. In 1983 the federal budget deficit was 3.8 percent of GNP, a level far above the postWorld W ar II average and nearly equal to the postwar peak realized in 1975. In the same year, inflation measured by the consumer price index fell to 3.2 percent—a 16-year low. As the decade proceeded, the deficit relative to GNP rose, fell, and rose again to its present level above 5 percent. The inflation rate was impervi ous to these patterns. Astute observers might question the relevance of the early- and mid-1980s to the fiscal dominance proposition, because deficits as they are conven tionally measured do not necessarily reflect the government’s long-run fiscal operations. To name just a few o f the problems, the value o f longrun government net liabilities is inherently ambiguous, the path of future revenues is un certain and the appropriate method of discounting future tax and expenditure flows is problematic. Although sympathetic to this view, I am still left with the strong suspicion that if any period in recent history was ripe for the emergence of fiscal dominance, it was the last 10 years. Indeed, as the decade progressed and the predictions of the fiscal-dominance theory failed to materialize, more sophisticated variants of the relationship between fiscal and monetary policy began to find their way into economic research. The fiscal authority’s reign over the subservient monetary authority was replaced by a more subtle and complicated institutional structure, a world in which fiscal and monetary authorities played a game of chicken, the out come of which left both parties less than fully satisfied.7 Although deficits may be detrimental to economic performance, the ability of the Federal Reserve to resist monetizing debt has protected the economy from even worse conse quences. The Federal Reserve’s decision to resist monetizing the federal debt resulted in lower infla tion and contributed to fiscal reforms that started with the Gramm-Rudman-Hollings legislation. In my view the Federal Reserve has sufficient independence to achieve price stability. The coreproblem, however, is that the Federal Reserve is not accountable for that objective. Without accountability, the policy process will be neither credible nor predictable. The more credible the commitment to the policy goal, the few er wrong decisions will be made by the markets. The more predictable the policy reaction to unforeseen economic events, the more limited will be the market reaction to those events. Credibility and predictability can substantially lower the costs 7See Sargent (1985). MARCH/APRIL 1993 48 of achieving and maintaining a stable price level. Yet with the disintegration of the monetary aggregates as intermediate policy guides, discre tionary monetary policy actions may seem espe cially hard to predict because policy objectives and accountability for them are unclear. The exist ing policy process, with its focus on short-term economic or financial developments does not provide credibility. How can we change the process to reinforce the credibility o f a consistent goal? I think the most secure way would be to give the FOMC a legislative mandate to meet a consistent, attainable and unchanging economic goal. Passage of House Joint Resolution 409, introduced by Rep resentative Stephen Neal, would provide that crucial reinforcement. The Neal resolution simply directs the Federal Reserve to make price stabil ity the primary goal of monetary policy and to achieve that goal within five years. History gives us little basis for expecting price stability or even a stable rate of inflation because the FOMC has had no mandate to produce that result. Giving the FOMC that mandate and knowing that the FOMC intended to stabilize the inflation rate at zero, would provide one gigantic piece of policy information to any rational decision-maker in any dollar-denominated market. The Federal Reserve would remain independent, and it would retain complete discretion about how to carry out policy. The only change would be that Congress would provide more direction about the basic policy objective, and the Federal Reserve would be accountable for achieving it. True accountability would also require an incentive or enforcement mechanism for achiev ing the objective. The FOMC can deliver lower inflation without a legislative mandate. Of that you should have no doubt! Inflation is a monetary phenomenon, and the FOMC is the sole custodian of the quan tity of money in the United States. If a zeroinflation mandate were in effect, short-term deviations from zero inflation might occur, but one way or another the FOMC could provide a stable price environment. As many scholars have urged, the FOMC might impose accounta bility on itself by tying policy actions to some intermediate target variable by an agreed-on formula that would ensure price stability. These days, the most popular candidates for an inter mediate policy target seem to be nominal GDP and M2, either of which is thought capable of producing reasonable price stability. Another FEDERAL RESERVE BANK OF ST. LOUIS approach would be for the Committee to specify achieving the ultimate policy goal as the rule, while using discretion in choosing actions to achieve the goal. Of course having today’s FOMC impose account ability on itself (by adopting an explicit rule tying an instrument to a goal) is not a foolproof way to achieve an official policy goal. Credibility would have to be earned through predictable actions consistent with the goal. To adopt an explicit rule, at least a majority of today’s FOMC members must not only agree on an overriding macroeconomic goal, but also renounce some discretion to pursue other goals. Moreover, tomorrow’s FOMC could decide to change the goal and hence the rule. In the current policy regime, today’s policy choice can in no way bind tomorrow's. Unless directed by society through specific mandate, tomorrow’s FOMC always has the discretion to change the goal. And with shifting goals there is no accountabil ity. I believe that the lack of accountability for a dominant policy goal of price stability is the major cause of the inflationary bias in the U.S. economy since W orld W ar II. Although the specifics of the Federal Reserve charter differ from those of other central banks, the problems of conflicting objectives and the lack of secure independence and explicit accountabil ity are common to all central banks in varying degrees. Experience around the world and through time repeatedly demonstrates that cen tral banks require independence from day-today political life to perform their price-stability role. If we could create legal and cultural condi tions that truly fix a central bank with account ability for anchoring the price level, the structure of the central bank itself would become less important. Those circumstances would be a joy to behold, but I am afraid they will be some time in coming. W HY A ZERO-INFLATION OBJECTIVE? I strongly believe for three reasons that the dominant objective of monetary policymakers should be price stability. First, in the long run, a central bank can control the price level of goods and services denominated in its own cur rency, but it cannot control the growth of out put (potential or actual). Second, a credible commitment to a price-stability objective enables a central bank to promote economic efficiency 49 and growth (potential and actual). Third, pricelevel stability, popularly called zero inflation, is superior to inflation-rate stability. Among economists, support for the first reason is nearly universal. There is also widespread agreement on the second point. A central bank that pursues price stability promotes economic efficiency and growth. I would venture further to say that experience shows that central banks that have sought to enhance economic growth directly have failed miserably at providing sta ble price levels and ironically have undercut eco nomic growth in the process. The last reason—that no inflation is preferable to stable, non-zero inflation—is most contentious, particularly when people attempt to compare the transitional costs of achieving price stability with the costs of stabilizing the inflation rate at the status quo. The argument that the cost of pursuing a zeroinflation target would outweigh the benefit of reaching that target has two dimensions. The first is that the benefit of achieving zero inflation would be small. The second deals with the costs of moving from a 4 percent trend rate of infla tion to zero inflation. This is the transition-cost argument, which essentially says that even if zero is the place to be, getting there is not worth the ride. I believe that the benefits of zero inflation are great and that the transition costs can be reduced if the Federal Reserve commits to an explicit plan for achieving zero inflation. The interaction between inflation and our cur rent tax system, especially as it applies to income generated by capital, represents one of the more significant channels through which non-zero inflation can exact economic costs.8 This channel of distortion is often not taken seriously because people think that its effects are minimal or that it would be easy to index the tax system. Correcting the tax code is a good idea of course, but until that happens, what possible excuse is there for not letting the monetary authorities do what is necessary to improve social welfare? It is clear that the horrendous U.S. inflation ary experiences o f the 1970s and early 1980s created the impetus for the limited inflation indexation of the current tax system; however, the job is far from complete. Capital gains, cor porate depreciation and interest expenses, and personal interest income remain untouched by 8See Altig and Carlstrom (1990). efforts to index the tax system for inflation. Even the bracket indexation implemented by recent tax reform does not fully protect tax payers from bracket creep, or nonlegislated increases in marginal tax rates created by infla tion. Complete indexation o f the tax code, how ever desirable it may be, will be extremely difficult to achieve. Will another inflationary experience like that of the 1970s be required to induce further progress on tax indexation? I fail to understand why some feel that these inflation-tax interactions are a significant drag on the economy, yet argue that only Congress should be concerned with the problem. The prob lem exists because o f the interactions between inflation and a tax system based in current dol lars. Therefore it seems that the responsibility for minimizing these costs lies as much with the monetary authorities as with Congress. Doesn’t it make more sense for monetary authorities to try to correct the inflation part of the problem rather than simply hoping that Congress will implement changes that it may be unable or unwilling to pursue? W e speak about the costs of achieving zero inflation, but what about the costs o f fully indexing the tax system? Surely they would be significant. Another area of concern is the role of uncertainty as a source of inflation costs. How important are the distortions that arise from price—level uncertainty? There is a class of models—the market-clearing, imperfect-information paradigm associated with Robert Lucas and others—in which inflation uncertainty harms the economy by distorting the period-to-period relative price signals that facilitate the efficient allocation of scarce resources.9 Despite the pervasive intellec tual influence exerted by the Lucas framework to this day, the empirical evidence accumulated since the development o f the paradigm in the early 1970s has not been entirely supportive. This point is not lost on critics, who think that the lack of evidence on short-term distortions should persuade us that inflation uncertainty is simply not that important to social welfare. Surely the relative-price/aggregate-price confu sion stressed by the Lucas-type models is a spe cial type of uncertainty. The failure to find significant effects from uncertainty that is resolved within a few quarters tells us next to nothing about the type o f long-run uncertainty with which the zero-inflation position has always been fundamentally concerned. 9See Lucas (1972). MARCH/APRIL 1993 50 Indeed, it seems likely that the uncertainty occurring over extended time horizons is pre cisely what is most affected by the average inflation rate.1 This is one reason why I favor a 0 price-level target. An inflation-rate target ena bles the price level to drift without bound, and with no enforcement mechanism to ensure that inflation mistakes will be corrected, the longrun variance of the price level is infinite. When people have reason to believe that this standard will erode over time, they invest numerous resources to protect themselves. Those who have nominal debt outstanding will drag their feet in paying it back, whereas creditors will invest in ways to accelerate the collection of funds. The private gains to self-protection are clear, as are the social costs. Recent experience is the best testimony to the real resource cost of inflation. During the 1970s, people could see that inflation acceler ated each year. They guessed, reasonably at the time, that financial assets had limited value in protecting their wealth from inflation. Conse quently, farmland, commercial and residential property, and precious metals became much more expensive as people sought to shelter their wealth. Not only was time spent seeking these investments, which was socially wasteful, but also the resource misallocation itself resulted in a great waste o f land, labor and capital that society is still paying for today. assume some frictions. These frictions, coupled with the inability o f markets to clear, make end ing inflation appear as costly as it does. Isn’t it sensible to assume that the implicit sources of frictions that make lowering the inflation rate costly would also contribute to making inflation costly in and of itself? For instance, a variety of explicit and implicit nominal contracts already exist, and a transition to zero inflation could alter the real values of payments from those that were originally intended. But surely the entire institutional apparatus that generates these con tracts must involve resource costs that are posi tively related to the average rate of inflation. One should not compare the costs of achieving zero inflation in non-market-clearing models, where such costs are high, to the benefits of being at zero inflation in frictionless, continu ously clearing models, where the benefits are low. If we use a model with frictions to meas ure the cost of getting to zero inflation, then we should also use such a model to examine the benefits o f being there. This is one reason I am skeptical of so many cost/benefit estimates of reducing inflation. It is difficult to comprehend how efficient planning within the public and private sectors could not be inhibited by this type of long-run uncertainty. Furthermore, the intuition that longrun inflation uncertainty is costly has empirical support. In cross-country comparisons, economic growth is negatively related to the variability of inflation.1 One finds that the case for reducing 1 price level uncertainty is far more compelling than a cursory analysis might indicate. I am also skeptical about transition-cost estimates that do not account for the possibility that a price-stability objective will be regarded as credible by the public. Economic theory and reasonable model simulations persuade me to believe that with credible precommitment, a central bank can greatly minimize private-sector planning errors during the transition period. I think that much of the disagreement among economists on the size of transition costs cen ters on the ability of a central bank to commit itself credibly to achieving its objective. Until I see some hard evidence to dissuade me, I plan to continue my advocacy o f price stability as the overriding objective of central banks. In evaluating the costs of attaining zero infla tion, economists almost always use models in which markets do not clear or do not clear without cost. Gone is the market-clearing, flexible-price, rational-expectations model. In its place is a model with price contracts that make the transition to zero inflation extremely costly. The source of the friction is usually not entirely explicit, but the implication is that we must It still puzzles me that volumes of research have been published on central bank operating proce dures and management of monetary aggregates, yet relatively little research lias been published on the value of a credible precommitment to a price-stability objective. My intuition tells me that the latter is far more important than the former in terms of economic welfare. Of course, credibility depends on policy information avail- 10See Ball and Cecchetti (1990). "S e e Grier and Tullock (1989) and Lebow, Roberts and Stockton (1990). FEDERAL RESERVE BANK OF ST. LOUIS 51 able to market participants so that they can monitor progress toward the objective. One major benefit of imposing an explicit inten tion on monetary policy is that policy actions in the money market would become far less momentous than they are now. Currently, detecting a change in the federal funds rate target from the pat tern of open market operations is crucial because it provides markets with one of the few clues as to what monetary policy the Federal Reserve is pursuing. Canvassing the positions of individual FOMC members is a way of predicting future policy. If policy intent were explicit and credi ble, however, finding the clues in open market operations would have less significance. I see the greatest payoff in more information about policy intentions. An explicit FOMC com mitment to price stability would allow markets to shift resources from watching the Federal Reserve to watching the economy for produc tive investment opportunities . Focusing on the intent of policy contrasts markedly with conven tional concerns for more certainty about the current degree of reserve restraint. There are many ways to reduce uncertainty about the immediate funds-rate implications of policy, just as there are many time schedules by which the FOMC directive might be released. More certainty about the immediate policy implications of the federal funds rate might make Fed-watching a bit easier, but it would not do much to help identify policy intentions beyond short horizons. Releasing Fed directives early might provide a slightly brighter glimmer o f policy intentions, but only for a slightly longer policy horizon. We do not need better information about the latest directive; we need better information about the process through which all future directives will be crafted—that is, policy inten tions. Nothing would provide more insight than a clearly stated goal MONETARY POLICY AND MONETARY PROTECTIONISM Let me turn now to the effects of interna tional policy coordination on the pursuit of zero inflation.1 Exchange-rate regimes and attempts 2 at monetary union are currently undermining the price-stability objective. Many actions taken by central banks are not aimed at price stability, but rather are attempts to establish monetary protectionism. By monetary protectionism, I refer to attempts to alter real exchange rates through manipulation of monetary policies and with the hope of ultimately promoting a balance-ofpayments objective. In the case o f a deficit country, monetary protectionists call for an expansion o f money growth (or lower nominal interest rates). A monetary expansion, other things being equal, will produce a nominal depreciation. If individuals are unable to adjust prices immediately, or if they are slow in per ceiving the inflationary aspects of this policy, a real depreciation will accompany the nominal depreciation. As most economists realize, how ever, the inflation rate will eventually respond to the monetary expansion, offsetting the nominal depreciation and returning the real exchange rate to its initial position. Nevertheless, the tenuous, short-lived relationship between money and the real exchange rate is seductive enough to con vince politicians and other fine-tuners that monetary policy can serve mercantilist designs. My focus on this issue stems from a firm belief that central banks can do no better than guar antee long-run price stability and that any efforts to limit this guarantee are not likely to raise world welfare. Central banks can juggle a real exchange rate and inflation target no better than they can slide back and forth along a stable Phillips curve. A central bank that attempts to maintain price stability and a nominal exchange rate target has more policy targets than policy instruments. At times, these two objectives might be compatible. For example, in the late 1970s, limiting rapid dollar depreciation through inter vention could have been compatible with a conractionary monetary policy to eliminate inflation. As often as not, however, these two policy objectives will be incompatible, and the central bank must trade one objective for the other. Under such conditions, markets will view nei ther price stability nor exchange-rate stability as a credible policy. The knowledge that central banks will deviate from a policy of price stabil ity to pursue an exchange rate objective will raise uncertainty about real returns and will distort the allocation of resources across sectors and through time. The resources devoted to protecting wealth from possible inflation could ,2This section summarizes ideas presented in Hoskings and Humpage (1990). MARCH/APRIL 1993 52 be applied to more productive uses under a pol icy of price stability. Moreover, attempts to maintain nominal exchange rates will not eliminate exchange rate uncertainty because countries will inevitably resort to periodic exchange-rate realignments. Hedging exchange risk will remain an important aspect of international commerce. Although monetary protectionism seems most prevalent under the present system of floating exchange rates, it does not follow that floating exchange rates promote its use. Monetary pro tectionism can result any time a government accepts nonmarket criteria for exchange rates. In principle, a gold standard or a fixed exchange rate regime can limit the scope of monetary protectionism because, if all countries play by the rules of the game, they link money supplies closely to the flow of international reserves. In practice, however, such regimes do not destroy the political motives for monetary pro tectionism, and examples o f monetary protec tionism under fixed exchange rates abound. By allowing some discretion in the choice of exchange rate adjustments, fixed exchange rate regimes often produce a mechanism that weakens the allocative efficiency of exchange markets and promotes mercantilist objectives. In contrast to the interventionist literature, which presupposes an all-wise government act ing in the public's best interest, a rich, growing literature on political economy characterizes elected officials as seeking to enhance their own power, prestige and wealth by maximizing their ability to gain votes. Politicians and bureaucrats attempt to extend the scope of their influence by responding to the demands of the most polit ically active constituencies.1 A political justifica 3 tion for exchange rate manipulation is that it defers criticism and postpones more fundamental actions. For instance, in 1985 dollar exchange rates were at their zenith, the U.S. current account was deteriorating rapidly and evidence sug gested that the United States was becoming a debtor country for the first time since World War I. U.S. manufacturers, facing increasingly stiff competition worldwide, besieged Congress for trade legislation. Most important, analysts increasingly linked the deterioration in the exter nal accounts with the fiscal policies o f the Reagan 13See Quibria (1989). 14The Group of Seven countries are Canada, France, Italy, Japan, the United Kingdom, the United States, and West Germany. FEDERAL RESERVE BANK OF ST. LOUIS Administration and Congress. The opportunity cost of government inaction, measured in terms of votes lost, seemed to rise sharply in the early 1980s. The U.S. current account deficit reflected imbalances between savings and investment in the United States, West Germany and Japan. Politicians, however, cannot easily redress such structural relationships through fiscal policies because of strong vested interests in maintaining various tax and expenditure patterns. Unable to address these structural problems directly and quickly, policymakers might resort to exchangemarket intervention. When coordinated through the Group of Seven, such intervention offers a highly visible signal that governments are respond ing to the desires of their constituencies.1 4 Exchange rate policies can also offer temporary benefits to specific constituencies. When goods prices are slow to adjust, a nominal currency depreciation is equivalent to a temporary, across-the-board tax on imports and a subsidy to exports. With the terms o f trade temporarily altered, certain groups in the traded-goods sectors can realize benefits from monetary protectionism similar to those afforded by more traditional forms of protectionism. Ultimately, any benefits from monetary protectionism dissipate with a high inflation rate and with reduced credibility of monetary policy. The inflation costs of mone tary protectionism, however, are dispersed across a wider spectrum of individuals and over a longer time horizon than the benefits. A constit uency that receives net benefits from monetary protectionism (export- and import-competing firms) can exist. Such a constituency is likely to be more politically cohesive than any constit uency for price stability. Consequently, a policy that seems myopic from an economic perspec tive can be politically attractive. Another seemingly attractive aspect of mone tary protectionism is that Congress and the administration can justify it in terms o f broader macroeconomic considerations, such as exchange rate misalignment or current account imbalance, instead of industry—specific considerations, such as automobile and steel employment. Consequently, the rent-seeking aspects o f mone tary protectionism are less obvious than those of standard protectionist policies. 53 Countries interested in establishing exchange rate targets have a strong incentive to collude in their efforts with foreign governments.1 In 5 the case where countries attempt to alter nomi nal exchange rates, such collusion provides tacit foreign approval of these policies and limits the chances that a foreign government will take steps to neutralize the exchange policies of another government. Sometimes such collusion involves having cartel members delay policy negotiations, or exchange rate adjustments, when individual cartel members face critical elections. Bretton Woods and the European Monetary System (EMS) are examples of collu sion that were fairly successful for a period. The competitive currency devaluations of the 1930s show what can happen when govern ments attempt to fix a price but their cartel breaks down. Coordinated efforts to fix exchange rates can allow individual countries to influence the policies of others and to defer some of the adjustment burdens o f maintaining the peg. Such mechanisms are found in the EMS and figure in some proposals for target zones and for fixed exchange rates. Many support the pro posal for a European Central Bank for just this reason. The alternative is to sacrifice monetary sovereignty to maintain a fixed exchange rate and to follow the monetary policy of a major trading partner. Under floating exchange rates, a rapid depreciation in the nominal exchange rate in response to such inflationary policies signals the market's displeasure and constrains governments. Through collusion to fix the exchange rate, however, governments can temporarily blunt the exchange rate reaction to their policies and reduce the political costs of pursuing inflation ary policies. Coordination to limit exchange rate fluctuations is politically attractive because it eliminates an important, immediate barometer of the market’s opinion of government policies. For their part, central banks often are willing participants, viewing exchange rate management as a legitimate aim o f monetary policy. Exchange rate movements can impart useful information for policymaking, and as already noted, exchange rate targets can sometimes be consistent with a monetary policy of price stability. As often as not, however, exchange rate policies conflict with price stability. For example, U.S. purchases of foreign currencies in 1990 seemed inconsis 15See Vaubel (1986). tent with a goal of price stability. When these objectives conflict, the Federal Reserve System is torn between its independence and its accounta bility to the broad national policy goals set by Congress and the Administration. The Federal Reserve does not wish to appear to the public as unresponsive to the objectives of Congress and the administration. Participation also enables a central bank to influence policy formulations that it is powerless to prevent. Such reasoning is a certain sign of a central bank unsure of its objective and insecure about its independence. In countries with independent central banks, intervention policies might enable fiscal agents to extend their influence beyond the foreign exchange market to domestic monetary policy. Elected officials often seek more stimulative monetary policies than do central banks, hoping to lower nominal interest rates and to stimulate real growth and employment. In choosing a nominal exchange-rate target, intervening and encouraging the central bank not to sterilize the intervention, fiscal agents have a mechanism for such influence that would usually not be open. At times, however, such as when the central bank policy committee is not in unanimous agreement, such an influence, marginal though it may be, could prove decisive in charting future monetary policy actions. INTEGRATED MARKETS AND POLICY CONSTRAINTS I have attempted to instill a healthy skepticism for exchange market manipulation, arguing that it is a form o f monetary protectionism that harms economic welfare. Monetary protectionism stems as a near-term palliative from the political inter actions between policymakers and constituen cies with vested interests in particular market outcomes. Any international monetary order willing to accept nonmarket criteria for exchange rates and failing to bind governments with a price-stability objective is ripe for monetary pro tectionism. To counter the political incentives toward monetary protectionism, nations should adopt monetary mandates, such as the Neal Resolution in the United States, that focus monetary policy on achieving and maintaining long-term price stability.1 This would do more 6 to eliminate exchange market uncertainty and foster the efficient worldwide use of real ,6See Hoskins (1990). MARCH/APRIL 1993 54 resources than any program to manipulate nominal exchange rates. My comments are not meant as a blanket con demnation o f international policy cooperation. I strongly support cooperation that makes price stability the dominant objective and recognizes market-determined exchange rates. Only cooper ation based on these conditions seems both feasible and credible because it recognizes that nations want monetary sovereignty and will pursue different economic policy objectives. Contrary to what some might infer, this approach does not preclude European monetary unification in the future, but it suggests a different approach than currently seems to be favored. European governments are not likely to relinquish national monetary sovereignty on adoption of a single market. Consequently, greater exchange rate flexibility than the EMS currently provides seems necessary to ensure that exchange rates do not interfere with the efficient flow of goods, labor and capital following the removal of restrictions. The free flow o f resources, if it occurs, will foster a convergence of policy preferences within Europe as governments com pete for these resources by providing stable economic and political environments. Govern ments that fail to provide such an environment will lose resources as markets vote on policies. The resulting convergence of monetary and fis cal policies will lead to greater exchange rate stability. If in time, governmental competition for resources attains a convergence of macroeconomic policy, issues of national policy sover eignty will be muted. Only then will monetary union augment the efficiency gains of a single market. As seems obvious from recent develop ments in Europe, efforts to rush monetary union are efforts that put the cart before the horse and may well interfere with the progress toward a single market. To fix exchange rates before a convergence of policy preferences within the European Economic Community seems to ensure that interest rates and prices will bear more of the adjustment burden. Moreover, judging from the experience of Bretton Woods, fixed exchange rates would seem to guarantee speculators periodic exchange rate adjustments and to encourage governments to impede the flow of goods and capital through the reintroduction o f restraints. The dynamics of achieving monetary union are as important FEDERAL RESERVE BANK OF ST. LOUIS as the goal, and price stability is a more impor tant goal than either. Scores of new nations are busy constructing central banks to implement monetary policy. Using history as a guide, these new central banks will try to pursue objectives other than price stability, especially since they are being coun seled by central bankers with weak records on price stability. Short-term political agendas will likely dominate their policy actions and push them away from the pursuit of price stability. Yet it seems that there are powerful market forces that will crimp the efforts of central banks to mismanage their currencies. The integration of world markets, particularly financial markets, is limiting the degree to which policymakers are willing to drift away from price stability, at least for the major economies. Twenty years ago the Federal Reserve paid scant attention to the effect of foreign markets on the price of U.S. government securities and interest rates in the United States. Yet when I participated in FOMC deliberations, we almost always discussed the effect o f a policy action on long-term Treasury rates, currency values or the shape of the yield curve. The FOMC now looks at how world financial markets assess the credibility of its policy actions with respect to inflation expectations. This process, in effect, limits the degree to which the FOMC is willing to risk inflationary policy actions. In Europe, smaller countries often peg their currencies to the German mark, allowing the Bundesbank to determine their monetary pol icies. The German central bank is also limited by world markets in terms of the inflation path it chooses to pursue. I am not so bold as to argue that markets will cause central banks to wither away to agencies that simply pump out monetary growth rates that provide price stability. It does seem to me, however, that market forces are strengthening the hand of central banks in fighting political pressures for short-term "quick fixes” to economic problems. Perhaps even poli ticians will learn the limits of governments in solving economic problems. If this view proves incorrect, central banks will face the prospect o f market participants developing private money to a much greater degree than exists today. When government management o f particular institutions results in failure, private-sector alternatives appear— witness the privatization trend in U.S. schools 55 and courts. Perhaps those who yearn to revisit the Scottish system of free banking may live to see a version of it replace central banking. If so, we are likely to pay a heavy price along the way. REFERENCES Alessina, Alberto. “ Macroeconomics and Politics,” Macro economics Annual (The MIT Press, 1988), pp. 38-43 and table 9. Altig, David, and Charles T. Carlstrom. “ Inflation and the Personal Tax Code: Assessing Indexation,” Federal Reserve Bank of Cleveland, Working Paper 9006 (July 1990). Ball, Laurence, and Stephen G. Cecchetti. “ Inflation and Uncertainty at Short and Long Horizons,” Brookings Papers on Economic Activity, vol. 1, (1990), pp. 215-54. Banaian, King, Leroy O. Laney, and Thomas D. Willett. “ Central Bank Independence: An International Comparison,” Federal Reserve Bank of Dallas, Economic Review (March 1983), pp. 6-8. Hoskins, W. Lee. “ The Case for Price Stability,” Economic Commentary, Federal Reserve Bank of Cleveland (March 1990). _______ , and Owen F. Hum page. “ Avoiding M onetary Protectionism: The Role of Policy Coordination,” Cato Journal (Fall 1990), pp. 541-55. Lebow, David E., John M. Roberts, and David J. Stockton. “ Economic Performance under Price Stability,” unpub lished manuscript, Board of Governors (December 1990). Lucas, Robert E. Jr. “ Expectations and the Neutrality of Money,” Journal of Economic Theory (April 1972), pp. 103-24. NABE Policy Survey. Economists Expect Further Monetary Easing; Do Not Favor FOMC Restructuring; Oppose Govern ment Action to Curb Stock Market Volatility (National Association of Business Economists, 1990). Quibria, M.G. “ Neoclassical Political Economy: An Application to Trade Policies,” Journal of Economic Surveys (1989), pp. 107-36. Friedman, Milton, and Anna J. Schwartz. “ Has Government Any Role in Money?” Journal of Monetary Economics (January 1986), pp. 37-62. Sargent, Thomas J. “ ‘Reaganomics’ and Credibility,” in Albert Ando and others, eds., Monetary Policy in Our Times: Proceedings of the First International Conference Held by the Institute for Monetary and Economic Studies of the Bank of Japan (MIT Press, 1985), pp. 235-52. Goodhart, Charles. The Evolution of Central Banks (MIT Press, 1988). Vaubel, Roland. “ A Public Choice Approach to International Organization,” Public Choice, vol. 51(1), (1986), pp. 39-57. Grier, Kevin B., and Gordon Tullock. “ An Empirical Analysis of Cross-National Economic Growth, 1951-80,” Journal of Monetary Economics (September 1989), pp. 259-76. White, Lawrence H. Free Banking in Britain: Theory, Experi ence, and Debate, 1800-1845 (Cambridge University Press, 1984). MARCH/APRIL 1993 56 G eorg Rich Georg Rich is a director and the deputy head of Department I at the Swiss National Bank. Commentary FEE HOSKINS HAS WRITTEN a line paper on monetary policy. I share most of his views on the role and duties of central banks. Hoskins discusses why the conduct of monetary policy has been entrusted to central banks. He also examines the conditions that must be satisfied for c e n tra l b an k s to p la y an e ffe c t iv e p o lic y ro le. Hoskins’ principal thesis is that central banks are needed to manage a standard based on fiat money. But a fiat standard imposes few constraints on central banks. If central banks are permitted to issue fiat money, there is always the risk that they will abuse their powers. Consequently, under a fiat standard it is necessary to ensure that central banks act in the public interest. Why do central banks frequently harm the pub lic interest by debasing the currency? Hoskins discusses several possible reasons. He dismisses the answers offered by public-choice economists and also rejects the notion that unsatisfactory performance of central banks is due to the pursuit of inappropriate targets or operating proce dures. Instead, he maintains that "central bank ers are suffering from a Keynesian hangover.” Frequently they do not direct monetary policy solely at price stability but attempt to pursue FEDERAL RESERVE BANK OF ST. LOUIS multiple objectives that often conflict. Many central bankers attempt to achieve at least two goals—to keep prices stable and to smooth cycli cal fluctuations in output and employment. Too often, Hoskins maintains, central bankers also try to manipulate the exchange rate with a view to strengthening the competitive position of domestic industry. Of course they do not pur sue multiple objectives because of a character defect. They merely reflect prevailing opinions held by politicians, bankers, economists and other members of the general public. In Hoskins' view, the performance of central banks could be much improved if they were granted independence from governments and given a single objective—price stability. The cen tral banks—though independent—would not be allowed to choose policy objectives but would be given a clear legislative mandate to achieve and maintain price stability. Moreover, they would be accountable to the public for their policy actions. I am largely sympathetic to Hoskins' sugges tions. An independent central bank with a clear mandate to pursue price stability is likely to perform better than an institution attempting to 57 respond to diverse and conflicting political pres sures. I also agree with Hoskins that the social value of a credible price-stability objective is often underestimated, whereas the costs of eradicating inflation are overstated. Thus I support Hoskins’ call for committing central banks to a price-stability objective. In my view, however, the story does not end here. A clear price-stability mandate by itself is not enough to improve the performance of central banks. Even if we agree that the objective of monetary policy should be price stability, we still have to address a second question: How should central banks achieve and maintain a stable price level? Hoskins plays down the problems of designing operational policy rules consistent with the price-stability mandate. Yet as practitioners of monetary policy know, the translation o f such a mandate into specific policy rules is far from trivial. Switzerland offers a good case in point. I argue that the Swiss National Bank (SNB) pos sesses a clear mandate to achieve and maintain price stability even though Swiss law does not precisely define the objectives of monetary pol icy. This mandate, albeit informal, rests on a remarkable consensus among the Swiss public about the objectives of monetary policy. The SNB’s informal mandate explains why the inflation rate in Switzerland has tended to be low by international standards. Since the begin ning of 1975—when Switzerland shifted to money stock targeting—inflation in Switzerland has averaged 3.5%. This average, however, still far exceeds the SNB’s stated inflation target of 0 percent to 1 percent. Consequently, the SNB has failed to achieve price stability despite the informal mandate. The SNB’s failure to meet its stated target results largely from two short episodes of accelerating inflation. From 1979 to 1981 and from 1989 to 1991, Swiss inflation temporarily rose to more than 7 percent and 6 percent, respectively. NEED FOR OPERATIONAL RULES The SNB's failure to achieve price stability did not reflect a Keynesian hangover. Rather, the SNB encountered various problems when it attempted to translate its price-stability mandate into suitable operational policy rules. The need for operational rules arises because monetary policy affects the inflation rate witli a long and frequently variable lag. In Switzerland the time lag may be as much as three years. Therefore mone tary policy decisions do not affect the inflation rate until long after they are implemented. Because of the lag, such decisions invariably entail a great deal of uncertainty. Central banks may err even if they try to adhere closely to their mandate. Once they recognize their mistakes, it is usually too late to take corrective action. To lower the danger of policy blunders, cen tral banks require reliable early warning signals or leading indicators of inflation. Operational rules centered on these leading indicators give central banks a good chance of accomplishing a goal of achieving and maintaining price stability. Do central banks possess reliable leading indica tors of inflation? This question cannot be answered straightforwardly. Monetarists tend to empha size the close relationship between money growth and the inflation rate. They maintain that the money stock serves as a good leading indicator o f price movements. Therefore central banks are likely to meet the price-stability objec tive if they adopt an operational rule providing for steady growth in the money supply. Most central banks today share the monetarist view that inflation is due largely to excessive money growth. Nonetheless, they hesitate to opt for strategies of steady money growth. The SNB is no exception. In Switzerland the growth in both the monetary base and the money stock M l tend to lead inflation. Therefore the SNB focuses attention on these two aggregates and sets an intermediate target for the Swiss mone tary base. It strives to increase the monetary base at a rate of 1 percent per year. The SNB views this target as consistent with price stabil ity in the medium and long runs. Although the SNB follows a money-growth rule, it need not augment the monetary base by 1 percent year after year. Depending on the cir cumstances, it may temporarily undershoot or overshoot the 1 percent target. For this reason, the SNB frames its money-growth rule in terms of a medium-range target, to be met on the average of a five-year period. Temporary deviations from the 1 percent growth path may be required if serious unexpected shocks hit the Swiss economy. Two kinds of shocks may prompt the SNB to deviate: unexpected shifts in money demand and other unexpected shocks such as excessive movements in the exchange rate. MARCH/APRIL 1993 58 SHIFTS IN MONEY DEMAND A strategy of steady money growth is effective only if money demand is stable. In contrast to many other countries, Switzerland has been blessed with reasonably stable money-demand pat terns. But this does not imply that instabilities have not occurred. Serious instabilities arose in the late 1980s as a result of two financial inno vations. A new electronic interbank payments sys tem and a major overhaul of liquidity requirements, or minimum reserve requirements, imposed on banks caused a huge permanent drop in the demand for base money. Much of that decline occurred in the first half of 1988, but stability was not restored until about 1990 or 1991. With hindsight, various students of Swiss monetary policy attribute the most recent surge in the Swiss inflation rate to the SNB's cautious reaction to the demand shift. The SNB, they assert, should have acted more aggressively. The SNB's cautious response no doubt was equivalent to a temporary easing of monetary policy. Nonetheless, it cannot be regarded as the main cause of the rise in inflation. I am not aware of any economic theory able to explain how six months of easy money, which the mar ket correctly regarded as transitory, could have generated three years of high inflation. For this reason, I still maintain that central banks should react cautiously to shifts in money demand. OTHER UNEXPECTED SHOCKS It is clear that central banks must adjust the money supply to permanent demand shifts or long-lasting temporary demand shifts if they are to keep the price level stable. It is not always advisable to react quickly to demand shifts, however. Money demand is subject to frequent transitory movements that do not call for a central-bank response. Moreover, demand shifts are hard to detect. They often become fully apparent only after considerable time has elapsed. For these reasons, Meltzer (1987) and McCallum (1989, Ch. 16) recommend a slow reaction pattern. They propose mechanical rules that would prompt central banks to adjust the money supply gradu ally to demand shifts. I support Meltzer and McCallum’s call for a gradual response, but I doubt that central banks should be committed to a mechanical reaction pattern. The speed of the response is likely to depend on the nature of these shifts. For example, if central banks know in advance that a major shift will occur, they should adjust the money supply quickly. Confronted with the demand shift of the late 1980s, the SNB opted for caution. SNB officials knew that a shift would occur but did not know how big the shift would be or how fast base-money demand would fall. As a result of the SNB’s cautious response, short-term domes tic interest rates fell sharply at the beginning of 1988 but rose again as the SNB gradually lowered the supply of base money. By summer 1988, short term domestic interest rates returned to their pre-shift levels. Long-term rates, however, did not budge. Thus market participants correctly regarded the fall in short-term interest rates as transitory. FEDERAL RESERVE BANK OF ST. LOUIS Similar problems arise from other unexpected shocks that may impinge on the central banks’ anti-inflationary monetary policies. In small countries like Switzerland, central banks are frequently compelled to take the real exchange rate into account when setting monetary policy. Real exchange rate movements often fail to reflect economic fundamentals. As I pointed out before, Swiss inflation picked up temporarily in the early 1980s and early 1990s. Although the SNB attempted to keep the monetary base on a growth path consistent with medium-run price stability, the Swiss franc weakened sharply in real terms during both periods of high inflation; that is, the depreciation was much larger than would have been expected on the basis of infla tion differentials between Switzerland and other countries. Therefore the exchange-rate deprecia tion reinforced the inflationary pressures in Switzerland. The SNB reacted to this situation by tightening monetary policy. As a result, the monetary base fell below the medium-run growth path. The tightening of the monetary reins eventually caused the Swiss franc to appreciate again. In this way, the SNB counter acted the inflationary pressures emanating from the exchange rate. Lee Hoskins takes a dim view o f central-bank attempts to manipulate the exchange rate. How ever, he considers only central-bank efforts to stimulate domestic employment by means of an exchange-rate depreciation. Such policies, I agree, may be inconsistent with the mandate to achieve and maintain price stability. But we should not overlook the situations in which exchange-rate movements undermine central banks' antiinflationary policy stances. 59 Nevertheless, Hoskins’ objections to exchangerate policy are often valid. Exchange-rate policy may or may not be consistent with price stability. Swiss experience offers examples of both types of exchange-rate policy. The SNB did more than try to counteract excessive real depreciations of the Swiss franc. In 1978 and 1987 it reacted to an excessive real appreciation by relaxing mone tary policy. Although the real appreciation supported the fight against inflation, the SNB tried to halt or even reverse the upward movement in the exchange rate. The SNB thought that its efforts to curb the appreciation of the Swiss franc were consistent with its mandate to stabilize the price level. In 1978 and 1987 inflation was low and declining. In principle it followed an opera tional strategy of gradually lowering the infla tion rate. In its view a gradual approach would minimize the real costs of achieving and main taining price stability. Considering its preference for gradualism, the SNB did not welcome the real appreciation of the Swiss franc because it affected the domestic economy in the same way an unnecessary tightening of monetary policy would. Therefore the SNB allowed money growth to rise temporarily above the level con sistent with medium-run price stability. Unfortunately, the SNB’s strategy of adjusting money growth to the real appreciation of the Swiss franc turned out to conflict with the price-stability objective. In both periods inflation rose again in due course. The two short episodes of rising inflation are largely explained by the SNB’s efforts to counteract an excessive real appreciation of the Swiss franc. Thus Swiss experience lends at least partial support to Hoskins’ objections to exchange-rate policy. However, strict compliance with a pricestability mandate need not imply that central banks should abstain totally from manipulating the exchange rate. Even if the SNB tried to rule out any risks of erring on the side of inflation, it could not afford to ignore real exchange rate movements entirely. Instead it had to react asymmetrically. With an excessive real apprecia tion of the Swiss franc, the SNB would keep the monetary base on the medium-run growth path. Faced with an excessive real depreciation, on the other hand, it would push the monetary base below that path. The resulting policy might be closer to shock therapy than to gradualism. The real costs of the shock therapy would con stitute the price the SNB would have to pay for playing it safe. In practice, I doubt that central banks are able to disregard entirely the real costs of eliminating inflation. The SNB has repeatedly emphasized that it cannot stabilize the price level without accepting a temporary increase in unemployment. But the Swiss public also expects the SNB to keep the real costs of its anti-inflationary mone tary policy as low as possible. Therefore the SNB, in principle, must follow a gradualist approach. We could probably improve our per formance if in the future we display greater reluctance to react to excessive real apprecia tions of the Swiss franc than we have in the past. CONCLUSIONS Let me conclude by emphasizing again that I agree with the thrust of Hoskins’ reasoning. Monetary policy should be entrusted to indepen dent central banks with a clear legislative man date to achieve and maintain price stability. But in my view, independence and a clear mandate are not sufficient to guarantee a good monetary policy performance. It is also important that central banks adopt operational policy rules consistent with their mandate. Although central banks should be free to choose appropriate operational rules, they should be committed to spell out explicitly how they intend to fulfill their mandates. In particular, they should state how they intend to respond to shifts in money demand and other unexpected disturbances. REFERENCES McCallum, Bennett T. Monetary Economics (Macmillan, 1989). Meltzer, Allan H. “ Limits of Short-Run Stabilization Policy: Presidential Address to the Western Economic Association, July 3, 1986,” Economic Inquiry (January 1987), pp. 1-14. MARCH/APRIL 1993 61 Harold Dem setz Harold Demsetz is the Arthur Andersen chair in business economics at the University of California-Los Angeles. Financial Regulation and the Competitiveness o f the Large U.S. Corporation A CENTRAL QUESTION OF THE DAY is whether U.S. business firms are capable of suc cess in highly competitive world markets. The question is embedded in hotly debated calls for the United States to develop an explicit indus trial policy, in frequently expressed concerns about our loss of market leadership in the computer chip, television and automobile industries, and in charges of excessive executive compensa tion. It is important to consider the efficiency of the large corporation when answering this question, and what I discuss here is a connection between corporate efficiency and the regulation of capital market institutions. The legal setting of a large U.S. corporation is usually thought of in terms of regulations that bear directly on the activities of business firms. These include business tax policy, environmental protection legislation, worker safety and health regulation, and antitrust. Because legal settings for business vary from nation to nation, regula tions undoubtedly affect relative efficiencies of business firms differently in different parts of the world. Business regulation of this type has been discussed explicitly on many occasions, so I set it aside here. Instead, I give attention to the neglected connection between corporate efficiency and the regulation of capital market institutions. My purpose is to show how the regulation of banks, insurance companies and mutual funds impinges on shareholder control of top management in U.S. corporations. Because most persons at this conference do not work in corporate economics, it is useful to begin by considering the potential control problem created by the diffuse ownership struc ture on which the modern large corporation rests—separation of ownership and control. This well-known agency problem has been around for some time. Even Adam Smith voiced con cern in The Wealth o f Nations, precisely because he believed that those who manage the funds of others cannot be expected to do as good a job as if their own funds w ere at stake. Along with many contemporary economists, the works of Veblen (1921), Berle and Means (1932), and Galbraith (1967) build heavily on this corporate control problem. Their works assert that owners of shares each have an ownership stake in the cor poration that is too small to motivate efforts to control management and that is too small to convey disciplinary weight even if such efforts were made. MARCH/APRIL 1993 62 Dissatisfied shareholders can do little better than sell their shares. If such sales are large, the price per share will fall and adversely affect the terms on which management can raise new capital from the capital markets. This price effect penalizes errant management, but it does so only indirectly and only to the extent that the corporation finds it necessary to raise new capital. The alleged weakening of the link between ownership and control makes the proposition correlating private ownership and efficient resource allocation more problematic in the minds of some students of the corporation. Uncontrolled professional management is likely to see its interest served, at least partly, by high management compensation, large firm size, altruism toward friends and community, leisure and other forms of on-the-job consumption and by indulging in these to an extent that seems inimical to shareholder interests. The thesis appears to be much like that which popularly explains the failings of socialism. If all citizens are in principle owners of state property then no person qua citizen can exercise control over this property. Ownership is simply too diffuse to be effective. Managing this property then becomes the task of state bureaucracies. State employees, however, have interests that do not coincide with those of the population at large, and the pursuit of these interests is not guided by market incentives. A separation between ownership and control arises and undermines the credibility of socialism. The separation thesis as applied to the large corporation substitutes professional management for state bureaucracies.1 Studies of corporate takeovers, mainly corporate takeovers undertaken in the United States during the 1980s, provide evidence of some instances of separation between ownership and control. These show that shareholders of target companies benefit considerably from a takeover of their firm. Successful takeovers increase share prices of target firms by an average of about 30 percent.2 Increases in share price may derive in part from several aspects of takeovers. The dominant view 'Socialized property and jointly owned corporate property are, however, far from equivalent. Socialized ownership is coerced into being, whereas corporate ownership is devised voluntarily. Given the facts of economic development and per capita wealth in East and West, we can surmise only that if there is separation between ownership and control in the large private corporation, it is less severe by several orders of magnitude than it is in the socialist state. 2See Jarrell, Brickley and Netter (1988). FEDERAL RESERVE BANK OF ST. LOUIS is that most of target shareholder gains derive from the removal of inept management, whose presence is consistent with the separation thesis.3 It should be noted, however, that only a small fraction of corporate assets has become the target of takeover attempts. This can be interpreted as statistical support for a proposition contrary to the separation thesis—that most modern corpo rations are not afflicted by significant separation between ownership and control. The indictment of the modern corporation implicit in the separation thesis creates its own puzzle. Because the corporation, including its ownership structure, arises from contractual agreements voluntarily entered into, the separa tion thesis implies that serious, systematic and persistent errors are made by owners of the corporation in relying on ownership structures that are too diffuse. Owners fail to anticipate that they are abandoning control over their assets. This is inconsistent with the belief held by most economists that all parties to an agree ment reached voluntarily expect to benefit from the agreement and that if the agreement is used repeatedly and extensively, this expectation is usually correct. However, the empirical supposition of the sepa ration thesis, the "fact” to which all adherents to the thesis have subscribed, is not at all fact. It is simply not the case that the ownership structure of the typical large corporation is so diffuse that it undermines the incentive and power of shareholders to influence manage ment. That thousands of shareholders jointly own the typical large corporation is true, but recent studies show that not every owner of corporate stock owns an insignificant number of shares. A few of the thousands o f share holders usually own a relatively large fraction of the firm's equity.4 In fact, the typical large corporation has a more concentrated ownership structure than serves the separation thesis well. For Fortune-500-sized U.S. corporations, the aggregate fraction of equity owned by the five largest shareholders is about one-fourth, and in Japan and several important European countries 3Evidence to date seems to indicate that target company gains do not derive from wealth transferred from bond holders [Jarrell, Brickley and Netter (1988)] or from most lower level employees. Although management personnel are released in disproportionately large numbers from target companies when a takeover occurs, the mass of laborers are not. 4Demsetz and Lehn (1985) and Shleifer and Vishny (1986). 63 this fraction is much larger. The typical case then is one in which a relatively small number of shareholders have well focused interests and nontrivial blocks of votes. Facing such concen trated share holdings, professional management cannot be as unguided by shareholder interests as the separation thesis supposes, although there surely are some cases in which ownership structure has become too diffuse to serve share holder interests well. When this occurs, owner ship should be restructured.3 Restructuring occurs in two ways. Corporate takeovers provide a dramatic mechanism for concentrating existing diffuse ownership struc tures. Less dramatically but more continuously, ownership is restructured through the normal issuing and purchasing of equity shares. At any given time the diffuseness with which shares of firms are held varies across corporations, buL restructuring should adapt ownership structures to the different situations confronted by different firms. This implies that observed structures should bear a sensible relationship to these situ ations. More specifically, we may posit that vari ations in ownership structure reflect the benefits and costs to shareholders of controlling profes sional management tightly. must commit to own a controlling share of equity, and hence the greater is their exposure to firmspecific risk. The risk-adjusted, utility-maximizing ownership structure for large firms, contrary to what is suggested by the separation thesis, is not the single-owner firm.6 It is a more diffuse ownership structure because the cost of bearing firm-specific risk should be reflected in the optimal ownership structure. Nonetheless, this structure should be one in which enough shares are owned by a few shareholders that they can exercise more than a modicum of control over professional management. The data reveal pre cisely this—greater diffuseness in ownership structure for larger firms accompanied by enough concentration of ownership to imbue large share holding interests with influence over management. This pattern of ownership, which suggests that shareholders choose ownership structures that maximize the value of their firms, has been con firmed for Swedish, Japanese and South African firms.7 Concentrated ownership (and consequent tight control over management) comes at a cost. If this cost is high, the ownership structure that is truly profit maximizing must look much like that of the single-owner firm. This is the case in partic ular for large firms because size of firm corre lates with one of the major costs of concentrated ownership—the bearing of firm-specific risk. Because controlling shareholders would tend to have a large fraction of their wealth invested in a single firm if this firm is large, they would be exposing themselves to firm-specific risk. The larger the firm, the larger is the wealth they Of course, management cannot be disciplined so thoroughly by controlling shareholders and by the threat of corporate takeovers that inept ness is dethroned at once wherever and when ever it exists. In this respect, it is important to remember that ownership is not structured exclu sively f o r the purpose o f dealing with management ineptness. Other things matter to ownership structure and to risk-adjusted profit, such as the avoidance of firm-specific risk. If ownership is structured to maximize risk-adjusted utility, it must not be so tightly structured that all error in judging professional management is eliminated. Moreover, because there is a cost to altering the structure of ownership quickly, profits are also maximized by tolerating a lag between evidence of ineptness and altering ownership structure appropriately. Things will get out of whack on 5There are several ways by which professional management can be guided to serve shareholder interest in the modern corporation— concentrated ownership (achieved through the normal financing of corporations or through corporate takeovers), the consequences of the capital market’s measurement of management performance, legal proceed ings, and compensation systems. Time and space allow me to consider here only concentrated ownership. This is unfortunate especially in regard to executive compensation, for there are new empirical results to report about this. It is improbable that all these mechanisms transform the modern corporation into a precise analog of the firm pictured in neoclassical theory, but they do raise serious questions about the Berle and Means thesis. professional managers without error to pursue his chosen ends. Although risk-adjusted profit always looms important to this owner, it need not be his sole concern. He might derive satisfaction from owning a larger firm even if it is less profitable, or from using the firm ’s assets to cater to personal utility maximization. The reduction in profit he thereby bears must not be thought of as a loss sustained because an agency problem separates his interests from management's behavior. There is no agency problem here, there is simply the recognition that in cases such as this, profit maximization for the owner does not equate to utility maximization for the owner. This may also hold for degrees of ownership concentration less than the 100 percent just assumed. 6I speak somewhat superficially in reference to risk-adjusted utility maximization. Suppose a real corporation is owned by a single person, and suppose further that he guides his 7See Bergstrom and Rydqvist (1990), Prowse (1991), and Gerson (1992). MARCH/APRIL 1993 64 occasion, and when they do, dramatic restruc turing of ownership is more likely to be called forth in the guise of a corporate takeover. What seems to be true then is that professional management is not free of substantive guidance by shareholders, but that the degree of guidance, because it responds to problems of risk and other similar concerns, will not generally be designed solely for the purpose of controlling management malfeasance. From a shareholder’s perspective, the optimal amount of management malfeasance is positive, not zero. Just what is optimal, however, is affected by the legal environ ment, and especially by laws bearing on the operation of capital market institutions. Ownership will tend to be more concentrated, and manage ment malfeasance will consequently be less pervasive to the extent that these laws do not raise the cost of maintaining concentrated owner ship structures. Recent data reveal a puzzle regarding ownership concentration. After standardizing for variables that should influence the ownership structure of corporations, such as firm size and firmspecific risk, studies of corporate ownership reveal large differences across countries in the typical degree to which ownership is concen trated. Ownership is noticeably more diffuse in U.S. corporations than in Japanese, European and South African corporations. In the typical large corporation in the United States, the top five shareholders, as a group, own about onefourth of the firm's outstanding voting stock. Most corporations traded on South Africa’s Johannesburg Stock Exchange are controlled by small shareholder groups who own 50 percent or more of voting stock.8 Ownership structures in Germany and Sweden are more like South Africa than the United States.9 In Japan, the five largest shareholders own about 33 percent of voting shares.1 0 The differences between the United States and these other countries are so large that we must suspect that the cost of concentrating cor porate ownership differs substantially from one country to another and for reasons not captured by the variables being used to index this cost. If a five-shareholder group owning one-fourth of the voting equity of the typical large corporation is 8See Gerson (1992). 9See Sundqvist (1986). ,0Prowse (1991). FEDERAL RESERVE BANK OF ST. LOUIS a suitable ownership structure in the United States, why is it not in other countries? A plau sible source of this difference is in variation across nations of regulations that impinge on ownership structure and which make it more costly to maintain control in the typical large U.S. corporation than in the typical large nonU.S. corporation. Important capital market insti tutions in the United States do bear special costs to hold large stakes in the equities of other com panies, and our banks are barred from holding any stake.1 1 One potential source of equity capital is the investment company, but the Investment Company Act of 1940 restricts the ability of investment companies to take concentrated equity positions in the firms in which they invest if they advertise themselves as diversified investment companies. There is a tax advantage to registering as a diversi fied investment company, since this entitles the company to pass income through to its investors without paying taxes, but even investment com panies that do not register as diversified are barred from exercising control over any firm engaged in interstate commerce. Hence funds channeled through investment companies are unlikely sources of controlling positions in the equity of corporations. Insurance companies are another potentially important source of equity capital, and most states do allow insurance companies to invest a percent age of their assets in common stock. The per centage varies from state to state but is commonly about 20 percent. New York, a particularly important state for insurance, limits the amount that insurance companies can invest in one company to 2 percent of the insurance company’s assets. Most other states have similar restrictions, but the percentage varies over a large range. States generally bar insurance companies from owning more than a stipulated percentage of the shares of other companies. A common upper bound is 10 percent. Finally, there fre quently is a penalty borne by insurance compa nies that invest in common stock; most states require that capital be set aside to maintain a financial cushion against declines in the price of stock held for investment purposes. Although it is not impossible to use funds channeled through 1 My summary discussion of the details of some of the relevant 1 legislation rests heavily on work by Paul S. Clyde (1990). 65 insurance companies to take a concentrated equity position in a given corporation, it clearly is an investment tactic that is generally dis couraged by state-imposed restrictions. Hence, a second capital market institution is handicapped in such an undertaking. For m ore than 60 years the Glass-Steagall Act has barred banks from directly ow ning equity in U.S. corporations. There is no counterpart to this law in South Africa and in much o f W estern Europe, and only recently has Japan adopted a similar law. Although banks w ould seem to be low-cost conduits o f equity capital, Glass-Steagall forces corporations to raise equity funds from other sources. In fact, banks play important equity roles in other nations, w h ere they supply enough equity to ow n sizeable positions in cor porate ownership structures. The possible con nection betw een Glass-Steagall and ownership structure, how ever, is not generally suspected even though banks are potentially a major source o f equity investment capital in the United States.1 If the behavior o f foreign banks 2 in their ow n countries is a guide to what U.S. banks would do if allowed to invest in corporate equity, it seems likely that U.S. banks w ould be important sources not only o f equity capital, but also o f concentrated ownership positions. A third major source o f concentrated ownership is thus barred by legislation. Because o f recent court decisions, em ployment retirem ent funds remain one important source o f capital that is free to take equity positions, even concentrated equity positions. In fact, w e find a fe w o f these funds playing key roles in monitoring and disciplining corporate manage ment by virtue o f their large holdings o f stock in particular corporations. Most notable in this regard is the California State Employees Retire ment Fund, but others have also becom e activist. For reasons discussed later, how ever, I do not believe that these funds o ffe r m onitoring and disciplining services as good as those likely to come from capital market institutions presently barred or penalized from taking large equity positions in specific corporations. The consequence o f these legal barriers is that corporations housed in the United States rely on capital that is secured directly from individual investors to a much greater extent than corporations located overseas. Really large controlling positions in the equity o f U.S. corpo rations are taken mainly by individual and family investors in the United States. Because o f greater portfolio specialization, these individuals and families are exposed to m ore firm-specific risk than capital market institutions w ould be. M ore over, individual or family wealth is seldom large enough to allow concentrated holdings o f the equity o f large corporations. The heavy reliance in the United States on this source o f equity capital results in corporate ownership structures much m ore diffuse than those that exist fo r comparable foreign corporations. The optimal degree o f control exercised by shareholders over the managements o f their U.S. corporations, as a result o f such legislation, is less than elsewhere. Arguments pro and con can be made in regard to the various legal hurdles that keep important institutional conduits o f capital from accessing the equity markets easily. W h atever the truth in this regard, the effect o f these legal hurdles on ownership structure and control has not yet been taken into account. The control problem created by these hurdles, taken by itself, offers a novel basis fo r opposing such legislation. But can institutional investors—fo r example, investment companies, insurance companies and banks—be relied on to perform the ownership function well? Since their capital comes from diffuse sources, it would seem that their ow n operations should be subject to the separation problem believed to plague large corporations. If so, institutional investors holding controlling positions in the equity structure o f large corpo rations cannot be expected to perform the duties o f ow ner as w ell as investors whose ow n wealth is at stake. I discuss this issue in the rem ainder o f this paper, showing that the control problem can be ameliorated by such institutions but not as completely as if individuals ow ned concentrated ownership positions in corporations directly. There are institutional investors that seem capable o f circum venting the problem created by their ow n diffuse ownership structure, and there are others that seem not so capable o f doing this. The distinction betw een the tw o lies in the ease with which individual investors can reclaim their funds from the institution. The openended mutual stock fund is organized so that investors can insist that the fund buy back their shares at the net asset value they represent in 12Exceptions include Prowse (1990) and Gerson (1992). MARCH/APRIL 1993 66 the fund’s portfolio. The closed-end stock fund has no such obligation; an investor w ho wants to convert his shares in such a fund to cash may sell them at w hatever market-determined price they fetch, but he cannot demand their redem ption by the fund. This is an important difference w hen it comes to the issue o f separa tion betw een ownership and control. T o see its importance, let us reconsider the separation problem in the context o f the corporation. T w o conditions must exist fo r the separation problem to be severe in a corporation. One is the generally recognized condition that ow n er ship structure be diffuse. The other is the con dition that assets made available to a corporation by shareholders must belong to it and not to shareholders. This second condition has not been recognized explicitly in economic litera ture, but it is important. It refers to the fact that, although the shareholder may sell his shares if he is dissatisfied, the shareholder can not insist that the corporation be the buyer o f his shares. Thus the corporation, not the share holder, has title to the productive assets it has purchased with funds secured from its initial issue o f stock.1 If shareholders could reclaim 3 these assets, the severity o f the separation problem w ould be lessened even fo r diffuse ownership structures. It would be lessened even m ore if share ownership w ere concentrated, because shareholders with much at stake will be m ore attentive to what management has been doing with the firm ’s assets. It is not practical to allow shareholders to reclaim their share o f the firm ’s assets in the general case o f the business firm. The typical corporation makes commitments to supply goods and services that, if they are dependably honored, require the corporation to have continuing con trol o f its assets. A steel company cannot relia bly stand by a commitment to fill an order fo r steel if its shareholders can force it to sell its assets to purchase back their stock. The typical corporation th erefore must be organized in a w ay that bars investors from reclaiming their fraction o f the firm ’s assets, and once the typi cal corporation sells a new issue o f shares, the funds it acquires belong to it, not to those w ho purchased the shares. 13Subsequent sale of shares by shareholders has a depressing effect on the price of the corporation’s stock if enough share holders offer to sell, and this has some disciplining effect FEDERAL RESERVE BANK OF ST. LOUIS Continued control by the firm over its assets is not a prerequisite to doing business if credi ble commitments o f this sort are not necessary. Consider the open-ended mutual stock fund. This firm gathers capital from investors and uses its skill to place these funds in the shares o f other corporations. These shares can be sold by the mutual stock fund on a m om ent’s notice should it decide to do so, and in doing so it will not be jeopardizing any business commitments. Consequently, investors w h o place their capital at the disposal o f open-ended mutual funds can w ithdraw their p ro rata share o f the value o f the fund's assets should they becom e displeased with the fund's perform ance. De fa cto , the openended mutual fund is obligated to repurchase pro rata investment positions. These investors are not shareholders in the conventional sense. They are purchasers o f investment services, but they also are providers o f the capital that is in turn invested in shareholdings o f other compa nies. In the absence o f the Glass-Steagall Act, the same arrangem ent could w ork fo r banks w ho reinvest depositor funds in the shares o f other corporations (but probably could not w ork w ell fo r that part o f bank investments that constitutes time-commitment loans to busi ness firms). Should those depositors w ho have made no commitment to keep their funds with a bank decide to w ithdraw deposits, the bank could sell its shareholdings in other corpora tions to cover the withdrawals. It is this characteristic, the ability o f investordepositors to reclaim capital from a firm , that distinguishes these institutions from others fo r our purposes. The closed-end fund does not have this characteristic. It is organized like the typical corporation. It issues shares and converts the funds from their sale to assets that belong to it. Dissatisfied shareholders may sell their shares, but they cannot force the closed-end fund to be the purchaser. This allows the fund’s management to make its investment plans without fear o f being forced to alter them should investor desires fo r cash or beliefs about the investment environm ent change, but it also eliminates the threat to management that it w ill lose control o f fund assets if the fund perform s poorly. It is this threat in the case o f the open-end mutual investment fund that reduces the on management, but even so, the corporation remains in control of the assets it has acquired. 67 severity o f the separation problem. Should investors becom e dissatisfied in large numbers, mass withdrawals would diminish the assets available to a fund’s management, forcing it to sell the shares they ow n in other companies. This reduction in the wealth available to the managements o f these institutions can take place even if no single investor or small group o f investors has provided a lion’s share o f the capi tal being invested. This disciplines the manage ments o f these institutions in a w ay not available to stockholders w hen they are disappointed with the managements o f typical corporations. The large scale sale o f shares in the typical corporation depresses share price but does not reclaim assets from management control. W hat this means is that managements o f institutional investors o f the open-ended mutual fund variety can be disciplined directly by providers o f capital even w hen there is no con centrated provision o f this capital. The diffuse ow n er problem is ameliorated, but only to a limited extent. It is m ore effectively defused if capital is provided in concentrated fashion to the institutional investor, fo r concentration o f rewards and penalties makes the large share holder more attentive and astute. Now suppose that this type o f institutional investor has taken controlling positions in the equity o f the firms whose shares it has purchased. The ability o f even diffuse contributors o f its capital to w ith draw their assets surely makes the institutional investor represent its investors’ interests better than if the threat o f w ithdraw al did not exist— as long as the ability o f the institution to make long-term commitments is not important to its productivity. Because o f this effect, capital secured from even diffuse sources can be combined w ith out suffering fully from a separation betw een ownership and control.1 4 One final point may be raised about another source o f diffuse ownership in the United States. The N ew York Stock Exchange (NYSE) requires that firms it lists raise their equity capital on a one-share, one-vote basis. The NYSE did not always use this standard. It was adopted during the 1920s under considerable pressure from governm ent and intellectuals w ho feared that the grow in g use o f differin g vote entitlements was disenfranchising many equity capital pro 14ln fact, a doctoral dissertation recently completed at UCLA [Clyde (1990)] gives evidence that institutional investors behave much as do individual and family shareholders viders. Nonvoting equity shares are used much m ore extensively in other countries. This makes fo r a low er cost o f establishing controlling equity positions in a company because only vo t ing shares must be reckoned with w hen con sidering the direct control o f management. Discussion o f this issue, how ever, cannot be undertaken here. REFERENCES Bergstrom, Clas, and Kristian Rydqvist. “ The Determinants of Corporate Ownership — An Empirical Study on Swedish Data,” Journal of Banking and Finance (August 1990), pp. 237-53. Berle, Adolf A., and Gardiner C. Means. The Modern Corpo ration and Private Property (Macmillan, 1932). Clyde, Paul S. The Institutional Investor as an Effective Moni tor of Management, UCLA Economics Dept., Ph.D. disser tation, 1990. Demsetz, Harold. “ The Structure of Ownership and the Theory of the Firm,” Journal of Law and Economics (1983). _______ . “ Corporate Control, Insider Trading, and Rates of Return,” American Economic Association: Papers and Proceedings (May 1986), pp. 313-16. _______ , and Kenneth Lehn. “ The Structure of Corporate Ownership: Causes and Consequences,” Journal of Political Economy (December 1985), pp. 1155-77. Galbraith, John K. The New Industrial State (Houghton Mifflin, 1967). Gerson, Josell. “ The Determinants of Corporate Ownership and Control in South Africa,” UCLA Economics Dept., Ph.D. dissertation, June 1992. Jarrell, Gregg, James A. Brickley, and Jeffrey M. Netter. “ The Market for Corporate Control: The Empirical Evidence Since 1980,” Journal of Economic Perspectives (Winter 1988), pp. 49-68. Prowse, Stephen D. “ Institutional Investment Patterns and Corporate Financial Behavior in the United States and Japan,” Journal of Financial Economics (September 1990), pp. 43-66. ________“ The Structure of Corporate Ownership in Japan,” (1991), unpublished manuscript. Smith, Adam. The Wealth of Nations (Modern Library, 1937), pp. 713-14. Shleifer, Andrei, and Robert W. Vishny. “ Large Shareholders and Corporate Control,” Journal of Political Economy (1986), pp. 461-88. Sundqvist, S. Owners and Power in Sweden’s Listed Companies (Dagens Nyheters Forlag, 1986). Veblen, Thorstein. The Engineers and the Price System (1921). who own controlling positions in the ownership structure of a corporation. MARCH/APRIL 1993 68 Charles I. Plosser Charles I. Plosser is the John M. Olin Distinguished Professor of Economics and Public Policy, W E. Simon Graduate School . of Business Administration, University of Rochester. Financial support of the John M. Olin Foundation is gratefully acknowledged. Commentary H Mil >11 ( DEMSETZ RARELY FAILS to deliver a creative and thought-provoking paper, and this one is no exception. I have learned a great deal from Harold’s writings over the years and usually find myself persuaded by his arguments. In this paper Demsetz explores the implications o f cer tain restrictions on the behavior o f financial insti tutions fo r the efficiency o f the market for corporate control. This is a potentially important consideration and one that, to my knowledge, has not been systematically investigated. The basic thesis o f the paper can be summa rized in three steps. The first step is to recog nize that the diffusion o f ownership o f large corporations creates a control problem fo r ow n ers (that is, stockholders). This well-known agency problem has long been the focus o f in tense study by economists. It is important to recognize, how ever, that the degree o f diffusion o f ownership reflects both costs and benefits to shareholders. Th e benefits arise from the reduc tion o f firm-specific risk borne by owners through the diversification o f their holdings. The costs arise from the potential loss o f control over management. The second step in the analysis is to argue that financial regulations limiting the extent o f ownership in a corporation by certain types o f institutional investors, including insurance com panies, investment companies and commercial banks, potentially raise the costs o f controlling management. The third step in the argument is to suggest that this reduced ability o f ow ners to monitor and control managers reduces the e ffi ciency o f large corporations and thus tends to FEDERAL RESERVE BANK OF ST. LOUIS make them less competitive than corporations in countries w h ere institutional investors are not subject to such restrictions. I have no difficulty w ith the logic or thrust o f this line o f reasoning. It is rare that regulations are neutral and thus fail to distort resource allo cations. The questions I am interested in focus ing on have to do w ith w hether the market has created alternative means o f monitoring manage ment. I f so, then the question is, which means is most cost efficient? Demsetz recognizes that concentrated ow n er ship is not the only means o f exercising control over management. Boards o f directors p ro vid e important control mechanisms and have begun to reassert their authority. The recent cases o f General Motors, Am erican Express and IBM are good examples. Management compensation is another means to align management and share holder interests. In general, the market fo r cor porate control is an important monitoring device. Though it requires individuals or firms to obtain a concentrated ownership in a company, it does not necessarily depend on large financial institu tions acting as the investors doing the m onitor ing. Large pension funds that are not regulated like banks or investment companies have be come increasingly active in m onitoring manage ment. CALPERS is one o f the most well-known funds and has been deeply involved in pressur ing fo r management changes in several com panies. Demsetz stresses that there must be a cost to regulation that prohibits investment companies, 69 insurance companies and banks from taking po sitions that encourage them to monitor manage ment m ore closely. This is undoubtedly true. But it is an empirical question as to the im por tance o f these restrictions. I w ould like to sug gest that there may be reasons to believe the effects are small—if fo r no other reason than the marketplace is innovative in getting around such restrictions, particularly w hen there are large rewards fo r doing so. Institutional investors can provide tw o sorts o f services—risk sharing or diversification and management monitoring. Th ere is no particular reason w h y expertise in one activity implies ex pertise in the other. In fact, it is easy to imagine that some institutional investors would special ize in one activity or the other. For example, Dean LeBaron o f Batterymarch Funds and Rex Sinquefield o f DFA view themselves prim arily as portfolio managers. Neither o f them seems to have the slightest interest in m onitoring manage ment. W hy? Even though they must file 13d's indicating w hen they ow n a significant share o f a particular company, probably only a small percentage o f their portfolios is made up o f companies o f which they ow n a significant share. These fund managers specialize in risk sharing, not control or monitoring. W h y should they have a comparative advantage in monitoring management? Just because they are skilled at managing risk does not mean they are skilled at management control. LeBaron has even pushed the idea o f selling voting rights that w ould allow the separation and specialization in monitoring and risk sharing. W h y should one expect that managers o f regulated insurance companies or investment companies have a comparative advantage in m onitoring management? I f they do not, the regulation is likely to have little substantive effect.1 Researchers sometimes feel that banks are different in this regard. Some view banks as having access to an informational advantage over other parties and thus being in a particu larly good position to exercise control over management. Indeed, in Japan and to a lesser extent in Germany, this has been standard prac tice. I f banks held both debt and equity then they w ould clearly have a strong interest in managerial monitoring. It is not clear, how ever, that they w ould always represent shareholder interests. It is w orth recalling that the GlassSteagall A ct was not m otivated b y a desire to limit managerial control by banks but from a desire to stabilize the payments system involv ing the other side o f the bank’s balance sheet. In fact, certain types o f banks are not subject to these limitations because they are not deposito ry institutions, and they sometimes do take con centrated ownership positions. Th e marketplace has clearly responded to the demand fo r corporate control through a variety o f mechanisms and institutional arrangements that go far beyond the regulated financial inter mediaries. In the case o f monitoring manage ment, new funds and partnerships have been created that specialize in seeking concentrated ownership fo r the purpose o f control. One o f the earliest o f these was WESRAY, which was a partnership betw een W illiam E. Simon and Ray mond Chambers. Th ey engineered successful leveraged buyouts (LBOs) fo r Gibson Greeting Cards, Avis and W ilson Sporting Goods. Kohlberg, Kravis and Roberts (KKR) is another suc cessful partnership that specializes in obtaining concentrated ownership fo r the purpose o f con trolling management. In fact, by 1990, almost every major investment bank had created its ow n LBO fund (fo r example, J. P. Morgan and The First Boston Corp.). These funds and their managers specialize in ownership and control, not in providing risk sharing fo r investors. Thus it w ould appear that financial institutions and the market have responded to the demand for corporate control in innovative ways that cir cumvent some o f the distortions caused by financial regulations on banks and other institu tions regarding ownership. There are other areas w h ere regulation o f financial institutions may be affecting the m ar ket fo r corporate control. Many o f the LBO funds frequently obtain bridge financing from commercial banks. Unfortunately, under the Financial Institutions Reform, Recovery and En forcem ent Act, banks are now much m ore limit ed in their ability to deliver such financing because o f direct restrictions on purchases o f high-risk securities and generally higher capital requirements. Thus, there remain potentially 1Of course, given the history of regulation, skilled monitors are likely to have migrated out of managing funds in these regulated firms. MARCH/APRIL 1993 70 important avenues fo r financial regulations on ownership to affect the efficiency o f the co r porate control market. The final element I w ould like to comment on briefly is what, if anything, all this says about international competition. It is certainly true that U.S. corporations operate in a global m ar ket and that to the extent w e reduce the e ffi ciency or raise the costs o f corporate control mechanisms serving to make U.S. companies better managed, w e put ourselves at a com peti tive disadvantage. The tw o observations that Demsetz makes are that many other countries do not have the same restrictions on ownership by financial institutions and that in some foreign countries, structures are less diffuse than in the FEDERAL RESERVE BANK OF ST. LOUIS United States. Though both observations are potentially relevant, it w ould be m ore interest ing if someone could gather evidence that linked the cross-country patterns o f ownership and regulation to patterns o f corporate p e rfo r mance and corporate control. In summary, I think this is an interesting paper that helps focus attention on a set o f is sues that deserves m ore study. Regulations often have subtle and unintended effects and in some cases those may turn out to be o f first-order im portance. Th e issues discussed in this paper may fall into this category. Nevertheless, w e must never underestimate the creative genius o f m arket participants in circum venting regulations w hen large profit opportunities exist. 71 Carl F. Christ Carl F. Christ is a professor of economics at The Johns Hop kins University. Helpful comments on an earlier draft were made by Jonathan Ahlbrecht, Stephen Blough, Pedro de Lima and William Zame. Any remaining shortcomings are my responsibility. Assessing Applied Econometric Results M . T IS A GREAT HONOR to be asked to participate in this conference to celebrate the w ork o f Ted Balbach, w ho has long upheld the standard o f relevant, independent, intelligible econom ic studies at the Federal Reserve Bank o f St. Louis. My invitation to this conference asked fo r a philosophical paper about good econom etric practice. I have organized my view s as follows. Part I o f the paper defines the concept o f an ideal econometric model and argues that to tell w hether a model is ideal, w e must test it against new data—data that w ere not available when the model was formulated. Such testing suggests that econom etric models are not ideal, but are approxi mations to a changing reality. Part I closes with a list o f desirable properties that w e can realisti cally seek in econom etric models. Part II is a loosely connected set o f comments and criticisms about several econom etric techniques. Part III discusses methods o f evaluating econometric models by means o f their forecasts and summa rizes some results o f such evaluations, as proposed in part I. Part IV resurrects an old, plain-vanilla equation relating m onetary velocity to an interest rate and tests it with m ore recent data. The rather remarkable result is that it still does about as w ell today as it did nearly 40 years ago. Part V is a b rief conclusion. H O W TO RECOGNIZE AN IDEAL M ODEL IF YO U MEET ONE The Goal o f Research and the Concept o f an Ideal M odel The goal o f econom ic research is to im prove knowledge and understanding o f the economy, either fo r their ow n sake, or fo r practical use. W e want to know how to control what is con trollable, how to adapt to what is uncontrollable, and how to tell which is which. Th e goal o f economic research is analogous to the prayer o f Alcoholics Anonymous (I do not suggest that economics is exactly like alcoholism)—"God grant me the serenity to accept the things I cannot change; the courage to change the things I can; and the wisdom to know the difference.” The goal o f applied econometrics is quantitative knowledge expressed in the form o f mathemati cal equations. I invite you to think o f an ideal econometric model, by which I mean a set o f equations, com plete or incomplete, with numerically estimated parameters, that describes some interesting set o f past data, closely but not perfectly, and that MARCH/APRIL 1993 72 Figure 1 Three Methods of Formulating and Estimating a Model and Checking Its Correspondence with 1950-1991 Data Method 1 Date of model’s formulation Estimation period 1991 1992 1950 Method 2 Date of model’s formulation Estimation period Prediction period 1971 1972 1950 Method 3 1991 1992 Date of model’s formulation Estimation period 1950 Prediction Period 1971 1972 ______________________ __________ . . I 1991 1992 Period of data available when model was formulated Period of data not yet available when model was formulated ■H w ill continue to describe all future data o f that type. The N eed f o r Testing Against N ew Data H ow can w e tell w hether w e have found an ideal econom etric model? W e can certainly tell how w ell a model describes a given set o f past data. (W e w ill discuss what is meant by a good description later). Suppose w e have a model in 1992, w ith estimated parameters, that closely describes past data fo r 1950-91. T o tell w hether it is the ideal model w e seek, w e must try it with future data. Suppose that after three years w e try the model with data fo r 1992-94, and it describes them closely also. Still, in 1995 all w e will be sure o f is that it describes data closely for a past period, this time from 1950 through 1994. In principle w e can never be sure w e have found an ideal model because there will always be m ore future data to come, so w e w ill never be able to say that a model is ideal. The longer the string o f future data that a model describes closely, how ever, the m ore confidence w e have in it. Is this only a matter o f the amount o f data that the model describes, or is there something else involved? I argue that something else is involved. FEDERAL RESERVE BANK OF ST. LOUIS Suppose again that in 1992 w e have a model that closely describes an interesting data set fo r the past period 1950-91. Consider the follow ing three methods, shown in figure 1, by which this model might have been obtained and by which its ability to describe data fo r 1950 through 1991 might have been assessed: 1. It was form ulated in 1992, and fitted to data fo r the entire period 1950-91. 2. It was form ulated in 1992, fitted to data fo r the sub-period 1950-71, and used to predict data from 1972 through 1991. 3. It was form ulated in 1972, fitted to data fo r the sub-period 1950-71, and used to predict data from 1972 through 1991. Methods 1 and 2 d iffer in that method 1 fits the model to all the available data, whereas method 2 fits it to the first part only and uses the result to predict the second part, from 1972 onward. 1972 is not a randomly chosen date. It was the year before the first oil crisis. Method 3 differs in that the model builder did not y et know about the oil crisis w hen form ulating the model. N ow consider the follow ing question: Given the goodness o f fit o f this model to data fo r the whole period 1950-91, does you r confidence in the model depend on which o f these three methods was 73 used to obtain it? I argue that it should. In par ticular, I argue that an equation obtained by a method similar to method 3, which involves testing against data that w ere not available to the model builder when the model was formulated, deserves m ore confidence than the same equation obtained by either o f the other tw o methods. The argument has to do with the goal o f an econom etric m odel—to describe not only past data, but also future data. It is easy to form u late a model that can describe a given set o f past data perfectly but cannot describe future observations at all. O f course, such a research strategy should be avoided. Here is a simple example. Imagine a pair o f vari ables whose relationship w e want to describe. Suppose w e have tw o observations on the pair o f variables. Then a line, whose equation is linear, will fit the data perfectly. N ow suppose w e obtain a third observation. It w ill almost certainly not lie on the line determined by the first two observations. But a parabola, whose equation is quadratic (of degree 2), w ill fit the three observations perfectly. N ow suppose a fourth observation becomes available. It w ill almost certainly not lie on the parabola. But a sort o f S-curve, whose equation is cubic (of degree 3), w ill fit the fou r observa tions perfectly. And so on. In general, a poly nomial equation o f degree n w ill fit a set o f n + 1 observations on tw o variables perfectly, but a polynomial o f higher degree will be required if the number o f observations is increased. Methods o f this type can describe any set o f past data perfectly but almost cer tainly cannot describe any future data. If a model is to describe future data, it needs to capture the enduring systematic features of the phenomena that are being modeled and it should avoid conform ing to accidental features that w ill not endure. The trouble w ith the exactfitting polynomial approach just discussed is that it does not try to distinguish betw een the enduring systematic and the tem porary accidental features o f reality. In the process o f fitting past data per fectly, this approach neglects to fit enduring systematic features even approximately. This relates to the choice among methods 1, 2 and 3 fo r finding a model that describes a body o f data. W hen form ulating a model, researchers typically pay attention to the behavior o f avail able data, which perforce are past data. One tries 1See Mitchell (1927). different equation forms and different variables to see which formulation best describes the data. This pro cess has been called data mining. As a method o f formulating tentative hypotheses, data mining is fine. But it involves the risk o f being too clever, o f fitting the available data too well and hence o f choosing a hypothesis that conform s too much to the tem porary accidental and too little to the enduring systematic features o f the observed data. In this respect it is similar to the exactfitting polynomial approach described earlier, though not as bad. The best protection against having done too good a job o f making a model describe past data is to test the model against new data that w ere not available w hen the model was formulated. This is what method 3 does, and that is w hy a model obtained by method 3 merits m ore confi dence, other things equal. T rygve Haavelmo once said to me, not entirely in jest, that what w e economists should do is form ulate our models, then go fishing fo r 50 years and let new data accumulate, and finally come back and confront our models with the new data. W esley Mitchell put the matter very well when he w rote the follow in g:1 1'he proposition may be ventured that a competent statistician, with sufficient clerical assistance and time at his command, can take almost any pair of time series for a given period and work them into forms which will yield coefficients of cor relation exceeding + .9. It has long been known that a mathematician can fit a curve to any time series which will pass through every point of the data. Performances of the latter sort have no significance, however, unless the mathe matically computed curve continues to agree with the data when projected beyond the period for which it is fitted. So work of the sort which Mr. Karsten and Professor Fisher have shown how to do must be judged, not by the coefficients of correlation obtained within the periods for which they have manipulated the data, but by the co efficients which they get in earlier or later periods to which their formulas may be applied. Milton Friedman, in his review of Jan Tinbergen’s pioneering model o f the U.S. economy, referred to Mitchell’s comment and expressed a similar idea somewhat differently:2 Tinbergen’s results cannot be judged by ordinary tests of statistical significance. The reason is that 2See Friedman (1940) and Tinbergen (1939). MARCH/APRIL 1993 74 the variables with which he winds up, the parti cular series measuring these variables, the leads and lags, and various other aspects of the equations besides the particular values of the parameters (which alone can be tested by the usual stati stical technique) have been selected after an exten sive process of trial and error because they yield high coefficients of correlation. Tinbergen is seldom satisfied with a correlation coefficient less than .98. But these attractive correlation coeffi cients create no presumption that the relationships they describe will hold in the future. The multi ple regression equations which yield them are simply tautological reformulations of selected economic data. Taken at face value, Tinbergen’s work “explains” the errors in his data no less than their real movements. That last statement can be strengthened. Tin ber gen's method, which has been the method o f most model builders ever since, explains w hat ever tem porary accidental components there may be in the data (regardless o f w hether they are measurement errors), as w ell as the enduring components. Most m acroeconom etric models formulated before the 1973 oil crisis had no variables rep re senting the prices and quantities o f oil and energy. Most o f these models w ere surprised by the oil crisis and its aftermath; and most of them made sub stantial forecast errors thereafter. Many models form ulated after 1973 pay special attention to oil and energy. O f course many o f those models provide better explanations o f the post-oil-crisis data than do models that ignore oil and energy. But my point is different. A model that was fo r mulated after the oil crisis w a s s p e c ific a lly designed to conform to data during and after the crisis, and if there are tem porary accidental va r iations, the model w ill conform to them just as much as to the systematic variations. Hence the task o f explaining data betw een the onset o f the 1972 oil crisis and 1992 is easier fo r a model that was form ulated in 1992 than fo r a model that was form ulated before the crisis. Th erefore if both models do equally well at describing data from 1950 to 1991, the one formulated before the crisis has passed a stricter test and merits m ore con fidence. W hat about the relative merits o f methods 1 and 2? Sometimes method 2 is recom m ended; that is, it is recom mended that researchers esti mate a model using only the earlier part o f the available data and use the later part as a test o f the m odel’s forecasting ability. W hen thinking about this proposal, consider a model that has FEDERAL RESERVE BANK OF ST. LOUIS been form ulated w ith access to all o f the data. It does not make much difference w hether part o f the data is excluded from the estimation pro cess and used as a test o f that model, as in method 2, or w hether it is included, as in method 1. Either way, w e draw the same con clusions. If the model with a set o f constant coefficients describes both parts o f the data well, method 1 w ill yield a good fit fo r the w hole period and method 2 w ill yield a good fit fo r the estimation period and small errors fo r the forecast period. If the model with a set o f constant coefficients does not describe both parts o f the data well, in m ethod 1 the residuals, if examined carefully, w ill reveal the flaws, and in method 2 the residuals, the forecast errors or both w ill reveal the flaws. And w ith both methods 1 and 2 w e have a risk that the model was form ulated to conform too much to the temporary accidental features o f the available data. One notew orthy difference betw een methods 1 and 2 is that if the m odel’s specification is correct, method 1 w ill yield m ore accurate estimates o f the parameters because it uses a larger sample and thus has a smaller sampling error. Econom etric M odels Are Approximations W hen I began w ork in econometrics, I believed a premise that underlies much econometric w ork— namely, that a true model that governs the behavior o f the econom y actually exists, with both systematic and random components and with true parameter values. And I believed that ultimately it would be possible to discover that true model and estimate its param eter values. My hope was first to find several models that could tentatively be accepted as ideal and even tually to find m ore general models that would include particular ideal models as special cases. (One w ay to top you r colleagues is to show that their models are special cases o f yours. N ow a days this is called "encompassing.”) Experience suggests that w e cannot expect to find ideal models o f the sort just described. W hen an estimated econom etric model that describes past data is extrapolated into the future fo r more than a year or two, it typically does not hold up well. T o try to understand how this might happen, let us tem porarily adopt the premise that there is a true model. O f course, w e do not know the form or parameters o f this true model. They may or may not be changing, but if they are changing according to some rule, then in principle it is 75 possible to incorporate that rule into a more general unchanging true model. Suppose that an economist has specified a model, which may or may not be the same as the true model. If the form and parameters o f the economist’s model are changing according to some rule (not necessarily the same as the rule governing the true model), again in principle it is possible to incorporate that rule into a m ore general unchanging model. N ow consider the follow ing possible ways in which the economist's model might describe past data quite w ell but fail to describe future data: 1. The form and parameter values o f the economist’s model may be correct fo r both the past period and the future period, but as the forecast horizon is lengthened, the fo re casts get worse because the variance o f the fo re cast is an increasing function o f the length o f the horizon. This w ill be discussed later. 2. Th e form o f the economist’s model may be cor rect fo r both the past period and the future period, but some or all o f the true param e ters may change during the future period. 3. Th e form o f the economist’s model may be correct fo r the past period but not fo r the future period because o f a change in the form o f the true model that is not matched in the economist’s model. 4. The form o f the economist’s model may be incorrect fo r both periods but more nearly correct fo r the past period. Th e last possibility is the most likely o f the four in view o f the fact that the econom y has millions o f different goods and services produced and consumed by millions o f individuals, each with distinct character traits, desires, knowledge and beliefs. These considerations lead to the conjecture that the aforem entioned premise underlying econo metrics is w ron g—that there is no unchanging true model with true parameter values that governs the behavior o f the econom y now and in the future. Instead, every estimated econom etric model is at best an approximation o f a changing econom y—an approximation that becomes w orse as it is applied to events that occur further into the future from the period in which the model was formulated. In this case w e should not be surprised at our failure to find an ideal general model as defined earlier. Instead, w e should be content w ith models that have at best only a tem porary and approximate validity that deteri orates with time. W e should sometimes also be con tent with models that describe only a restricted range o f events—fo r example, events in a parti cular country, industry or population group. Desiderata f o r an Econom etric M odel If no ideal model exists, what characteristics can w e realistically strive fo r in econometric models regarded as scientific hypotheses? The follow ing set o f desiderata are within reach: 1. The estimated model should provide a good description o f some interesting set o f past data. This means it should have small residuals rela tive to the variation o f its variables—that is, high correlation coefficients. The standard errors o f its param eter estimates should be small relative to those estimates, that is, its t-ratios should be large. If it is estimated fo r sep arate subsets o f the available data, all those esti mates should agree w ith each other. Finally, its residuals should appear random. (If the residuals appear to behave systematically, it is desirable to try to find variables to explain them.) 2. The model should be testable against data that w ere not used to estimate it and against data that w ere not available w hen it was specified. 3. The estimated model should be able to describe events occurring after it was form ulated and estimated, at least fo r a fe w quarters or years. 4. The model should make sense in the light o f our knowledge o f the economy. This means in part that it should not generate negative values fo r variables that must be non-negative (such as interest rates) and that it should be consistent with theoretical propositions about the econom y that w e think are correct. 5. Other things equal, a simple model is p re fer able to a complex one. 6. Other things equal, a model that explains a w ide variety o f data is preferable to one that explains only a narrow range o f data. 7. Other things equal, a model that incorporates other useful models as special cases is p re fer able to one that does not. (This is almost the same point as the previous one.) MARCH/APRIL 1993 76 In offerin g these desiderata, I assume that the purpose o f a model is to state a hypothesis that describes an interesting set o f available data and that may possibly describe new data as well. Of course, if the purpose is to test a theory that w e are not sure about, the model should be constructed in such a w ay that estimates o f its parameters w ill tell us something about the validity o f that theory. The failure o f such a model to satisfy these desiderata may tell us that the theory it embodies is false. This too is useful knowledge. COMMENTS A N D CRITICISMS A B O U T ECONOMETRIC TECHNIQUES Theory vs. Empiricism T w o general approaches to form ulating a model exist. One is to consult economic theory. The other is to look fo r regularities in the data. Either can be used as a starting point, but a combination o f both is best. A model derived from elegant economic theory may be appealing, but unless at least some o f its components or implications are consistent w ith real data, it is not a reliable hypothesis. A model obtained by pure data min ing may be consistent with the body o f data that was mined to get it, but it is not a reliable hypothesis if it is not consistent w ith at least some other data (recall what was said about this earlier), and it w ill not be understood if no theory to explain it exists. The V A R A pproach Vector autoregression (VAR) is one w ay o f looking fo r regularities in data. In VAR, a set o f observable variables is chosen, a maximum lag length is chosen, and the current value o f each variable is regressed on the lagged values o f that variable and all other variables. No exogenous variables exist; all observable variables are treated as endogenous. Except fo r that, a VAR model is similar to the unrestricted reduced form o f a conventional econom etric model. Each equation contains only one current endogenous variable, each equation is just identified, and no use is made o f any possible theoretical information about possible simultaneous structural equations that might contain m ore than one current endogenous variable. In fact, no use is made o f any theoretical information at all, except in the choice o f the list o f variables to be included and the length o f the lags. In macroeconomics it is FEDERAL RESERVE BANK OF ST. LOUIS not practical to use many variables and lags in a VAR because the num ber o f coefficients to be estimated in each equation is the product o f the number o f variables times the number o f lags and because one cannot estimate an equation that has m ore coefficients than there are obser vations in the sample. The A R IM A Approach The Rox-Jenkins type o f time-series analysis is another w ay to seek regularities in data. Here each observable variable is expressed in terms o f purely random disturbances. This can be done with one variable at a time or in a multivariate fashion. In the univariate case an expression involving current and lagged values o f an observable variable is equated to an expression involving current and lagged values o f an unobservable white-noise disturbance; that is, a serially inde pendent random disturbance that has a mean o f zero and constant variance. Such a formulation is called an autoregressive integrated m oving aver age (ARIM A) process. Th e autoregressive part expresses the current value o f the variable as a function o f its lagged values. Th e integrated part refers to the possibility that the first (or higherorder) differences o f the variable, rather than its levels, may be governed by the equation. Then the variable's levels can be obtained from its differences by undoing the differencing operation—that is, by integrating first d iffe r ences once, integrating second differences twice, and so on. (If no integration is involved, the process is called ARM A instead o f ARIMA.) The moving average part expresses the equa tion’s disturbance as a m oving average o f cur rent and lagged values o f a white-noise disturbance. To express a variable in A RIM A form , it is necessary to choose three integers to character ize the process. One gives the order o f the auto regression (that is, the number o f lags to be included fo r the observable variable); one gives the order o f the m oving average (that is, the number o f lags included fo r the white-noise dis turbance); and one gives the order o f integration (that is, the number o f times the highest-order differences o f the observable variable must be integrated to obtain its levels). The choice o f the three integers (some o f which may be zero) is made by examining the time series o f data fo r the observable variable to see what choice best conform s to the data. A fter that choice has been made, the coefficients in the autoregression and m oving average are estimated. The multivariate form o f ARIM A modeling is a generalization o f the 77 univariate form . And, o f course, VAR modeling is a special case o f multivariate ARIM A modeling. ory but permit the adjustment path to be deter mined largely by data. VAR and ARIM A models can be useful if they lead to the discovery o f regularities in the data. If enduring regularities in the data are discovered, w e have something interesting to try to under stand and explain. In m y view, how ever, one disadvantage o f both approaches is that they make almost no use o f any knowledge o f the subject matter being dealt with. To use univari ate ARIM A on an economic variable, one need know nothing about economics. I think o f univariate ARIM A as mindless data mining. To use multivariate ARIMA, one need only make a list o f variables to be included and choose the required three integers. To use VAR, one need only make a list o f the variables to be included and choose a maximum lag length. Knowledge o f the subject the equations deal with can enter into the choice o f variables to be included. Testing Residuals f o r Random ness It may seem that the ARIM A approach and the conventional econom etric model approach are antithetical and inconsistent with each other. Zellner and Palm (1974), however, have pointed out that if a conventional model's exogenous variables are generated by an ARIM A process, the m odel’s endogenous variables are generated the same way. General-to-Specific Modeling General-to-specific modeling starts w ith an esti mated equation that contains many variables and many lagged values o f each. Its approach is to pare this general form down to a more specific form by omitting lags and variables that do not con tribute to the explanatory p ow er o f the equation. Much can be said fo r this technique, but o f course it w ill not lead to a correct result if the general form one starts w ith does not contain the variables and the lags that belong in an equation that is approximately correct. The E rror Correction Mechanism The error correction mechanism (ECM) provides a w ay o f expressing the rate at which a variable moves toward its desired or equilibrium value when it is away from that value. Economic theory is at its best when deriving desired or equilibrium values o f variables, either static positions or dynamic paths. ECM has so far not been good at deriving the path follow ed by an economy that is out o f equilibrium. Error correction models are appealing because they perm it the nature o f the equilibrium to be specified with the aid o f the I have already discussed testing residuals fo r randomness. If an equation's residuals appear to fol low any regular or systematic pattern, this is a sig nal that there may be some regular o r systematic factor that has not been captured by the form and variables chosen fo r the equation. In such a case it is desirable to try to m odify the equation’s specification, either by including additional vari ables, by changing the form o f the equation, or both, until the residuals lose their regular or systematic character and appear to be random. Stationarity It is often said that the residual o f a properly specified equation should be stationary, that is, that its mean, variance and autocovariances should be constant through time. H ow ever, fo r an equation whose variables are grow in g over time, such as an aggregate consumption or moneydemand equation, it would be unreasonable to expect the variance o f the residual to be constant. That would mean that the correlation coefficients fo r the equation in successive decades (or other time intervals) w ould approach one. It w ould be m ore reasonable to expect the standard deviation o f the residual to g ro w roughly in proportion to the dependent variable, to one o f the indepen dent variables, or to some combination o f them. The Lucas Critique Robert Lucas (1976) w arned that w hen an estimated econom etric model is used to predict the effects o f changes in governm ent policy var iables, the estimated coefficients may turn out w ron g and hence the predictions may also turn out wrong. Under what conditions can this be expected to occur? Lucas says that this occurs w hen policymakers follow one policy rule during the estimation period and begin to follow a different policy rule during the prediction period. Th e reason fo r this, he argues, is that in many cases the parameters that w ere estimated are not constants that represent invariant economic relationships, but instead are variables that change in response to changes in policy rules. This is because they depend both on constant parameters and on varying expectations that private agents form ulate by observing policy makers and trying to discover what policy rule is being followed. Jacob Marschak (1953) fo re shadowed this idea w hen he cautioned that MARCH/APRIL 1993 78 predictions made from an estimated econometric model w ill not be valid if the structure o f the model (that is, its mathematical form and its parameter values) changes betw een the estimation period and the prediction period. Th erefore, to make successful predictions after a structural change, one must discover the nature o f the structural change and allow fo r it. I take this w arning seriously. It need not con cern us w hen policy variations whose effects w e want to predict are similar to variations that occurred during the estimation period. But when a change in the policy rule occurs, private agents w ill eventually discover that their previous expectation formation process is no longer valid and will adopt a new one as quickly as they can. As they do so, some o f the estimated parameters will change and make the previously obtained esti mates unreliable. Goodhart’s Law Lucas’ warning is related to Goodhart’s Law, which states that as soon as policymakers begin to act as if some previously observed relation ship is reliable, it w ill no longer be reliable and w ill change.3 A striking example is the shortrun, downward-sloping Phillips curve. Are P olicy Variables Exogenous? Most econom etric models treat at least some policy variables as exogenous. But public policy responds to events. Policy variables are not exogenous. The field o f public choice studies the actions o f policymakers, treating them as maximizers o f their ow n utility subject to the con straints they face. Econometric model builders have so far not made much use o f public choice eco nomics. BY THEIR FORECASTS YE SH ALL K N O W THEM (MODELS, T H A T IS) Methods o f Evaluating M od els’ Forecasts A conventional econom etric model contains dis turbances and endogenous and exogenous varia bles. Typically some o f the endogenous variables appear with a lag. Consider an annual model with data fo r all variables up to and including 1992. Suppose that at the end o f 1992 w e wish to forecast the endogenous variables fo r 1993, one 3See Goodhart (1981). FEDERAL RESERVE BANK OF ST. LOUIS year ahead. This is an ex ante forecast. For this w e need estimates o f the m odel’s parameters, which can be computed from our available data. In addition, w e need 1993 values fo r the lagged endogenous variables. These w e already have because w e have values fo r the years 1992 and earlier. Further, w e need predicted 1993 values fo r the disturbances. W e usually use zeros here because disturbances are assumed to be serially independent with zero means. (Some modelers, how ever, would use values related to the residuals fo r 1992 and possibly earlier years if the disturbances w ere thought to be serially correlated.) Finally, w e need predicted 1993 values fo r the exogenous variables. These p re dictions must be obtained from some source outside the model. Our predictions o f the endogenous variables fo r 1993 w ill be conditional on our estimated model and on our predictions o f the disturbances and exogenous variables. If w e make errors in forecasting the endogenous variables, it may be because our estimated model is wrong, because our predictions o f the disturbances or exogenous variables are w rong, or because o f some combi nation o f these. It is possible—and desirable—to test the fo re casting ability o f an estimated model independently o f the model user's ability to forecast exogenous variables. This is done w ith an ex post forecast. An e,x post forecast fo r one period ahead, say fo r 1993, is made as follows: W ait until actual 1993 data fo r the exogenous variables are avail able, use them instead o f predicted values of the exogenous variables to compute forecasts o f the 1993 endogenous variables, and examine the errors o f those forecasts. W hen comparing forecasts from different m od els, bear in mind that the models may differ in their lists o f exogenous variables and that this may affect the comparison. For example, a model that has hard-to-forecast exogenous variables is not going to be helpful fo r practical ex ante forecasting, even if it makes excellent ex post forecasts. Errors o f ex ante and ex post forecasts tell us different things. Ex ante forecasting errors tell us about the quality o f true forecasts but do not allow us to separate the effects o f incorrect estimated models from the effects of bad predictions o f exogenous variables and disturbances. Ex post forecasting errors tell us how good an estimated model has been as a scientific hypothesis, which is 79 Table 1 Root Mean Square Percentage Errors of Ex Post Forecasts with No Subjective Adjustments of the Forecasts, from about 1965 to 1973, Averaged over Eight Models__________________ Horizon Variable Nominal GNP Real GNP GNP Deflator 1 quarter 0.7 . 0.7 0.4 distinct from anyone’s ability to forecast exogenous variables and disturbances. If you are interested in the quality o f practical forecasting, you should evaluate ex ante forecasts. If you are interested in the quality o f a model as a scientific theory, you should evaluate ex post forecasts. Ex post forecasts are usually m ore accurate than ex ante forecasts because the predictions o f the exogenous variables that go into ex ante forecasts are usually at least somewhat wrong. W hat if w e want to make forecasts two years ahead, fo r 1994, based on data up to and including 1992? W e need 1993 values fo r the endogenous variables to use as lagged endogenous values for our 1994 forecast; however, w e do not have actual 1993 data. Hence w e must make a one-year-ahead fo re cast fo r 1993 as before. Then w e can make our 1994 forecast using our 1993 forecasts as the lagged values o f the endogenous variables fo r 1994. Thus the errors o f our 1994 forecast w ill depend partly on the errors o f our 1993 fo re cast and partly on the values w e use fo r the 1994 exogenous variables and disturbances. If w e want to make forecasts fo r n years ahead instead o f tw o years ahead, the situation is similar except that n steps are required instead o f two. W e can still consider either ex ante or ex post forecasts. As before, ex post forecasts use actual values of the exogenous variables. W hen making ex ante forecasts, the typical econom etric forecaster does not automatically adopt the forecasts generated by a model. Instead the forecaster compares these forecasts with his subjective judgment about the future o f the economy, and if there are substantial dis crepancies, he makes subjective adjustments to his m odel’s forecasts. This is usually done with subjective adjustments to the predicted distur bances. Thus the accuracy o f ex ante forecasts 4 quarters 8 quarters 2.0 1.9 0.6 4.5 2.5 1.9 typically depends not only on the adequacy of the estimated model, but also on the model builder’s ability to forecast exogenous variables and to make subjective adjustments to the m od el's forecasts. Paul Samuelson once caricatured this situation at a meeting some years ago by likening the process that produces ex ante econom etric forecasts to a black box inside which w e find only Law rence R. Klein! Errors o f Forecasts f r o m Several Econom etric M odels Most presentations o f forecasting accuracy are based on ex ante rather than ex post forecasts, often with subjective adjustments, perhaps because o f the interest in practical forecasting. I like to look at ex post forecast errors without adjust ments because I am interested in econometric models as scientific hypotheses. Fromm and Klein (1976) and Christ (1975) discuss root mean square errors (RMSEs) o f ex post quarterly forecasts o f real GNP, nominal GNP and the GNP deflator one quarter to eight quarters ahead by eight models with no subjec tive adjustment by the forecaster. The models w ere form ulated by Brookings, the U.S. Bureau o f Economic Analysis, Ray Fair, Leonall Ander sen o f the Federal Reserve Bank o f St. Louis, T. C. Liu and others, the University o f Michigan and the W harton School (two versions). For GNP they show RMSEs rising from 0.7 percent to 2.5 or 4.5 percent o f the actual value as the horizon increases from one quarter to eight quarters. For the GNP deflator they show RMSEs rising from 0.4 percent to 1.9 percent, as shown in table 1. In a series o f papers over the past several years, Stephen McNees (1986, 1988 and 1990) has reported on the accuracy o f subjectively MARCH/APRIL 1993 80 adjusted e^ ante quarterly forecasts o f several m acroeconom etric models, fo r horizons o f one to eight quarters ahead, and has compared them w ith tw o simple mechanical forecasting methods. One is the univariate ARIM A method o f Charles Nelson (1984), which is called BMARK (for benchmark). The other is the Bayesian vec tor autoregression method o f Robert Litterman (1986), which is called BVAR. The models dis cussed in McNees (1988) are those form ulated by the U.S. Bureau o f Economic Analysis, Chase Econometrics, Data Resources Inc., Georgia State University, Kent Institute, the University o f Michigan, UCLA and Wharton. McNees' results fo r quarterly forecasts may be summarized in the follow ing five statements: 1. The models' forecast errors w ere usually smaller than those o f BMARK.4 2. The models’ forecast errors w ere usually slightly smaller than those o f BVAR fo r nominal GNP and most other variables and slightly larger than those o f BVAR fo r real GNP. Thus BVAR was usually better than BMARK fo r real GNP.5 3. Forecast errors fo r the levels o f variables became w orse as the forecast horizon length ened from one quarter to eight quarters, roughly quadrupling fo r most variables and increasing tenfold fo r prices. However, fo re cast errors fo r the growth rates o f many vari ables (but not fo r price variables) im proved as the horizon lengthened. In other words, fo r many variables, the forecasts fo r grow th rates averaged over several quarters w ere better than the forecasts fo r short-term fluc tuations.6 4. Mean absolute errors (MAEs) o f the models' forecasts o f the level o f nominal GNP w ere usually about 0.8 percent o f the true level for forecasts one quarter ahead and increased gradually to about 2.2 percent fo r forecasts one year ahead and about 4 percent fo r fo re casts tw o years ahead. Real GNP forecast errors w ere somewhat smaller. Errors fo r other variables w ere comparable. Price-level forecast errors w ere smaller fo r the onequarter horizon but g rew faster and w ere larger fo r the two-year horizon.7 5. W hen subjectively adjusted forecasts w ere compared w ith unadjusted forecasts, the adjustments w ere helpful in most cases, though sometimes they made the forecast worse. Usually the adjustments w ere larger than optimal.8 One-year-ahead annual forecasts o f real GNP by the University o f Michigan’s Research Center in Quantitative Economics, by the Council o f Economic Advisers and by private forecasters covered by the ASA/NBER survey all had MAEs o f about 0.9 percent to 1.1 percent o f the true level, and RMSEs o f about 1.2 percent to 1.5 percent o f the true level.9 (The relative sizes o f the MAEs and RMSEs are roughly consistent with the fact that fo r a normal distribution, the RMSE is about 1.25 times the MAE.) Implications o f Worsening Ex Post Forecast Errors Because the root mean square error o f an econom etric m odel’s post forecasts roughly quadruples w hen the horizon increases from one quarter to eight quarters as in table 1, can w e conclude that the model is no longer correct fo r the forecast period? The answer is possibly, but not certainly. For a static model w e could conclude this because the error o f each forecast w ould involve distur bances only fo r the period being forecast, not fo r periods in the earlier part o f the horizon. Hence there is no reason to expect great changes in the size o f the forecasting erro r fo r a static model as the horizon increases. Small increases will occur because o f errors in the estimates o f the m odel’s parameters if the values o f the m od el's independent variables m ove fu rth er away from their estimation-period means as the hori zon lengthens. This is because any errors in the estimates o f equations’ slopes w ill generate larger effects as the distance over which the slopes are projected increases. But most econom etric forecasting models con tain lagged endogenous variables. Th erefore, as noted previously, to forecast n periods ahead, w e must first forecast the lagged endogenousvariable values that are needed fo r the n-periods- 4See McNees (1988 and 1990). 7See McNees (1988). 5See McNees (1990). 8See McNees (1990). 6See McNees (1988). 9See McNees (1988). FEDERAL RESERVE BANK OF ST. LOUIS 81 ahead forecast. This involves a chain o f n steps. The first step is a forecast one period ahead, whose error involves disturbances only from the first period in the n-period horizon. The second step is a forecast tw o periods ahead, w hose erro r involves disturbances from the sec ond period in the horizon and also disturbances from the first period because they affect the one-period-ahead forecast, which in turn affects the two-periods-ahead forecast. And so on, until the nth step, whose forecast error involves dis turbances from all periods in the horizon from one through n. Thus, fo r a dynamic model, the variance o f a forecast n periods ahead w ill depend on the variances and covariances o f disturbances in all n periods o f the horizon, and except in very special circumstances, it w ill increase as the horizon increases. T o decide w hether the evidence in table 1 shows that the estimated models it describes are incor rect fo r the forecast horizon o f eight quarters, w e need to know w hether the RMSEs o f a correct model w ould quadruple as the forecast horizon increases from one quarter to eight quarters. If they would, then the quadrupling observed in the table is not evidence o f incorrectness o f the estimated models. If they would not, then evi dence o f incorrectness exists. W e do not have enough information about the models underly ing the table to settle this issue definitively, but some simple examples w ill illustrate the principle involved. Suppose the model is linear and perfectly cor rect, and suppose it contains lags o f one quarter or m ore (as most models do). Then the variance o f the error o f an n-periods-ahead forecast will be a linear combination o f the variances and covariances o f the disturbances in all periods o f the horizon. In the simple case o f a single-equation model, if the disturbances are serially indepen dent and if the coefficients in the linear com bi nation o f disturbances are all equal to one, the variance o f the linear combination o f distur bances fo r a horizon o f eight quarters w ill be eight times that o f one quarter. So the RMSE o f e,x post forecast errors from a correct model w ill increase by a factor o f the square root o f eight (about 2.8) as the horizon goes from one quarter to eight quarters. I f the coefficients in the linear combination are less than one, as in the case o f a stable model w ith only one-period lags, the variance o f the linear combination fo r eight quarters w ill be less than eight times that fo r one quarter. Hence the RMSE o f e?c post forecast errors from a correct model will increase by less than a factor o f the square root o f eight as the horizon goes from one quarter to eight quarters. In such a case, if the observed RMSEs approximately quadrupled, it would cast some doubt on the validity o f the model. Consider a single-equation model with a single lag, and no exogenous variables as follows: y, = a + 0yt_, + e, w h ere £ is a serially independent disturbance with zero mean and constant variance o". Suppose that the values o f a and P are known and thus no fo re cast error is attributable to incorrect estimates of these coefficients. Then the variance o f the error 2 o f a one-period-ahead forecast is a", that o f a twoperiods-ahead forecast is (1 + p ) o~, that o f a three-periods-ahead forecast is (1 + p + p )cf, and so on. The variance o f an n-periods-ahead forecast is ' /T' o', which is equal to (1 - /TN a"/( 1 - p i. ) Table 2 shows how the standard deviation o f such a forecast error increases as the horizon increases from one quarter to eight quarters fo r several values o f the parameter p. Table 2 sug gests that if the RMSE o f a m odel’s forecasts quadruples as the horizon increases from one quarter to eight quarters, either P (the rate o f approach o f the model to equilibrium) must be large or close to one, or the model is inade quate as a description o f the forecast period. Corresponding expressions can be derived fo r multi-equation models w ith many lags and serially correlated disturbances, but they are rather cumbersome. AN OLD, P L A IN -V A N IL L A E Q U A TIO N T H A T STILL W O RK S, R O UG H LY Nearly 40 years ago Henry Allen Latane pub lished a short paper in which he reported that fo r 1919-52 the inverse o f the GNP velocity o f M l is described by a simple least squares regression on the inverse o f a long-term, highgrade bond rate RL as follow s:1 0 (1) Ml/GNP = .100 + ,795/RL, r" = .75 (t-ratio) (10) 10See Latane (1954). MARCH/APRIL 1993 82 Table 2 Standard Deviation of Error of N-Periods-Ahead Forecast from the Equation yt = a + /?yt_, + et Relative to the Standard Deviation of t, as a Function of N and [i, when a and ft Are Known Horizon, N 2 p P 1 2 4 0.7070 0.8944 0.9486 1.0000 0.50 0.80 0.90 1.00 1.00 1.00 1.00 1.00 1.22 1.34 1.38 1.41 1.37 1.72 1.85 2.00 Here and in what follows, I have expressed inter est rates in units o f percent per year, so a 5 per cent rate is entered as 5, not as 0.05, and its inverse 0.20, not 20. The Appendix gives the definitions and data sources fo r variables in this and sub sequent equations. Latane showed the unad justed correlation coefficient r, but showed nei ther the standard deviation nor the t-ratio o f the slope. I calculated the adjusted r and the tratio. The latter is the square root o f r2 (df) 1(1 - r 2 ), w h ere df, the number o f degrees o f freedom , equals 32. This specification has some o f the properties o f a theoretical m oney demand equation—namely, a positive income elasticity (restricted to be con stant and equal to one by construction) and a negative interest elasticity (restricted to have an absolute value less than one and not constant). But its least-squares estimate w ould almost cer tainly be biased or inconsistent, even if the form o f the equation w ere correct, because the bond rate is almost certainly not exogenous and hence not independent o f the equation’s disturbances. Nevertheless, this specification has continued to w ork fairly w ell fo r other periods. Nearly 30 years ago Ml/GNP was described fo r 1892-1959 by a similar regression on the inverse o f M oody’s Aaa bond rate w ith almost the same coefficients, as follow s:1 1 (2) Ml/GNP = .131 + ,716/RAaa, r 2 = .76 (t-ratio) (14) 1 See Christ (1963). 1 FEDERAL RESERVE BANK OF ST. LOUIS 8 1.41 2.04 2.39 2.83 oo 1.41 2.24 3.16 oo For 1959-91 the same specification describes the ratio o f M l to GNP w ith almost the same coefficients, as follows: (3) Ml/GNP = .085 + .774/RAaa, r 2 = .90 (t-ratio) (13) (17) If GNP in equation (3) is replaced by the new output variable GDP fo r 1959-91, the result is almost identical, as follows: (4) Ml/GDP = .086 + ,771/RAaa, ? 2 = .91 (t-ratio) (13) (18) David Dickey’s discussion is based on the 1959-91 data that underlie equation (3). For 1892-1991 a similar result is again obtained, as follows: (5) Ml/GNP = .083 + ,874/RAaa, ? 2 = .89 (t-ratio) (11) (28) Table 3 shows the estimated equations (1) —(5) and several other estimated equations that w ill be described soon. Equations ( l 7 and (27 are ) ) attempts to duplicate the results in equations (1) and (2) using the same data base that is used in equations (3), (5) and later equations. The Appen dix gives data sources. Figure 2 shows the graphs o f Ml/GNP and 1/RAaa over time. Figures 3 and 4 show the scatter diagrams fo r equations (3) and (5), respectively. (I should add that, o f the four 83 Table 3 Regressions of M1/GNP or M1/GDP on 1/RAaa and Other Variables* (t-ratios are in parentheses)______________________________________ Estimated Coefficient (and t-ratio) Eq Sample Constant PLAIN-VANILLA EQUATIONS 1 1919-1952 .100 1' 1919-1952 .136(7) 2 1892-1959 .131 2' 1892-1959 .132(9) 3 1959-1991 .085(13) 4 1959-1991 .086(13) 5 1892-1991 .083(11) AR(1) EQUATIONS** 15 1960-1991 .013 17 1893-1991 .020 ECM EQUATIONS** 18 1960-1991 .016(2.2) 19 1893-1991 .016(2.1) 1/RAaa (M1/GNP)_, (1/R A aa )., .795(10) .713(10) .716(14) .712(14) .774(17) .771(18) .874(28) R DW .75 .75 .76 .73 .90 .91 .89 .62 .38 .36 .48 .56 .267(2.8) 711(7) .896(26) .831(12) -.2 3 9 -.5 9 1 .98 .95 1.82 1.60 .275(2.8) .593(5) .857(11) .807(12) - .212( —2.2) - ,428( -3 .6 ) .98 .95 1.78 1.59 'T h e dependent variable is M1/GNP in all equations except equation (4), where it is M1/GDP. Definitions and data sources for the variables M1, GNP, GDP and RAaa are given in the Appendix. As explained in the Appendix, a uniform set of data for M1, GNP and RAaa was used for equations (1'), (2'), (3), (5) and later equations; slightly different data were used for equations (1), (2) and (4). David Dickey’s discussion is based on equation (3) and the data underlying it. Equations (1') and (2') are attempts to reproduce equations (1) and (2), respectively, using the same data that were used for equations (3), (5) and later equations. **AR(1) means “ first-order autoregressive.” ECM means “ error correction mechanism.” equations that can be obtained by regressing either the velocity o f M l or its inverse on either RAaa or its inverse, the form that is presented here fits the best.) It is rather remarkable that this plain-vanilla specification continues to describe the relation betw een M l's velocity and the long-term Aaa bond rate with such similar regression and cor relation coefficients fo r the four periods, espe cially in view o f the changes in interest-rate regulation and in the definition o f M l that have occurred over the last century. How ever, the differences among the four estimated versions are not negligible, as seen in a comparison o f the computed values o f Ml/GNP that they yield. For 1959-91 these computed values are shown in figure 5 together with the actual values o f Ml/GNP. Note that those computed from equa tions (1) and (2) using 1919-52 and 1892-1959 data are ex post forecasts, whereas those from equations (3) and (5) using 1959-91 and 1892-1991 data are within-sample calculated values. Figure 6 shows the values o f Ml/GNP obtained when equation (3) based on 1959-91 data is used to backcast Ml/GNP fo r 1892-1958, and it also shows the actual values and the calculated values from equation (5) using 1892-1991 data. Th e forecasting and backcasting errors are by no means negligible, but the general pattern o f behavior o f Ml/GNP is reproduced. The estimates o f the plain-vanilla equation are rather stable across time, as indicated by fig ures 7 and 8 which show the behavior o f the slope as the sample period is gradually length ened by adding one yea r at a time. In figure 7 the sample period starts w ith 1959-63 and is extended a year at a time to 1959-91. In figure 8 the sample period starts w ith 1892-97 and is gradually extended to 1892-1991. In each figure the slope settles dow n quickly after jumping around at first and varies little as the sample is extended thereafter. H ow ever, this simple specification does not by any means satisfy all o f the desiderata listed previously. In particular, the 1959-91 DurbinWatson statistic is a minuscule 0.38, and the 1892-1991 Durbin-Watson statistic o f 0.48, is not much better, which suggests that the residuals have a strong positive serial correlation. This by itself would not create bias in the estimates if MARCH/APRIL 1993 84 Figure 2 M1/GNP and 1/RAaa, 1892-91 Figure 3 Regression of M1/GNP on 1/RAaa, 1959-91 M1/GNP 1/RAaa FEDERAL RESERVE BANK OF ST. LOUIS 85 Figure 4 Regression of M 1/GNP on 1/RAaa, 1892-91 M1/GNP 1/RAaa Figure 5 Actual, Computed and Forecast Values of M1/GNP from Regressions on 1/RAaa for Four Periods MARCH/APRIL 1993 86 Figure 6 Actual, Computed and Backcast Values of M1/GNP from Regressions on 1/RAaa for Two Periods Figure 7 Estimates of Slope in Regression of M1/GNP on 1/RAaa for Samples Starting in 1959 and Ending in 1963...1991 FEDERAL RESERVE BANK OF ST. LOUIS 87 Figure 8 Estimates of Slope in Regression of M1/GNP on 1/RAaa for Samples Starting in 1892 and Ending in 1897...1991 the equation form w ere correct and if the dis turbance w ere independent o f the interest rate and had zero mean and constant variance. But it certainly suggests strongly that the equation has not captured all its relevant systematic fac tors. The graph o f the residuals o f the 1959-91 equation (3) against time is illuminating. It shows an almost perfect 12-year cycle o f diminishing amplitude with peaks (positive residuals) in 1959 (or possibly earlier), 1970 and 1982 and troughs (negative residuals) in 1965, 1977 and 1990. It also suggests a negative time trend. The residuals o f the 1892-1991 equation (5) show a roughly sim ilar pattern. (See figures 9 and 10.) The very low Durbin-Watson statistics suggest that the equation should be estimated either using the first differences o f its variables, or better, using the levels o f its variables with a first-order autoregressive [AR(1)] correction applied to its residuals. Estimation in levels with an AR(1) correction w ould be appropriate if the disturbance u in the original equation w ere equal to its ow n lagged value times a constant, p, plus a serially independent disturbance, £, with constant variance, as follows: (6) ut = p u t_, + £, In this case, if the original equation is (7) y, = a + / x + u, = a + /J , + pu,_, + £t ?, x , the AR(1) correction subtracts p times the lagged version o f equation (7) from equation (7) itself and produces the follow ing equation: (8) y, = py,_, + (1 - p ) a + (ixl - pp\l t + £, This equation is nonlinear in the parameters because the coefficient o f lagged x, -ftp , is the negative o f the product o f the coefficients o f x and lagged y. If that restriction is ignored and the coefficient o f lagged x is denoted by y, the equation becomes as follows: (9) y, = Py,_, + (1 - p) a + Pxt + yx( j + £ , This equation can be given the follow ing error correction interpretation. Suppose that the equil ibrium value y * o f a dependent variable y is linear in an independent variable x, as follows: (10) y * = a + lixt and that the change in y depends on both the change in the equilibrium value and an error MARCH/APRIL 1993 88 Figure 9 Residuals from Regression of M1/GNP on 1/RAaa, 1959-91 Figure 10 Residuals from Regression of M1/GNP on 1/RAaa, 1892-1991 FEDERAL RESERVE BANK OF ST. LOUIS 89 correction term proportional to the gap betw een the lagged equilibrium and the lagged actual values, as follows: (11) Ay, = 0 A y *+ (1 - p) (y *_, - y,_,) +£, Substitution from equation (10) into equation (11) implies an equation with the same variables as the AR(1) equation (8) but with some different parameters, as follows: (12) y, = py, , + (1 - p )a + 0/Jx, + (1 - p - 0)/3x, , + £t If the adjustment parameter 9 in equation (12) w ere equal to one, then equation (12) would becom e the same equation as (8). Estimates in first differences would be appro priate if the value o f p in equation (6), (7) and (8) w ere one. In this case, equation (8) becomes a first-difference equation, as follows: The least-squares estimate o f equation (8) in levels with the AR(1) correction fo r 1960-91 is as follows: , = (1 - .896).126 + .267(1/RAaa - ,896/RAaa ,) (t-ratio) (8) (2.8) (26) with an adjusted R squared o f .98 and D W equal to 1.82. This is equivalent to the follow ing equation: (15) Ml/GNP = .896(M1/GNP) , + .013 + ,267/RAaa - ,239/RAaa , There is no evidence o f a trend. The least-squares estimate in levels with the AR(1) correction fo r 1893-1991 is as follows: (16) Ml/GNP - .83KM1/GNP) , = (1 - .831).117 + .71 l(l/RAaa - .831/RAaa ,) (t-ratio) (5) (7) (12) with an adjusted R squared o f .95 and DW equal to 1.60. This is equivalent to the follow ing equation: (17) Ml/GNP = ,831(M1/GNP)_, + .020 + ,711/RAaa - ,591/RAaa , Th ere is again no evidence o f a trend. (18) Ml/GNP = ,857(M1/GNP) , (t-ratio) (11) + .016 + ,275/RAaa - ,212/RAaa (2 .2) (2 .8) ( - 2 .2) with an adjusted R squared o f .98 and DW equal to 1.78. This is quite close to the AR(1) result in equation (15), which suggests that the adjustment coefficient 9 in equation (12) is not very different from one. The hypothesis that in equation (18) the coefficient o f lagged 1/RAaa is equal to the negative o f the product o f the coefficients o f 1/RAaa and lagged Ml/GNP, as required by equation (8) and as satisfied by equation (15), is strongly accepted by a Wald test (the p-value is .59). Least-squares estimation o f equation (12) for 1893-1991 (again without restricting 9 to be one) yields the follow ing equation: (13) Ay, = /?Ax, + e, (14) Ml/GNP - ,896(M1/GNP) Least-squares estimation o f the ECM equation (12) fo r 1960-91 (without restricting 9 to be one) yields the follow ing equation: (19) Ml/GNP = ,807(M1/GNP) , (t-ratio) (12) + .016 + ,593/RAaa - ,428/RAaa , (2.1) (5) (-3 .6 ) w ith an adjusted R squared o f .95 and D W equal to 1.59. This is quite close to the AR(1) result in equation (17), which again suggests that the adjustment coefficient 9 in equation (12) is not very different from one. The hypothesis that in equation (19) the coefficient o f lagged 1/RAaa is equal to the negative o f the product o f the coefficients o f 1/RAaa and lagged Ml/GNP, as required by equation (8) and as satisfied by equation (17), is accepted by a W ald test (the p-value is .11). Equations (15), (17), (18) and (19) are better than the plain-vanilla equations (3) and (5) in some respects, and w orse in others. They have substantially higher adjusted R-squared values, much less serial correlation in their residuals, no evidence o f a time trend, and significant coefficients. The ECM equations (18) and (19), how ever, are very unstable over time. In equa tion (18) the coefficient o f 1/RAaa varies from about .6 fo r 1960-70, to .05 fo r 1960-78 and 1960-81, to .3 fo r 1960-86 and 1960-91. In equation (19) the coefficient o f 1/RAaa varies almost as much but remains at about .7 or .6 for samples that include at least the years 1893-1950. MARCH/APRIL 1993 90 Table 4 Regressions of A(M1/GNP) on A(1/RAaa) without a Constant* (t-ratios are in parentheses)_______________________________ Eq Sample 24 25 1960-1991 1893-1991 Coef of A(1/RAaa) R2 DW .380(3.6) .494(4.1) .05 .15 1.23 1.76 ‘ Definitions and data sources for the variables M1, GNP and RAaa are given in the appendix. I conjecture that in the AR(1) equations (15) and (17) the coefficient o f 1/RAaa is also unstable across time because the AR(1) and ECM equa tion estimates are quite similar. By comparing equations (12) and (18), one can solve fo r the 1960-91 estimates o f the four parameters p, a, / and 0, in that order, to 3 obtain: (20) p = .857, a = .112, ft = .441 and 8 = .624 This implies that the equilibrium relation in equation (10) em bedded in the ECM is as follows: (21) (Ml/GNP)* = .112 + ,441/RAaa Similarly, by comparing equations (12) and (19) one can solve fo r the 1893-1991 estimates o f the four parameters as follows: (22) p = .807, a = .083, / = .855 and 6 = .694 ? This implies that the equilibrium relation in equation (10) embedded in the ECM is as follows: (23) (Ml/GNP)* = .083 + ,855/RAaa The tw o equilibrium relations in equations (21) and (23) fo r the tw o periods 1960-91 and 1893-1991 are quite different, which is consis tent with the instability o f the ECM specification across time. Now let us return to the first-difference equation (13). The least-squares estimate fo r 1960-91 is as follows: (24) A(M1/GNP) = .380A(l/RAaa), r 2 = .05 (t-ratio) (3.6) FEDERAL RESERVE BANK OF ST. LOUIS with D W = 1.23. For 1893-1991 it is as follows: (25) AIMl/GNP) = .494A( 1/RAaa), ? 2 = .15 (t-ratio) (4.1) w ith D W = 1.76. Table 4 shows the estimated equations (24) and (25). The estimates o f this first-difference specification are not quite as stable across time as those o f the specification in levels o f the variables. This can be seen by comparing equations (24) and (25) and also from figures 11 and 12, which show the values o f the estimates as the sample is increased one yea r at a time, starting respectively w ith 1960 and 1893. In each figure the estimates stabilize after an initial period o f instability, but the values at which they settle d iffer by a factor o f about .75. If a constant term is included in equation (24), which implies a trend term in equation (3), the constant is small but significantly negative, the slope falls to about .3, and the adjusted Rsquared and D W values im prove slightly. The estimated slope, how ever, becomes w ildly unsta ble across time. I f a trend variable is included in equation (3), its coefficient is small but signifi cantly negative, the interest-rate coefficient falls to .49 and remains highly significant, the adjusted R-squared and the D W values rise slightly, and again the estimated slope is w ildly unstable across time. If a constant term is included in equation (25), it is small and insignificantly negative, the rest o f the equation is almost unchanged, and the slope becomes quite unstable through time, varying from .6 to zero and back to .6 again. If a trend is included in equation (5), its co effi cient is small but significantly negative, the interest-rate coefficient is almost unchanged at .81, the adjusted R-squared value rises a bit, the D W value rises a bit, and the coefficient is again w ildly unstable across time. 91 Figure 11 Estimates of Slope in Regression of A(M1/GNP) on A(1/RAaa) for Samples Starting in 1960 and Ending in 1962...1991 10 5 \+ 2 Standard Error ♦ 0 Estimates of Slope .♦** - 2 Standard Error -5 -10 1962 ------- 1-------1 ------- 1 --------1 ------- 1 --------1 ------- 1 --------1 ------- 1 --------1-------1-------T 1 64 66 68 70 72 74 76 78 80 82 84 86 88 1990 Figure 12 Estimates of Slope in Regression of A(M1/GNP) on A(1/RAaa) for Samples Starting in 1893 and Ending in 1896...1991 MARCH/APRIL 1993 92 On the whole, the first-difference specification does not stand up well. W here do matters stand? On the one hand, w e have the plain-vanilla equation such as equa tion (3), which fits only m oderately w ell and has severe serial correlation in its residuals but has an estimated slope that is rather stable across time. On the other hand, w e have m ore compli cated dynamic equations such as the ECM equa tion (18), which fit much better and have nice Durbin-Watson statistics but have estimated coefficients that vary greatly across time. Nei ther is quite satisfactory, but if the aim is to find an estimated equation that w ill describe the future as w ell as it does the past, I think I would now bet on the plain-vanilla specification, even though the relation o f its estimated coefficients to structural parameters is unclear. CONCLUSION Econometrics has given us some results that appear to stand up w ell over time. The price and income elasticities o f demand fo r farm products are less than one. The income elastic ity o f household demand fo r food is less than one. Houthakker (1957), in a paper com m em orating the 100th anniversary o f Engel’s law, reports that fo r 17 countries and several different periods these income elasticities range betw een .43 and .73. Rapid inflation is associated w ith a high grow th rate o f the money stock. Some short-term m acroeconom etric forecasts, especially those o f the Michigan model, are quite good. But there have also been some nasty surprises about which econometrics gave us little or no w arning in advance. The short-run downwardsloping Phillips curve met its demise in the 1970s. (Milton Friedman [1968] and Edmund Phelps [1968] predicted that it would.) The oil em bargo o f 1973 and its aftermath threw most models off. The slowdown o f productivity grow th beginning in the 1970s was unforeseen. The money demand equation, which appeared to fit w ell and be quite stable until the 1970s, has not fit so w ell since then. H ow then should w e approach econometrics, fo r science and fo r policy, in the future? As fo r science, w e should form ulate and estimate models as w e usually do, relying both on economic theory and on ideas suggested by regularities observed in past data. But w e should not fail to test those estimated models against new data FEDERAL RESERVE BANK OF ST. LOUIS that w ere not available to influence the process o f formulating them. As fo r policy, w e should be cautious about using research findings to predict the effects o f any large policy change o f a type that has not been tried before. REFERENCES Christ, Carl F. “ Interest Rates and ‘Portfolio Selection’ among Liquid Assets in the U.S.,” in Christ et al., Meas urement in Economics: Studies in Mathematical Economics and Econometrics in Memory of Yehuda Grunfeid (Stanford University Press, 1963). ________“ Judging the Performance of Econometric Models of the U.S. Economy,” International Economic Review (February 1975), pp. 54-74. Friedman, Milton. Review of “ Business Cycles in the United States of America, 1919-1932” by Jan Tinbergen, American Economic Review (September 1940), pp. 657-60. _______ . “ The Role of Monetary Policy,” American Eco nomic Review (March 1968), pp. 1-17. Fromm, Gary, and Lawrence R. Klein. “ The NBER/NSF Model Comparison Seminar: An Analysis of Results,” Annals of Economic and Social Measurement (Winter 1976), pp. 1-28. Goodhart, Charles. “ Problems of Monetary Management: The U.K. Experience,” in A. S. Courakis, ed., Inflation, Depression, and Economic Policy in the IVesf (Barnes and Noble Books, 1981). Houthakker, Hendrik. “ An International Comparison of Household Expenditure Patterns, Commemorating the Centenary of Engel’s Law,” Econometrica (October 1957), pp. 532-51. Kendrick, John. Productivity Trends in the United States (Princeton University Press, 1961). Latane, Henry Allen. “ Cash Balances and the Interest Rate— A Pragmatic Approach,” Review of Economics and Statistics (November 1954), pp. 456-60. Litterman, Robert B. “ Forecasting with Bayesian Vector Autoregressions— Five Years of Experience,” Journal of Business and Economic Statistics (January 1986), pp. 25-38. Lucas, Robert E. Jr. “ Econometric Policy Evaluation: A Critique,” The Phillips Curve and Labor Markets, Carnegie-Rochester Conference Series on Public Policy, vol. 1, (North-Holland, 1976), pp. 19-46. Marschak, Jacob. “ Economic Measurements for Policy and Prediction,” in William C. Hood and Tjalling C. Koopmans, eds., Studies in Econometric Method, Cowles Commission Monograph No. 14 (Wiley, 1953), pp. 1-26. McNees, Stephen K. “ The Accuracy of Two Forecasting Techniques: Some Evidence and an Interpretation,” New England Economic Review (March/April 1986), pp. 20-31. ________“ How Accurate Are Macroeconomic Forecasts?” New England Economic Review (July/August 1988), pp. 15-36. . “ Man vs. Model? The Role of Judgment in Fore casting,” New England Economic Review (July/August 1990), pp. 41-52. Mitchell, Wesley C. Business Cycles: The Problem and Its Set ting (National Bureau of Economic Research, 1927). Nelson, Charles R. “ A Benchmark for the Accuracy of Econometric Forecasts of GNP,” Business Economics (April 1984), pp. 52-58. 93 Phelps, Edmund. “ Money-Wage Dynamics and Labor-Market Equilibrium,” Journal of Political Economy {Pan II, July/August 1968), pp. 678-711. Tinbergen, Jan. Business Cycles in the United States of America, 1919-1932, Statistical Testing of Business Cycle Theories, vol. 2, (League ot Nations, 1939). Zellner, Arnold, and Franz Palm. “ Time Series Analysis and Simultaneous Equation Econometric Models,” Journal of Econometrics (May 1974), pp. 17-54. A p pen dix On Data F or T ables 3 and 4 A . DataJ'or equations ( l ') , (2 '), (3), (5), (1 4 -1 9 ), and (2 4 -2 5 ): 1957 (Government Printing Office, 1960), p. 656, series X-333. 1939-91: Econom ic Report o f the President, 1992, p. 378. M l = currency plus checkable deposits, bil lions o f dollars 1892-1956, June 30 data: U.S. Bureau o f the Census. Historical Statistics o f the U.S. fr o m Colonial Times to 1957 (Government Printing Office, 1960), p. 646, series X-267. 1957-58, June 30 data: Economic Report o f the President, 1959, p. 186. 1959-91, averages o f daily data fo r December, seasonally adjusted: Eco nomic Report o f the President, 1992, p. 373. Note: Decem ber data, seasonally adjusted, are close to June 30 data. GNP = gross national product, billions o f dollars per year 1892-1928: Kendrick (1961), pp. 296-7. 1929-59: Economic Report o f the President, 1961, p. 127. 1960-88: Econom ic Report o f the President, 1992, p. 320. 1989-91: Survey o f Current Business, July 1992, p. 52. RAaa = long-term high-grade bond rate, percent per year 1892-1918: Macaulay's unadjusted railroad bond rate, U.S. Bureau o f the Census. Historical Statistics o f the United States fr o m Colonial Times to 1957 (Government Printing Office, 1960), p. 656, series X-332. 1919-91: Moody's Aaa corporate bond rate: 1919-38: U.S. Bureau o f the Census. Historical Statistics o f the United States fr o m Colonial Times to Note: For pre-1959 data I used sources that w ere available in 1960, in an attempt to make equation 2' reproduce the 1892-1959 equation 2, which originally appeared in Christ (1963). These same sources also yield equation 1', which is an approximate reproduc tion o f the 1919-52 equation 1, from Latane (1954). B. Data f o r 1959-91 f o r equation (4): M l = currency plus checkable deposits, billions o f dollars: same as above. GDP = gross domestic product, billions o f dol lars per year: Econom ic Report o f the President, 1992, pp. 298 or 320. RAaa = M oody’s Aaa corporate bond rate, per cent per year: same as above. C. Data f o r 1919— f o r equation (1), as 52 d escrib ed in Latane (1954), p. 457 :* M l: "demand deposits adjusted plus cur rency in circulation on the mid-year call date, (Federal Reserve Board Data).” U.S. Bureau o f the Census. Historical Statistics o f the United States fr o m Colonial Times to 1957 (Government Printing Office, 1960). Series X-267 GNP: "Departm ent o f Commerce series from 1929 to date; 1919-28 Federal Reserve Board estimates on the same basis (National Industrial Conference Board, Econom ic Almanac, 1952, p. 201).” 'Though Latane’s work was published in 1954, research analysts at the Federal Reserve Bank of St. Louis used more recent data to replicate his work. MARCH/APRIL 1993 94 RAaa: "interest rate on high-grade long-term corporate obligations. The U.S. Treasury series giving the yields on corporate high-grade bonds as reported in the Federal Reserve Bulletin is used from 1936 to date. Before 1936 w e use annual aver ages o f Macaulay’s high-grade railroad bond yields given in column 5, Table 10, o f his Bond Yields, Interest Rates, Stock Prices,” pp. A157-A161. Macaulay, Frederick R. Bond Yields, Interest Rates, Stock Prices (National Bureau o f Economic Research, 1938). D. Data f o r 1892—1959 f o r equation (2), as d escribed in Christ (1963), pp. 2 1 7 -1 8 ;2 M l: “ currency outside banks” plus “ demand deposits adjusted”, “ billions o f dollars as o f June 30.” 2Though Christ’s work was published in 1963, research analysts at the Federal Reserve Bank of St. Louis used more recent data to replicate his work. FEDERAL RESERVE BANK OF ST. LOUIS U.S. Bureau o f the Census. Historical Statis tics o f the United States fr o m Colonial Times to 1957 (Government Printing Office, 1960). Series X-267 U.S. Bureau o f the Census. Historical Statis tics o f the United States fro m Colonial Times to 1957; Continuation to 1962 and Revisions (Government Printing Office, 1965). Series X-267 RAaa: "long-term interest rate (M oody’s Aaa corporate bond rate, extrapolated before 1919 via Macaulay's railroad bond yield index)”, "percent per year.” GNP: "gross national product, billions o f dollars per year.” 95 David A. Dickey David A. Dickey is a professor of statistics at North Carolina State University. Commentary IRST, LET ME EXPRESS my appreciation fo r the invitation to participate in this conference. I have made several visits to the Federal Reserve Bank o f St. Louis and have enjoyed the hospitality o f Ted and his associates. Carl Christ’s paper was interest ing and thoughtful, prompting us to look again at some philosophical issues in econometric modeling. Tryin g to describe an ideal econom etric model makes sense to me. W hen I am in the market fo r a car, camera or other piece o f technological equip ment, I often look at the top-of-the-line item to see what it can do and then decide which features I can give up to make my purchase affordable. Carl Christ has done the same sort o f shopping fo r an econom etric model, searching fo r the best o f all possible models. W e likely cannot afford it, in the sense that w e cannot really afford to fo r mulate a model now and go fishing fo r several years w hile test data accumulates. Nevertheless, looking at the top-of-the-line type o f model will let us see an upper bound on what w e can expect models to give us and will give us a target point to m ove tow ard even though w e have no hope o f actually hitting the target. Researchers see some o f the same statistical strengths and weaknesses o f econometrics when they apply statistics to the biological and physical sciences. In both sciences you must decide which independent variables are o f interest. Often these are control variables like fertilizer, water, insec ticides or in our case, interest rates. In actual agricultural practice, insecticide and w ater are often applied in response to observations on the state o f the grow ing plants. Similarly in economics, it is often hard to tell if a control variable, the Aaa bond rate, fo r example, is a response to observations on the econom y or a driver o f them. Agronomists perform greenhouse experiments in which they fertilize plants in amounts long and short o f the perceived optimum to map out a response curve fo r yield as a function o f fe r tilizer. In contrast, economists are reluctant to experim ent by (knowingly) setting control varia bles at nonoptimal values. It is w ell known in agriculture that greenhouse results often do not translate directly to the field, so agronomists, like econometricians, distinguish micro from macro environments. Biologists also typically know the lag relationships, if any, involved in their experiments. Yield in August may be related to fertilizer application in June, but when do w e finish harvesting the effects o f a bank closure or a tax increase? Biological organisms in the field and the econom y respond to a great number o f inputs and a big decision is which to put into the model and which to leave as part o f the erro r term. An aspect o f model choice that Christ does not particularly stress is the choice o f model form . This is sometimes chosen to fit the data at hand w ell and so can be part o f a data mining operation. Many physical models, as w ell as econometric MARCH/APRIL 1993 96 models, are not linear. Einstein’s famous E = MC2 is an example. In economics, the well-known MV = PY can be made linear by taking logarithms. Such transformations have implications fo r variance on the original scale—a point I think is not often appreciated. If log(M) = log(P) + log(Y) - log(V) + e with e normal, then MV = PY[exp(e)] and therefore the error is multiplicative, causing heterogeneity o f variance in the untransformed data. Further, nonlinear models like MV = PY pose problems of aggregation. For example, suppose such a rela tionship holds in all segments o f an economy. W ill it then hold in the aggregate? Not necessarily. T o illustrate, note that (20M2) = (4)( 10) and (12)(6) = (18)(4). H ow ever, if w e average 20 and 12, average 2 and 6, average 4 and 18, and average 10 and 4, w e find that (16)(4) = 64, but that (H )(7 ) = 77. So apart from any estimation errors, even exact relation ships can hold on some scales but not on others. Despite all these potential problems, people have an inherent tendency to observe their environm ent and draw inferences. Th ere seems to be an optimism that with enough information w e can solve any o f our problems, regardless o f w hether they are economic problems, medical problems or other kinds o f problems. Attempts at problem solving w ill certainly persist, and analysis and criticism o f these attempts, such as Christ's, are w orthw hile activities. In fact, I think one o f his main points is that w e are all statisti cians, observing our w orld and m odifying our models based on the data. This may be done w ith or without numerical calculation. Model selection is influenced by our previous observa tions in a w ay that is hard to quantify. I found the Mitchell quote from 1927 somewhat offensive. The idea that with enough calculations, any tw o series can be found correlated at 90 percent surely cannot be true o f inform ed and careful statisticians and econometricians. N ever theless, I can agree w ith the nature, if not the extent, o f the problem. I can imagine someone noticing how a black cat had crossed his path on several occasions before a misfortune, thus giving birth to a superstition. Surely, how ever, the past must be somewhat like the future. Living in North Carolina, fo r FEDERAL RESERVE BANK OF ST. LOUIS example, I do not carry earthquake insurance, but I might if I lived in San Francisco. I suspect that early mankind anticipated being cold in w inter even without a good understanding o f m eteorology. I do not think w e should dismiss modeling as a w hole based on Lucas-critique types o f considerations. Christ gives an example o f a simple model that seems to have held up over a fairly long period. This is good news and I would go further to suggest that w e not give up on statistical modeling even if w e can’t get quite such good results every time. Along these lines, I agree with Christ that ARIM A and VAR are not as inform ative as a good econom etric model, but they may do less damage to our understanding o f the econom y than a m ediocre econom etric model. As a technical person, I feel obliged to address at least one or tw o technical points. I note that in the paper, some time was spent trying to decide w hether a quadrupling RMSE would b^reasonable in a good forecasting model. W hen w e look at the theoretical forecast erro r variances, w e can argue that this variance could not increase by more than a factor o f eight in going from a one-stepahead to an eight-step-ahead forecast. To com pensate fo r the difference betw een estimated and theoretical MSEs, it is then concluded that if the estimated error mean square goes up by a factor o f 16 (RMSE up by a factor o f four) our model would be suspect. The probability o f this quadrupling o f sample RMSE w ould depend on the autocorrelation and the number o f fo re casts used to estimate RMSE; fo r example, if w e just look at a single one-step-ahead residual and a single eight-step-ahead residual, the estimated RMSEs w ill simply be the ratio o f the absolute errors and hence w ill vary a lot around the true values. Suppose MSE is calculated by averaging the squares o f k independent one-step-ahead errors e(n + l ) and the squares o f the k corresponding eight-step-ahead errors z(n + 8) = e(n + 8) + r e(n + 7) + r2 e(n + 6) + .... + r 7 e(n + 1) from an AR(1) w ith autoregressive param eter r. I estimated the probability that the sum o f k values z(n + 8)~ is m ore than 16 times the sum o f the k corresponding values e(n + l ) by a Monte Carlo experiment with 10,000 replicates at each r and k. Figure 1 summarizes the results w ith r = 0.5, 0.6, 0.7, 0.8 and 0.9. The num ber o f forecasts 97 Figure 1 Probability of Quadrupling (versus number of squares in RMSE and first order autocorrelations) from which RMSE is calculated is k = l, 2, 3 or 4. It is seen that, because o f the variation in RMSE around its theoretical expected value, the probability o f the eight-step-ahead RMSE exceed ing four times the one-step-ahead RMSE can be reasonably large (greater than 0.2 in the case that k = 1) even with a perfect model and relatively mild autocorrelation. As k gets large, and hence as RMSE converges to the theoretical value dis cussed in the paper, the probability declines. Figure 1 shows the empirical frequency o f RMSE quadrupling. As another m inor technical point, I w ould like to say that a lot o f new ideas are the same old vanilla ones with a fe w sprinkles throw n on. In his figure 3, Christ plots the inverse velocity against the inverse Aaa bond rate data with connecting lines indicating the time order o f the data and w ith the regression line overlaid. W e MARCH/APRIL 1993 98 FEDERAL RESERVE BANK OF ST. LOUIS 99 MARCH/APRIL 1993 100 could think o f this as the end view o f a threedimensional picture, which w e rotate and tilt a bit in figures 2 and 3. The line is seen as the end o f a three-dimensional plane. The data wander pretty far up and dow n and right and left but never get too far from the plane. Projections into the wall and floor o f the plot show the tw o nonstationary looking series, also depicted in Christ’s figure 2. The tightness o f the data about the plane shows that a linear combination o f the tw o series looks fairly stationary. This is the idea o f cointegration. Regression is one w ay o f finding cointegration in FEDERAL RESERVE BANK OF ST. LOUIS bivariate series. Other methods may give slightly different planes, but w e can see that the main ideas o f this currently popular econometric method are quite close to simpler time-tested ones. In closing, I think w e are at an exciting time fo r econometrics. Some o f the computational burdens have been lifted, and w e can concentrate m ore on proper m odel form s and form ulation methods. Philosophical guidance such as that offered by Christ is important to keep in mind in our search. 101 D a v i d L a id le r David Laidler is a professor of economics at the University of Western Ontario, London, Canada. Commentary fA R L CHRIST’S PAPER is a w orth y tribute to Ted Balbach. It is broad ranging, thoughtful and provocative; and it deals with serious issues too. M oreover, no small matter fo r this discus sant, it is readily accessible to the stochastically challenged. The best compliment I can pay it is to add a fe w reflections o f my ow n on the questions it raises. It must now be at least 25 years since I first heard Carl Christ discuss the importance o f test ing models against data that had not been used to build them. Even then he distinguished be tween data generated before and after not just the m odel’s estimation period, but also the actu al time at which the model was constructed. This last distinction is not often made, but Carl convinced me that it is m ore important than w e might think. I am glad he still stresses it. The simple fact is that what w e know about eco nomic history influences how w e build our models in ways that w e barely recognize. Sup pose w e decided today to build a model o f the U.S. business cycle, to estimate it fo r the period 1948-70, and then to test it further against data fo r 1971-92. W hen w e constructed our model, w ould w e be able to ignore the tw o oil price shocks during the 1970s, and would w e even be right to ignore them if w e could? But if w e did rem em ber the activities o f the Organization o f Petroleum Exporting Countries, would it really be the case that the structure fitted to the data fo r 1948-70 would yield parameter estimates unaffected by any influence from data generated after 1970? It is at least safer, and more convincing too, if w e test our models against really new data—data o f which w e w ere unaware at the time those models w ere constructed. 1 must confess, though, that the first time I heard Carl Christ make this point, I was discomfited by his argument. In the 1960s I was estimating demand-for-money func tions, and I did not much like the idea o f waiting another decade or so before submitting my results to a journal. The right scientific approach was all w ell and good in its place it seemed to me, but there w ere more mundane matters to consider—promotion and tenure, fo r example. But here w e are 25 years later, and the back is sues o f economics journals are full o f empirical studies, which w ere influential in their time but are now half forgotten, whose results could be subjected to real tests. How would the Jorgenson investment equation or the Andersen-Jordan equation stand up?1 There is a market niche here waiting to be filled by applied econom etri cians, not least those currently w orryin g about the above-mentioned publishing criteria fo r pro motion and tenure. In his paper, Christ has shown us how to do such w ork with his investigations o f what he calls the plain-vanilla velocity equation, first pro posed by Henry Latane’ in 1954. This rather odd equation has held up surprisingly well. The use o f the inverse o f the rate o f interest as an argu ment surely (as Robert Rasche has suggested to me) reflects Latane’ 's reluctance to use logarithms to deal with a nonlinear relationship in an age when such a transformation o f data had to be carried out using tables and much tedious interpolation therefrom . In the light of Carl's results I am relieved to be able to report that, even before reading his paper, I had decid- 1See Jorgenson (1963) and Andersen and Jordan (1968). MARCH/APRIL 1993 102 ed to retain the paragraph dealing with L a ta n e’s study in the new edition o f Demand f o r Money.2 From a certain view'point, the survival o f the Latane" equation for a full three and a half de cades is remarkable. It is, after all, best inter preted as a rearrangem ent o f a supply-anddemand-for-money system, and as Carl also tells us, the last tw o decades have not been kind to empirical demand-for-money functions. But at least one precedent in the literature occurred, namely Robert E. Lucas Jr.’s demonstration that Allan Meltzer's long-run demand-for-money function also seems alive and w ell when viewed in light o f m ore recent data.3 N ow w e must not claim too much here, and Carl does not. The Latane" equation displays many faults calculated to shock the econom etric purist—fo r example, auto-correlated residuals. W hen these are attended to w ithin sample, the out-of-sample perform ance o f the m ore sophisti cated formulation seems to deteriorate. Similar ly, Lucas showed that though subsequent data still seemed to move around M eltzer’s relation ship, they did so w ith a great deal o f complex serial correlation. But still, I think there is a les son to be learned here, one which I began to develop in the second (1977) edition o f Demand f o r Money and w hich w ork using co-integration techniques is now tending to support. The les son is this: what w e call the long-run demandfor-m oney function is indeed a stable structural relationship, give or take ongoing institutional change, which w e often deal with by adapting our w ay o f measuring money. What w e call the short-run function, how ever, is not structural at all. It is rather an ill-understood, quasi-reduced form characterizing the mutual dynamic interac tion o f the money supply and the variables on which the demand fo r money depends in the long run. This w ay o f looking at things helps explain w hy co-integration studies produce evidence consistent with the existence o f a stable longrun demand-for-money function and w hy simple regressions o f the type estimated by Latane’ and M eltzer hold up rather well. As David Dickey has told us here, simple regression is one w ay o f looking fo r co-integration. It also helps ex plain w hy the error correction mechanisms as sociated with co-integration relationships are complicated and unstable, w hy the dynamics of 2See Laidler (1993). 3See Lucas (1988) and Meltzer (1963). FEDERAL RESERVE BANK OF ST. LOUIS so-called short-run demand-for-money functions have tended to break down as sample periods are extended, and w h y more sophisticated esti mation techniques, designed to cope with autocorrelated residuals, applied to relationships like the Latane equation produce results that are less robust over time than the plain-vanilla ver sion. Have w e not, after all, known all along that changes in the money supply affect the econom y with long and variable time lags, which, among other things, involve feedbacks to the money supply itself? And if w e have known that all along, should w e be surprised if w e get now here with studies o f m onetary dynamics that do not begin by specifying a model o f the aforem entioned interaction that will perm it us to identify the structural parameters o f the sys tem w e are investigating? It is all much easier said than done, o f course, but it w ill not be done if no one tries, and I hope therefore that Carl Christ’s striking results fo r Latane’s equation w ill prompt someone to carry his investigation further. REFERENCES Andersen, Leonall C., and Jordan, Jerry L. "Monetary and Fiscal Actions: A Test of Their Relative Importance in Eco nomic Stabilization,” this Review (November 1968), pp. 11-24. Jorgenson, D.W. “ Capital Theory and Investment Behavior,” American Economic Review (May 1963), pp. 247-59. Laidler, David. The Demand for Money—Theories, Evidence and Problems, 4th ed. (Harper-Collins, 1993). Latane, Henry. “ Cash Balances and the Interest Rate— a Pragmatic Approach,” Review of Economics and Statistics (November 1954), pp. 456-60. Lucas, Robert E. Jr. “ Money Demand in the United States: A Quantitative Review,” Money, Cycles and Exchange Rates: Essays in Honor of Allan H. Meltzer (Carnegie-Rochester Conference Series on Public Policy (Autumn 1988), pp. 137-67. Meltzer, Allan H. “ The Demand for Money—the Evidence From the Time Series,” Journal of Political Economy (June 1963), pp. 219-46. 103 Allan H. M e ltz e r Allan H. Meltzer is a professor of political economy and public policy at Carnegie Mellon University and a visiting scholar at the American Enterprise Institute. Thanks to Craig Hakkio, Carl Christ and Bennett McCallum who commented on an earlier draft and to Jeffrey Liang who contributed more than the usual assistance. A preliminary version was presented at the 1992 Western Economic Association session honoring Milton Friedman on his 80th birthday. Real Exchange Rates: Some Evidence f rom the Postwar Years HE MOVE TO FLEXIBLE EXCHANGE RATES early in 1973 is the type o f experim ent that economic researchers experience rarely. A marked change in m onetary regim e from fixed to flexible rates was follow ed by years o f floating rates. Initially, some governm ents may have thought o f flexible rates as a tem porary expedient to last only until new parities w ere firm ly established. Within a few years, how ever, the governm ents o f prin cipal developed countries, including the United States, accepted flexible rates as a durable arrange ment. Although there has been considerable inter vention in the currency markets, attempts at policy coordination and talk about target zones (particularly in recent years), the dollar and sev eral other currencies have continued to float. Most major trading countries have reduced or rem oved exchange controls and other restric tions on capital mobility. A frequent, and probably the dominant, assess ment o f experience with flexible rates is that they have not w orked as anticipated. Robert Aliber (1992, p. 44) w rites that "Few o f the advantages noted by proponents o f floating exchange rates have been realized in the 1970s and the 1980s.” Krugman and M iller (1992, p. 1) share this view and, in addition, criticize theories o f exchange rate determination. They write that “ interventionist economists believed that left to themselves exchange markets w ould introduce unnecessary and harmful volatility into the exchange rate.” These w riters summarize the current state o f research as showing that m onetary models "have had almost no empirical success. Indeed, money supplies, if they enter at all, typically enter w ith the w ro n g sign.” (ibid, p. 9) Singleton (1987, p. 9) reports the professional judgment that "b y most measures, exchange rates have been relatively unstable since 1973.” He re cognizes, how ever, that the instability may reflect uncertainty that the public faces in adjusting to information about the future. And he notes that observed variability o f exchange rates may have low er w elfa re costs than alternative regimes. Mussa (1986) studied fluctuations in bilateral exchange rates fo r the principal market econo mies. He showed that the variability o f bilateral real exchange rates from 1957 to 1984 was eight to 80 times higher in flexible-rate periods. There w ere no examples o f low er variability under flex ible rates among the 17 countries studied. The reason is clear from Mussa's data. Under flexi ble exchange rates the variability o f nominal exchange rates increases much m ore than the variability o f the ratio o f relative price levels declines. In fact, the variance o f bilateral rela MARCH/APRIL 1993 104 tive price levels was not always low er in flexiblerate regimes. Mussa did not draw any conclusion about the w elfare properties o f alternative regimes. The increased variance o f bilateral real exchange rates may substitute fo r the variance o f other variables, may be absorbed at relatively low cost by hedgers and speculators in financial markets, or in part may represent permanent shocks, such as the oil shocks o f the 1970s and 1980s, that require adjustment o f relative prices and real values. But the alternative is also plau sible. Some o f the higher variances under fluc tuating rates may be the source o f excess burden. A main problem in reaching a judgment about the operation o f fluctuating rates is that there is no benchmark fo r comparing alternative regimes. Economic models o f exchange rates have per form ed poorly compared with statistical models such as the random walk. Many papers report that there is no significant relation, often no evidence o f any reliable relation, betw een exchange rates and other economic variables. Meese and Rogoff’s (1983) well-known paper found that a random walk perform ed as w ell out o f sample as any estimated structural model. This suggests that many changes in exchange rates are random events, unrelated to policy or macroeconomic perform ance. Chinn (1991) summarizes recent tests fo r cointegration o f real and nominal ex change rates w ith standard economic aggregates such as money and output at home and abroad or, fo r nominal exchange rates, relative rates o f inflation. The tests reject cointegration, suggest ing that there is no long run relationship between exchange rates and any o f these aggregates. Critics have commented especially on the rela tively large change in dollar exchange rates in the 1980s. Even Haberler (1987), a long-time proponent o f floating refers to “ the widespread disenchantment with floating exchange rates.” Critics have not been satisfied with computa tions showing that the variances o f exchange rates, like the prices o f other traded assets, exceed the variances o f prices o f current pro duction. Nor have they accepted as sufficient explanation fo r observed variability that foreign exchange markets, like other markets fo r traded 'See Frenkel (1983). 2The qualification is needed because some testable proposi tions result. Changes may be unbiased or larger in periods of large shocks such as wars and oil price changes. 3See Wallich (1984). FEDERAL RESERVE BANK OF ST. LOUIS assets, respond to new information, which arrives continuously in a changing w orld .1 Without evidence showing that the news is systemati cally linked to exchange rate changes and that the adjustments are tow ard a new equilibrium, the proposition is nearly em pty.2 A longer summary o f the large literature on flexible rates w ould belabor the obvious. Neither the critics nor the proponents o f flexible exchange rates have produced much evidence on which to base comparative judgments about exchange rate regimes. Claims that variability is larger or too large are meaningless unless an alternative is specified and its properties compared. Yet it is common to find statements that flexible rates have not w orked as expected. They "do not sub stantially shield a country from events abroad”; that “ current account imbalances have been pro tracted”, and that "w ide movement and reversals have contributed to the widespread impression that floating rates tend to overshoot.” "Although clean floating has not yet becom e a dirty word, the simple faith that the market is always right has been shaken.”3 This paper reconsiders experience under flexible exchange rates. Section 1 summarizes the claims about flexible rates in Milton Friedman's classic 1953 paper to show that Friedman’s claims are m ore modest than is often supposed. Section 2 presents some key facts about exchange rates and comparative variability o f several variables under fixed and flexible rates. Section 3 esti mates a model o f the so-called real exchange rate under Bretton W oods and flexible rates and tests fo r the effect o f econom ic aggregates on the exchange rate. The m odel incorporates some o f the principal variables affecting exchange rates suggested by Friedman. Section 4 discusses some limitations o f the results. A conclusion completes the paper. FR IE D M AN ’S CASE FOR FLEXIBLE EXCHANGE RATES In "T h e Case fo r Flexible Exchange Rates,” w ritten shortly after the Bretton W oods System started, Friedman claims four benefits fo r flexi ble rates: (1) increased liberalization o f trade, (2) avoidance o f direct controls, (3) facilitation o f 105 rearmament, and (4) harmonization o f internal monetary and fiscal policies.4 The point that concerned later critics most, variability or insta bility, is dismissed early with the claim that exchange rate instability reflects instability in the econom y and is not a property o f a flexible or floating rate system. This claim is not selfevident, and it has not been accepted by the principal critics o f flexible rates. Friedman appears to have anticipated this outcome. He devotes m ore space to refuting or dismissing the charge o f instability than to making the positive case fo r the four benefits claimed for flexible rates. Friedman’s essay does not claim that flexible exchange rates are optimal fo r all countries or even fo r a single country. W hen discussing the form er sterling bloc, he considers a mixed sys tem in which groups o f countries may elect to maintain fixed exchange rates internally and flexible rates against all other groups or coun tries. Although there are structural differences betw een the sterling bloc and the proposed European M onetary Union, Friedman anticipates the principal issues: policy harmonization, avoid ance o f trade controls and exchange restrictions, absence o f a political authority and, in the absence o f controls, the need to choose betw een unem ploym ent and exchange rate changes in the short term. by the critics o f flexible rates. The critics typi cally argue that speculation is (or can be) destabilizing. Friedman considers and rejects some common conjectures about destabilizing speculation. His main argument is that there is no empirical foundation fo r these claims. Appearances to the contrary are misleading and subject to misin terpretation. A main problem in any study is to separate the actions o f speculators based on correct predictions o f parity changes and actions that cause parity changes that would have been avoided. These problems arise under an adjustable peg, but Friedman claims they would be prevented under continuous adjustment o f flexible rates. Friedman is cautious, how ever. He avoids a gen eral claim that speculation is stabilizing. Instead, he argues that if destabilizing speculation is com mon, governments (or exchange stabilization funds) would profit by intervening. And he recognizes that governm ents may have m ore information or m ore timely information that gives them an advantage over private speculators. He is willing to let a governm ent agency intervene to smooth tem porary fluctuations if they can do so profita bly (p. 188), but he is skeptical that they would be able to profit consistently. They are less likely to profit, he claims, than private speculators who risk their ow n wealth. Recognizing that optimality o f flexible rates can not be established, Friedman limits his claim to the judgment that flexible exchange rates are more desirable socially than the four alternative means o f offsetting changes in international position. The fou r alternatives are: (1) official changes in currency reserves; (2) changes in domestic price levels and incomes; (3) periodic realignment o f parities; and (4) direct controls. The reason fo r choosing flexible rates is that other means o f adjustment are less satisfactory. Fixed exchange rates w ere maintained in the 19th century because the public and govern ments tolerated larger fluctuations in domestic prices and employment than would be accepta ble in the late 20th century. Direct controls are least satisfactory because they introduce distor tions and do not correct permanent differences in relative prices in foreign and domestic markets. The key conditions are posited. First, with flexi ble exchange rates, there are "broad, active, and nearly perfect markets ... in foreign exchange” w henever they are permitted. Second, a fixed but adjustable exchange rate “ insures a maxi mum o f destabilizing speculation. Because the exchange rate is changed infrequently and only to meet substantial difficulties, a change tends to come w ell after the onset o f difficulty, to be postponed as long as possible.”5 These condi tions, it seems fair to say, have not been accepted Tim ing o f adjustments is a source o f variabil ity about which little is known with precision. Anticipating future discussion, Friedman con siders overshooting and undershooting o f ex change rates. Overshooting arises because initial adjustment is borne by prices that adjust most readily. The exchange rate is such a price. Later other prices adjust, and the overshooting reverses, although it may be replaced by undershooting o f the final change, follow ed by a series o f ad justments around the new equilibrium. 4The essay was written in 1950 but not published until 1953. 5See Friedman (1953), pp. 162-64. MARCH/APRIL 1993 106 Thus Friedman recognizes that there will be variability and fluctuations o f exchange rates, not prompt, rapid adjustment from the old to the new equilibrium. The possibility that the fluctuations, though not destabilizing, produce excess burden and w elfare loss is not addressed directly. Friedman’s main response to this central issue is comparative. His conclusion can be sum m arized in tw o paragraphs. First, comparison o f exchange rate regimes must include the costs o f adjustment under alternative policies. The comparison cannot be limited to the size o f changes in exchange rates or the variability o f exchange rates under dif ferent regimes. Changes in the relative prices o f goods and services are not the same under dif ferent policies. W ith gradual adjustment o f real wages and other relative prices, labor market adjustment, hence unemployment rates, w ill dif fe r under different regimes. And direct controls introduce distortions and w elfare losses. Second, there is no presumption that social costs could not be increased by flexible exchange rates. “ About all one can say ... is that there seems no reason to expect the timing or pace of adjustment under the assumed conditions [flexi ble exchange rates] to be systematically biased in one direction or the other from the optimum or to expect that other techniques o f adaptation through internal price changes, direct controls, and the use o f m onetary reserves with rigid ex change rates—w ould lead to a m ore nearly opti mum pace and timing o f adjustment.”6 EXCHANGE RATE CHANGES AN D V A R IA B IL IT Y , 1973-90 Excessive variability is one o f the main issues raised by the critics o f flexible rates. Evidence o f increased variability o f real or nominal ex 6See Friedman (1953). The conflicts in the system devel oped more slowly than Friedman predicted. He predicted that “ direct controls over exports and imports would be reimposed on a large scale within two or three years at the most.” This prediction was inaccurate. The United States introduced some controls on capital movements in the 1960s, but the trend in the 1950s and 1960s was toward reduction of trade barriers under General Agree ment on Tariffs and Trade rules. The conflicts in the sys tem were resolved partly by changes in parities abroad but mainly by inflation in the 1960s and early 1970s. 7The so-called real exchange rate measures the ratio of the price level in the United States to a weighted average of foreign price levels expressed in a common currency. FEDERAL RESERVE BANK OF ST. LOUIS change rates after 1973 is easy to produce. To draw any conclusion about the effects on w el fare, tw o issues must be resolved. First, as Friedman noted, increased variability o f exchange rates may reduce variability o f output, con sumption, em ployment or other variables o f interest to consumers. Reduced variability o f these variables can produce a w elfa re gain despite the increased variability o f exchange rates. Second, increased variability o f exchange rates may result from real shocks, such as an oil shock, or from policy activism, or it may reflect increased knowledge o f the operation o f exchange markets. This section considers changes and variability o f exchange rates and some other variables under Bretton W oods and flexible rates. Figure 1 shows the monthly trade-weighted nominal and real ex change rate fo r the United States, using Federal Reserve weights, fo r the period 1973-90. A rise in the index is an appreciation o f the dollar. T w o facts are immediately apparent. First, real and nominal exchange rates m ove together and by similar amounts.7 This fact has been demonstrated repeatedly fo r bilateral rates. See Mussa (1986) and Edwards (1989) fo r studies o f developed and developing countries. Second, trade-weighted exchange rates moved over a relatively w ide range during the 18-year period. The movement is dominated by a persistent appreciation from 1980 to 1985 follow ed by a persistent deprecia tion lasting to early 1987. Both exchange rates then returned to approximately the same range they had left in 1979. Other measures o f trade-weighted exchange rates developed by the International M onetary Fund (IMF) using wholesale prices or unit labor costs in the various countries to compute real exchange rates show the same general pattern. Experiments with different w eighing patterns 107 Figure 1 Trade-Weighted Nominal and Real Exchange Rates Index (1973:3=100) Monthly 1973:1 to 1990:12 160150140130120110 100- 9080- do not appear to change the general features, although computed variances and ranges d iffer fo r the individual measures.8 The exchange rate data shown in figure 1 raise tw o issues that w ill concern us. First, w hy do real and nominal exchange rates move together? Second, is the higher variability o f real exchange rates under fluctuating exchange rates caused by policy actions, or is there evidence o f excess burden arising from increased variability unrelated to policy action? sumer price indexes fo r the tw o countries. In the first years, the real and nominal exchange rates differ; consumer prices rose m ore rapidly in Japan than in Germany. In real terms Japan paid m ore yen per mark than in nominal terms. A fter 1976, the tw o price levels had about the same rate o f change, so the real and nominal exchange rates are often indistinguishable on figure 2. The similarity o f real and nominal exchange rate changes in figure 1 is not peculiar to U.S. data. Figure 2 shows monthly values o f the ex change rate o f the Japanese yen fo r the German mark during the period 1973-90. The real ex change rate is obtained using the relative con Mussa’s (1986) study o f changes in bilateral exchange rates fo r a broad sample o f developed countries during the years 1957-1984 found the same result. Under flexible exchange rates, changes in nominal and real exchange rates are highly correlated, but changes in nominal or real ex change rates are not closely correlated with changes in the ratio o f price index numbers. 8Becketti and Hakkio (1989) computed the correlation between innovations in seven alternative measures of trade-weighted exchange rates. Most of the correlations are above 0.9 using quarterly data for 1976 to 1988. They show that similar results hold for percentage rates of change of exchange rates. the start of each decade-1960, 1970 and 1980-to adjust for changes in relative trade shares. The main conclusion sensitive to the change in weights is that the variance of the trade-weighted real exchange rate is lower for the alternative measure. I have used the Federal Reserve index throughout. The Federal Reserve index uses weights reflecting country shares of world trade. I computed an alternative index based on U.S. trade weights and reweighted the index at MARCH/APRIL 1993 108 Figure 2 Real and Nominal Yen/DM Exchange Rates Yen/DM Monthly 1973:1 to 1990:12 150140- 120 - 110 - 100 - 60 I------ 1-----1----- 1-----1----- 1----- I------1----- 1----- 1-----1..................................................... - - - - - - - . 1973 74 75 76 77 78 79 80 81 M eltzer (1990) considered the variability of multilateral exchange rates using data from the IMF. Real exchange rates are based on both relative wholesale prices and relative unit labor costs, and variances are used to measure varia bility. Again, countries with flexible exchange rates had greater variability o f nominal and real exchange rates than countries in the European Monetary System (EMS) that maintained an ad justable peg with other members of the EMS. Changes in real and nominal exchange rates w ere highly correlated under flexible rates. H ow ever, the variability o f relative unit labor costs was typically low er in the countries with flexi ble exchange rates, whereas the variability of wholesale price ratios was higher. Table 1 summarizes these data. Both nominal (N) and real (R) exchange rate changes are more variable under flexible exchange rates than under fixed but adjustable rates, whereas relative prices are not. The variability o f R or N under flexible rates is significantly different at the 1 percent level from the variability experienced under EMS or the mixed regimes (denoted other) that had crawling pegs or some other type o f par tially fixed nominal exchange rate during this FEDERAL RESERVE BANK OF ST. LOUIS 82 83 84 85 86 87 88 89 90 1991 period. Changes in multilateral real exchange rates are 4 or 5 times m ore variable in flexiblerate countries than in the EMS. Generally, the variances fo r “ other” countries lie betw een the variances fo r the EMS and flexible-rate coun tries. The exception is IJ]L —the variability of C changes in relative prices based on unit labor costs, IJ C has been low er on average under ,L flexible rates, although the difference betw een regimes is not significant by the usual standards. The much-discussed increase in the variability o f real exchange rates in a flexible exchange rate regim e may reflect only that flexible exchange rates change m ore frequently, whereas the rela tive price ratios are not much affected by the change in regime. Using the terms o f trade as a measure o f relative prices, table 2 shows that the variances o f relative price ratios do not dif fe r systematically across exchange rate regimes. Variability o f the terms o f trade rose in all coun tries but to different degrees unrelated to the exchange rate regime. The comparatively high variability o f Japan’s terms o f trade suggests that there is no simple relation betw een the variability o f this measure and the grow th of trade. 109 Table 1 Variances of Changes in Relative Prices, Real and Nominal Exchange Rates 1/1979-111/1989__________________________ Average Quarterly Values x 100 Flexible Rates2 EMS1 Pwp R ulc RWp N Other3 .007 .037 .154 .158 .143 .013 .017 .037 .033 .025 P ULC .017 .030 .058 .059 .043 ’ Austria plus seven EMS countries (Belgium, Denmark, France, Germany, Ireland, Italy, Netherlands) 2Japan, Switzerland, United Kingdom and the United States 3Norway, Spain, Sweden, Finland and Canada Source: IMF where N s Ft + P NOTE: P R N is the first difference of the logarithm of the relative price of domestic to foreign goods or services. is the first difference in the logarithm of the real exchange rate is the first difference in the logarithm of the nominal exchange rate Table 2 Variances under Fixed and Flexible Rates United States, Germany, Japan and the United Kingdom (quarterly values at annual rates) Real GNP or GDP Period 1/1960— 111/1971 1/1973-111/1991 Relative value 1/1973-11/1975 111/1975-11/1980 111/1980— 111/1987 IV/1987-111/1991 U.S. 11.0 15.7 1.4 24.2 19.7 13.3 5.2* Germany 27.9 8.5 0.3 10.6* 7.1* 6.6* 3.5* Japan 32.3 11.3 0.3 37.4 5.1* 5.8* 7.8* TOT U.K. 30.0 32.6 1.1 70.5 59.8 10.6* 11.0* U.S. 15.5 92.2 5.9 202.3 60.2 61.7 71.0 Germany 26.0 92.5 3.6 285.2 52.4 78.2 24.8* Japan 23.5 390.8 16.6 237.7 431.1 388.8 170.8 U.K. 25.9 62.6 2.4 197.3 40.8 31.6 27.6 TOT is terms of trade; variances are squared deviations from x, = (Xt - X t _ 1)/Xt _ 1 NOTE: * denotes that the variance is lower than under Bretton Woods. Table 2 also compares real output variances under fixed and flexible exchange rates in four countries. Th ere is no relation betw een the rela tive variances and the m onetary system. Real ouput variability declined in the same prop or tion in Germany and Japan with (mainly) fixed and flexible rates respectively and rose m oder ately in the United States and the United Kingdom.9 The last fou r lines o f the table show variances fo r subperiods. The oil shocks o f the 1970s in creased the variances in table 2 in the early years o f flexible rates. Variability o f output fell in the United States in each successive period. In all countries the variance o f real GDP was low er in 1987-91 than under the Bretton W oods regime. 9Meltzer (1986) reports similar results for the four countries using unanticipated variances. Unanticipated variances were computed using forecasts obtained from a multistate, univariate Kalman filter. MARCH/APRIL 1993 110 The countries shown in table 2 have different exchange rate systems. Japan and the United Kingdom had flexible exchange rates during the period, although the United Kingdom fixed to the exchange rate mechanism (ERM) o f the EMS at the end o f the period. Germany has been in the fixed-but-adjustable-rate ERM system since March 1979, and it experim ented with other fixed-but-adjustable-rate systems with its neigh bors beginning in the mid-1970s. The mark fluc tuated, how ever, against the dollar, yen and many other currencies. Though the variability o f Germany’s output grow th is, on average, lowest o f the countries in table 2, this cannot be attributed entirely to the reliance on fixed-but-adjustable rates. Varia bility o f output grow th in Germany was also low er than in Japan or the United Kingdom during the Bretton W oods period, and the rela tive decline in variability is the same fo r Ger many and Japan. Further, during 1975-80 and 1980-87, periods o f declining inflation, variabil ity o f Japan’s output grow th is comparable to (and even slightly below ) Germany’s. The main conclusion drawn from table 2 is that there is no basis fo r a general proposition that output is more variable under fixed rates than under flexible rates. Relative prices (terms o f trade) are m ore variable in all countries after 1973, but the increase is smallest in the United Kingdom. POLICIES A N D REAL EXCHANGE RATES Friedman (1953) made tw o suggestions that have been overlooked. He gave prom inence to policy—particularly rearm am ent—as one o f the main factors affecting U.S. real exchange rates. Rearmament changes relative prices and the bal ance o f payments (Friedman, pp. 159-60). Also, Friedman distinguished permanent and transi tory changes in exchange rates. He noted the different response o f speculators to changes that w ere expected to reverse and those that w ere expected to persist.1 0 Real governm ent spending fo r defense rose and fell during the postwar years. Spending rose during the Vietnam W a r and declined dur,0See Friedman (1953, p. 162). I began work on the relation of permanent and transitory fiscal and monetary changes to real exchange rates before I reread Friedman’s essay. I was pleased to find that the results I had obtained provided evidence on some of his principal propositions. FEDERAL RESERVE BANK OF ST. LOUIS ing the 1970s both absolutely and relative to real output. Spending rose again in the 1980s, reached a peak in the mid-1980s and declined modestly to the end o f the decade. Maintained changes in the level o f real defense spending act like any fiscal change. Increases in real defense spending raise aggregate spending and interest rates. Higher interest rates attract a capital inflow, appreciating the exchange rate. In the absence o f capital controls and restric tions, the capital inflow reverses the rise in the interest rate. Reductions in real defense spend ing have the opposite effects.1 The sign o f real 1 defense spending per unit o f output should be positive. Real money balances also affect real exchange rates. Injections o f money tem porarily increase real balances, and if the price level does not adjust instantly, the increase in m oney depreci ates the real exchange rate. Reductions in real balances brought about by reductions in m oney or by a rise in prices fo r a given quantity o f money appreciate the exchange rate. Let r, the real exchange rate, have a perm a nent and transitory component, so that (1) r, = r, + ut w h ere r, is the permanent component and ui is the transitory disturbance. In the absence o f changes in defense spending, real U.S. money balances and foreign real balances, the expected value o f the exchange rate is the permanent value. The current permanent value is a weighted average o f last period's exchange rate and any persistent effect o f defense spending (relative to GDP) and real money balances at home and abroad as shown in equation (2). (2) rl = ar[ , + (1 - a) f (d {,m {,m ^ + v t, Combining equations (1) and (2) gives equation (3), a testable equation fo r the real exchange rate. (3) r, = art t + (l - a) w h ere + £ , has the usual properties. I f the real exchange rate is mainly a random walk, r, = r, j plus a transitory w hite noise term. "D efense spending is a large share of government spend ing on goods and services. It has the advantage of being independent of income and hence a good measure of the thrust of exogenous fiscal policy. It also permits a test of Friedman’s proposition. 111 Table 3 Determinants of the Real Exchange Rate* (t statistics in parentheses)______________ Periods RER,_1 m. mf dt constant R2/DW p(AR1) (1) 1962-91 Annual 0.80 (6.40) -0 .1 5 (4.45) 0.28 (2.55) 5.32 (2.92) 53.64 (3.26) 0.89/1.89 0.09(0.36) (2) 1972-91 Annual 0.72 (3.69) -0 .1 6 (4.49) 0.26 (1.72) 9.75 (1.92) 44.59 (2.45) 0.78/2.03 -0.11(0.26) (3) 11/1971 — IV/1991 Quarterly 0.82 (11.76) -0 .1 1 (2.60) 0.07 (0.52) 3.37 (1.62) 20.87 (2.42) 0.92/1.95 0.26(1.75) (4) 1962-91 Annual 0.67 (5.37) -0 .1 3 (3.87) 0.21 (1.77) 5.88 (2.82) 62.31 (3.16) 0.88/1.92 0.14(0.50) (5) 1972-91 Annual 0.55 (3.12) -0 .1 6 (4.43) 0.16 (1.06) 13.02 (2.45) 52.53 (2.63) 0.77/2.00 -0.10(0.23) *See appendix for definition of variables. p(AR1) is the coefficient of the AR1 serial correlation correction and its t-statistic. But if m onetary and fiscal actions have persist ent effects, these effects w ill be found signifi cant in estimates o f equation (3). Equation (3) therefore permits a test o f the influence o f the defense spending share and real money bal ances against the alternative hypothesis that real exchange rates are approximately a random walk and independent o f systematic m onetary and fiscal effects. If the real exchange rate is mainly a random walk, a is close to one. If there are persistent and systematic effects o f money and the defense spending share, current values o f these variables will have a significant effect on the real exchange rate. The first tw o columns o f table 3 show estimates fo r 1962-91 and 1972-91 based on annual data; the form er includes the fixed exchange rate period, whereas the latter does not. The tw o sets o f estimates are similar. The standard errors o f estimate fo r the tw o equations are 5.9 and 6.8, a difference o f approximately 1 percent o f the mean value o f the real exchange rate. The implied standard erro r o f estimate fo r the Bretton W oods period is 3.6, about half the value fo r the flexi ble rate period. These values suggest that transi tory random variation increased under flexible rates, but the increase is much smaller than is commonly alleged. A main reason is that the estimates here rem ove the effects o f permanent changes in m, m*, and d. These variables, par ticularly real money balances, have significant effects on the trade-weighted real exchange rate. One problem with these estimates is that the coefficient o f m* is much larger than the co effi cient o f mt using annual data. The difference may not be meaningful, how ever. The definitions o f money d iffer (as described in the Appendix), and the difference in coefficients is not significant. Figure 3 shows the actual and predicted val ues o f table 3 using equation (1). Many o f the claims about exchange rate instability are based on the relative changes in the 1980s. The chart suggests that much o f the swing in the tradeweighted real exchange rate during the 1980s is driven by the variables in the model. The defense spending share rose by m ore than a percentage point in the early 1980s then fell after the mid dle o f the decade. Real money balances moved in the opposite direction, falling through 1982, then rising, particularly in 1985 and 1986. The forecasts and actual values are extrem ely close fo r 1981-83. Th ere is some evidence o f o v er shooting by the actual rate in 1984-85, but the errors are not much larger than the standard error o f estimate. The subsequent decline in the forecast value lags the actual decline, however, in 1986 and 1987. Th e largest error in the 29-year span is in 1986. The third column in table 3 shifts the time interval from annual to quarterly data. The results are similar to the annual data except that mt* is no longer significant. Current real money balances remain significant at the usual level, and the defense spending share nearly so. MARCH/APRIL 1993 112 Figure 3 Trade Weighted Real Exchange Rate Index (1973:3=100) The dependent variable in the regressions reported in the first and third columns o f fig ure 3 is the average trade-weighted real exchange fo r the period. The fourth and fifth columns repeat the regressions fo r annual data using the monthly average value fo r Decem ber as the de pendent variable. The results are similar. Th e estimates in table 3 permit a test o f the unit coefficient on RER( , implied by the randomwalk hypothesis. All o f the estimates are below unity, but tw o are not significantly different from unity; these are in the first and second columns o f table 3. The estimates in the third and fifth columns d iffer from unity by m ore than tw o standard errors, so they reject this central implication o f the random walk. Recent w ork on the causes o f fluctuations em phasizes the importance o f real shocks to aggre gate supply as a cause o f fluctuations. The effects on the real exchange rate o f the rise and fall o f the relative price o f oil in the 1970s and 1980s is an obvious candidate fo r investigation. The relative price o f oil can be included in equation (2) as an additional variable affecting the perma nent component o f the real exchange rate. Annual data fo r 1972-90 and 1962-90 reject the effect; FEDERAL RESERVE BANK OF ST. LOUIS the coefficient o f the relative oil price is small (-0 .0 3 ) in each period and has a standard error larger than the estimated coefficient. The use o f real m oney balances combines the separate effects o f money and prices. To sepa rate the effect o f policy actions from the effects o f prices, I first differentiate m, = (M/p) then lag the denominator by one period to get dM, dp, P.-1 P, (4) dm, The first term is the real value (in past prices) o f the current change in nominal balances. The second is the revenue from the inflation tax on last period’s real money balances. T o estimate responses to these variables, I take first d iffe r ences o f equation (3) using equation (4) to re place dmr Table 4 shows estimates relating the annual change in the real exchange rate to changes in some policy variables and real shocks. I have omitted the change in m* to conserve a degree of freedom. Am* typically has a small negative coefficient and is not significant. Changes in money and changes in defense spending relative 113 Table 4 Response of ARER to Changes in Policy Periods ARERt _ 1 AM,/p, _ i Ad, Inflation tax A real debt (1) 1972-90 Annual 0.60 (3.05) -0 .3 9 (4.42) 23.91 (2.62) -0.01 (0.07) -0 .0 0 5 (0.93) A gov’t net worth* A RGDP constant R2/DW P(AR1) 0.07 (2.39) 13.95 (1.23) 0.59/1.79 -0.19(0.43) (2) 1972-89 Annual 0.72 (6.66) -0 .4 7 (8.41) 13.86 (3.60) -0.001 (0.00) 0.004 (0.30) 0.03 (1.81) 19.32 (2.47) 0.80/2.51 -0.68(2.51) t-statistics are in parentheses. Coefficients of ARER»_-| are significantly different from unity at the 2a level. 'T h e data are from Bohn (1992). These data are available only through 1989. For other data, see appendix. to GDP have considerable effect. For example a 0.1 percentage point change in the share o f defense spending changes the real exchange rate betw een 1.4 percentage points and 2.4 per centage points based on the tw o equations. The 1982 increase in defense spending alone appre ciated the dollar by 8.7 percentage points using the coefficient estimate fo r 1972-89.1 2 The inflation tax is not significant in the re gression or in alternative estimates. This is un satisfactory. Without a significant response to inflation, the equations imply that a change in nominal money has a permanent effect on the real exchange rate. If the equations are inter preted as short-term responses, they leave an important part o f the dynamics unspecified. Much recent discussion o f the appreciation o f the real exchange rate in the early 1980s, fo l low ing the Reagan tax cuts, linked the apprecia tion either to the budget deficit or to the increased after-tax return to real capital. The change in the real value o f governm ent debt measures the part o f the current federal budget deficit financed by borrow ing. I used the change in real GDP (RGDP) as a measure o f the real return to real capital. This variable also captures the effects o f changes in real output emphasized in the busi ness cycle literature. Because real output is close to a random walk, changes in RGDP are a meas ure o f unanticipated changes. The change in RGDP has a significant effect on the change in the real exchange rate. The size o f the coefficient is misleading because the changes are in billions o f dollars. A m ore sug gestive comparison is given by the change in the real exchange rate induced by changes in RGDP and the defense spending rates during four years o f appreciation— 1981-84. The total appreciation o f the real exchange rate fo r this period is 44. The coefficients in the first column o f table 4 assign slightly less than half o f this change to the change in the defense spending ratio and slightly m ore than half to the change in RGDP. These calculations neglect other varia bles, particularly changes in money and lags o f the real exchange rate. And the calculation overstates the importance o f supply shocks or changes in tax rates because the changes in RGDP include the recovery from the 1981-82 recession that would have occurred in the absence o f tax changes or supply shocks. The response to deficit finance, measured by the change in real governm ent debt, is small and insignificant. A problem with testing fo r effects o f the budget deficit is the incomplete and imprecise w ay in which the deficit is meas ured. Eisner and Pieper (1984) called attention to this problem and showed that there are large differences betw een current accounting meas ures and measures o f a m ore economically rele vant magnitude. Bohn (1992) computed a measure o f governm ent net w orth that includes principal government assets and liabilities other than Social Security and Medicare liabilities. The second column substitutes the change in real govern ment net w orth from Bohn fo r the change in the real value o f the federal debt as a measure o f the deficit. Governm ent’s net w orth is nega tive, and if properly measured, the level o f governm ent net w orth is the value o f future tax payments. Changes in net w orth have no signifi cant effect. The responses to changes in RGDP and changes in the defense spending share both fall. Each explains a smaller fraction o f the 12l neglect possible changes in the properties of the error term when taking first differences of equation (3). MARCH/APRIL 1993 114 Figure 4 Annual Changes in Real Exchange Rates change in the real exchange rate during 1981 to 1984 (and other periods). The implied change in the real exchange rate resulting from changes in RGDP and the defense spending share are now approximately 25 percent and 29 percent respectively. Figure 4 shows predicted and actual changes in the real exchange rate based on the estimates in the second column o f table 4. Inspection sug gests that the equation explains the annual changes more accurately fo r the 1980s than fo r the 1970s. This is particularly true in 1974 and 1975. There are only three years in which actual and p re dicted changes go in opposite directions—1975, 1978 and 1983. Actual and predicted changes move together during the appreciation and sub sequent depreciation o f the dollar in the 1980s. The equation suggests that contemporaneous changes in money and in defense spending are the principal factors keeping the predicted changes in step with actual changes. LIM ITATIO N S The empirical results are subject to some limi tations. This section briefly discusses some pro blems arising from the absence o f a structural FEDERAL RESERVE BANK OF ST. LOUIS model, neglect o f simultaneity, and problems o f stationarity. First, the estimates are obtained from a simple model o f permanent and transitory changes, not from a structural model. The equations are nei ther structural equations nor reduced form s o f a structural model. An advantage o f the model is that it nests the effects o f money and defense spending within a popular statistical model, the random walk. Second, several o f the variables such as the price level, output, the real value o f m oney and the defense spending share are simultaneously determined. Simultaneity has been neglected throughout. Th e changes reported in table 4 and the use o f lagged prices rem oves some o f these problems. That the principal results are unaffected suggests that simultaneity may not impart serious bias to the estimates in table 3. Third, many studies o f exchange rates have investigated the stationarity o f exchange rates. Tests o f non-stationarity at first seemed to sup port the hypothesis. M ore recent w ork using longer time series, how ever, casts doubt on this conclusion. Engel and Hamilton (1990) did not test fo r stationarity, but they found persistent 115 departures from a random walk. Earlier, Krasker (1980) coined the term peso problem fo r persist ent deviations o f exchange rates in a particular direction. Papers by Huizinga (1987), Hakkio and Joines (1990), Lothian (1991), and Diebold, Husted, and Rush (1991) are part o f the grow in g liter ature rejecting non-stationarity based on evidence that real exchange rates return to a mean value. A main reason fo r the differences in findings betw een earlier and later studies is the use o f a longer span o f years. Some early studies used daily or monthly data to obtain a larger number o f observations. Recent studies suggest that an increased number o f high-frequency observa tions is a poor substitute fo r the relative paucity o f low -frequency data.1 3 The principal conclusion to draw from many o f the studies is that the real exchange rate is subject to persistent and transitory changes. Some changes in the real exchange rate persist fo r long periods. Some o f the changes are re versed quickly. Diebold, Husted, and Rush (1991) conclude that on average the half-life o f a shock to the real exchange rate has been about three years. This finding is similar to the decay rates implied by the coefficients on annual values of the lagged real exchange rate in table 3. Inspection o f figure 1 suggests that the mul tilateral real exchange rate remained within a range o f 95 ± 15 from 1973 to 1980 and returned to approximately the same range in 1987. T o test fo r stationarity, I used quarterly data fo r first quarter 1973 to fourth quarter 1990 but omitted the sharp appreciation and depreciation from third quarter 1980 to first quarter 1987.1 The coefficient o f the lagged 4 multilateral real exchange rate on the change in the real exchange rate is -0.14 with a t-statistic o f 2.72. The Dickey-Fuller test statistic is 2.93 at the 5 percent level and 2.60 at the 10 percent level. On this basis, I reject non-stationarity. CONCLUSION Milton Friedman’s (1953) essay on flexible exchange rates anticipated much o f the discus sion and many o f the controversies o f the next 40 years. Friedman did not claim that flexible exchange rates would be stable rates. Stability depends on the size and frequency o f shocks. Friedman claimed that flexible exchange rates w ould (1) contribute to trade liberalization, (2) avoid reliance on direct controls, (3) facilitate rearmament and (4) allow countries to follow domestic policies to achieve price stability. Several o f these conjectures w ere correct. Direct controls on capital movements have been reduced since 1973 in all developed countries and in some developing countries. It seems likely that rearmament (defense spending) w ould have provoked greater conflict about payments imbalances in the 1980s under fixed exchange rates than under the system that prevailed. Flexible rates permitted countries to choose how much o f the stimulus emanating from the United States they wished to absorb. Many countries, indeed most developed coun tries, both purchased dollar securities and appreciated their currency. The average rate o f inflation has been brought dow n under flexible rates, and some countries have achieved price stability at times. Trade restrictions, how ever, increased in the 1980s, particularly in the United States, and the movement toward trade liberalization slowed. Friedman did not argue that exchange rates would be stable. He argued that the path fo l low ed by real exchange rates w ould depend on the real and m onetary disturbances to which the econom y is subject and on the persistence o f shocks. Critics argued that destabilizing specula tion and random movements dominate exchange rate changes and create an excess burden. This burden, some suggested, could be reduced by fixing exchange rates or establishing target zones. The paper does not address the issue o f excess burden. H ow ever, I compare variability o f output and the terms o f trade fo r fou r coun tries under the Bretton W oods System and the different regimes adopted after 1973. There is no evidence that real output is generally more variable under flexible exchange rates. Term s of trade are m ore variable after 1973, but the data do not suggest that the increased variability is mainly the result o f the exchange rate regime. Further, I compare levels and changes in real exchange rates to the values predicted by a model. 13Hakkio and Rush (1991) reach the same conclusion based on more formal tests. 14The hypothesis implies and the data suggest that the ap preciation and depreciation in this period is mainly the result of policy action. MARCH/APRIL 1993 116 The forecast errors do not give evidence o f large, persistent errors. On the contrary, the models call the turning points in the level and changes in the exchange rate with considerable accuracy. Th e data suggest, how ever, that there is m ore unexplained variability o f real exchange rates after 1973 than before when measured by the standard error o f estimate fo r the regres sion equation. The evidence also suggests that much o f the m ovem ent in both levels and changes in annual values o f the U.S. multilateral real exchange rate is explained by permanent or persistent changes in a fe w variables. The principal varia bles are real m oney balances and the share o f defense spending in GDP. W hen the change in real balances is separated into variables measur ing the current change in nominal money and the current change in the price level, the data suggest that the change in nominal money (measured at past prices) has a m ore important short-run effect. Quarterly data on levels o f the variables support the principal findings. M onetary and fiscal variables are nested w ith in a random walk model o f the real exchange rate. If the random-walk component dominated the exchange rate, the data would reject the re levance o f the m onetary and fiscal variables. The tests based on annual and quarterly data and on annual changes support the opposite conclusion: m onetary and fiscal effects are persistent and reliable, and their effect is contemporaneous— w ithin the current year or quarter. O f course, none o f the findings here deny that the random walk may dominate levels or changes o f the ex change rate at higher frequencies. T w o principal observations about fluctuating exchange rates during the past 20 years are: (1) the close relation betw een real and nominal exchange rates and (2) the sharp appreciation and subsequent depreciation o f the real dollar exchange rate in the 1980s. 1 conjecture that the principal reason fo r the correspondence betw een movements in real and nominal ex change rates is that real exchange rates are driven by contemporaneous permanent changes in real variables, particularly real defense spend ing and real m oney balances, whereas nominal exchange rates are driven by the nominal val ues o f the same variables. Much o f the short term effect o f money on the real exchange rate appears to be the result o f changes in nominal money, so it would not be surprising to find that changes in nominal m oney balances have a sig nificant effect on the nominal exchange rate also. FEDERAL RESERVE BANK OF ST. LOUIS D A T A A P PE N D IX Nom inal exchange rate (FNER): Index o f the trade-weighted foreign exchange value o f the United States dollar compiled by the Federal Reserve. The index is a geom etric average o f 10 industrialized countries’ dollar value o f their currencies w eighted by their shares o f w orld trade betw een the years 1972 and 1976. The 10 countries are Germany, Japan, France, United Kingdom, Canada, Italy, Netherlands, Belgium, Sweden and Switzerland. Trade-weighted price level (TW CPI): Geometric average o f 10 industrialized countries’ consumer price indexes w eighted by their shares o f total w orld trade. Real exchange rate: FNER deflated by the ratio o f the United States consumer price index (CPI) to the 10 countries’ trade-weighted CPI. Real money balances (m): United States M l m onetary aggregate deflated by the United States CPI. Defense spending share (d): Ratio o f the United States defense spending in current dollars to GDP in current dollars. Foreign money balances (T W N M ): Arithm etic average o f indexes o f M2 m onetary aggregates o f Canada, Germany, Great Britain, and Japan (M2 &, CD), w eighted by their shares o f total w orld trade betw een the years 1972 and 1976. Foreign real money balances: T W N M deflated by TW CPI. Real government net worth: Real governm ent deficit measured by the US governm ent real net w orth from Bohn (1992). Relative price o f oil: Oil price measured by com posite refiners' acquisition cost deflated by GNP deflator. Real federal debt: Gross federal debt net o f Fed eral Reserve holdings deflated by the CPI. REFERENCES Aliber, Robert Z. “ The Case for Flexible Exchange Rates Revisited,” University of Chicago (unpublished). Becketti, S., and Hakkio, C. “ How Real is the 'Real Exchange Rate’?” draft Federal Reserve Bank of Kansas City (April). Bohn, H. “ Budget Deficits and Government Accounting,” Carnegie Rochester Conference Series on Public Policy (December 1992), pp. 1-83. Chinn, M. “ Some Linear and Nonlinear Thoughts on Exchange Rates,” Journal of International Money and Finance (June 1991), pp. 214-30. Diebold, F.X., Husted, S., and Rush, M. “ Real Exchange Rates under the Gold Standard,” Journal of Political Econ omy (December 1991), pp. 1252-71. 117 Edwards, S. Real Exchange Rates, Devaluation, and Adjust ment: Exchange Rate Policy in Developing Countries (MIT Press, 1989). Krasker, W. “ The ‘Peso Problem’ in Testing the Efficiency of Forward Exchange Markets,” Journal of Monetary Eco nomics (April 1980), pp. 269-76. Eisner, R. and Pieper, P. “ A New View of the Federal Debt and Budget Deficits,” American Economic Review (March 1984), pp. 11-29. Krugman, P., and Miller, M. “ Why Have a Target Zone?” Carnegie Rochester Conference Series on Public Policy (forthcoming). Engel, C. and Hamilton, J.D. “ Long Swings in the Dollar: Are They in the Data and Do Markets Know It?” American Economic Review (September 1990), pp. 689-713. Lothian, J. “ A History of Yen Exchange Rates,” in W.T. Ziemba, W. Bailey, and Y. Hamao, eds., Japanese Finan cial Market Research (Elsevier, 1991). Frenkel, J. “ Turbulence in the Foreign Exchange Markets and Macroeconomic Policies,” in L. Melamed, ed., The Merits of Flexible Exchange Rates (George Mason Univer sity Press, 1988), pp. 445-69. Meese, R. and Rogoff, K. “ Empirical Exchange Rate Models of the Seventies: Do They Fit Out of Sample?” Journal of International Economics (February 1983), pp. 3-24. Friedman, M. “ The Case for Flexible Exchange Rates,” in Friedman, M., ed., Essays in Positive Economics (Univer sity of Chicago Press, 1953). Meltzer, A.H. “ Size, Persistence and Interrelation of Nominal and Real Shocks,” Journal of Monetary Economics (Janu ary 1986), pp. 161-94. Haberler, G. “ The International Monetary System and Proposals for International Policy Coordination,” in Cagan, P., ed., Contemporary Economic Problems (American Enterprise Institute, 1987), pp. 63-96. _______ . “ Some Empirical Findings on Differences Between EMS and Non-EMS Regimes,” Cato Journal (Fall 1990), pp. 455-83. Hakkio, C. and Joines, D. “ Real and Nominal Exchange Rates Since 1919,” working paper, Federal Reserve Bank of Kansas City (September 1990). Mussa, M. “ Nominal Exchange Rate Regimes and the Behavior of Real Exchange Rates: Evidence and Implica tions,” Carnegie Rochester Conference Series on Public Policy (Autumn 1986), pp. 117-213. Hakkio, C. and Rush, M. “ Cointegration: How Short is the Long Run?” Journal of International Money and Finance (December 1991), pp. 571-81. Huizinga, J. “ An Empirical Investigation of the Long-run Behavior of Real Exchange Rates,” Carnegie Rochester Conference Series on Public Policy (Autumn 1987), pp. 149-214. Singleton, K. “ Speculation and the Volatility of Foreign Cur rency Exchange Rates,” Carnegie Rochester Conference Series on Public Policy (Spring 1987), pp. 9-56. Wallich, H. “ Floating as Seen from the Central Bank,” in L. Melamed, ed., The Merits of Flexible Exchange Rates (George Mason University Press, 1988), pp. 395-403. MARCH/APRIL 1993 118 Pedro Schwartz Pedro Schwartz is a professor in the history of economic thought at the Universidad Anto'noma de Madrid and executive vice president at NERA Madrid. Commentary A .LLA N MELTZER HAS PRESENTED a paper that rescues Milton Friedman's 1953 defense o f flexible exchange rates from its critics and p re sents an impressive amount o f empirical evidence against today’s received opinion favoring fixed exchanges. Nothing is m ore bracing than the refutation o f conventional wisdom nor m ore satisfying than the vindication o f the simple faith that the market is always right against the latest interventionist fads o f central bankers, w hich is what I take to be the practical conse quence o f the paper. Anything that could make Europeans think again about imposing a con trived m onetary union in the EC is most welcome. M y comments start w ith M eltzer’s summary o f Friedman’s case fo r floating rates. Then 1 make a quick evaluation o f the force o f the data presented in refutation o f those w ho have as serted that flexible exchange rates have not w orked as anticipated. I make this evaluation without in any w ay claiming, how ever, to have redone M eltzer’s calculations or amassed some different evidence. The reason fo r not focusing on the empirical part o f M eltzer’s paper, except as evidence o f how the w orld seems to func tion, becomes apparent in the follow ing section. A fter a simple-minded expose" o f what M eltzer’s results mean fo r the day-to-day business o f a central banker and a portfolio manager, I try to contrast the empirical relations given in M eltzer’s paper w ith the assumptions implicit in the project fo r a European M onetary Union. Finally, I reflect on the conditions that explain w hy m onetary zones appear and on w hether the benefits o f a freely floating independent curren cy becom e larger than the costs. 1See Friedman (1953). FEDERAL RESERVE BANK OF ST. LOUIS M eltzer underlines h ow Friedman refrained from claiming too much fo r floating exchange rates and also h ow prescient he was about problems that w ould plague fixed exchanges at the time w hen the Bretton W oods accord was being implemented. Friedman, as I have confirm ed by rereading his 1953 paper, did not present flexible exchange rates as optimal under all circumstances. He started by defining the new arrangem ent he wanted to criticize:1 the Western nations seem to be committed to a system of international payments based on ex change rates between their national currencies fixed by governments and maintained rigid ex cept for occasional changes to new levels. He then prudently said that “w hatever may have been the merits o f this [Bretton W oods] system fo r another day it is ill suited to current economic and political circumstances.” If the fixed exchange system must have had some merit w hen it was being used within currency zones—some as large as the one in which he lives—fo r Friedman, the flexible rates system had the superior political consequences, namely: (a) fostering the liberalization o f trade; (b) reducing the need fo r exchange controls; (c) easing the path fo r necessary extraordinary ex penditures, such as the rearmament that turned out to be necessary w ith the Cold W ar; and (d) alleviating the constant brushes betw een central bankers and treasuries over domestic m onetary and fiscal policies. 119 These indirect advantages are important and correspond contrastingly w ith the political ad vantages claimed by those w ho defend pegging exchanges o f countries or a group o f countries to some standard or to some m ore reliable cur rency. In this case the political advantage lies in putting a check on the oversupply o f money and forcing m onetary authorities to minimize their inflation tax and maintain the value o f money. An exchange fixed onto gold, some basket o f goods or another reliable currency such as the deutsche mark is a sort o f superin dependence clause in the central bank bylaws because it turns the Bank in effect into a cur rency board. In M eltzer's paper there is little discussion o f possible data related to these elements o f com parative advantages o f flexible vs. fixed exchange rates from the political and social point o f view. In his paper there is a useful explanation o f what Friedman really meant, but not an em piri cal evaluation o f the relative size o f the func tional relationship he posited. The elements o f a comparative analysis o f floating exchange rates vs. fixed are listed by Meltzer, but their quan titative importance is not evaluated. Recognizing that optimality of flexible rates can not be established, Friedman limits his claim to the judgment that flexible exchange rates are more desirable socially than the...alternative means of offsetting changes in international po sition... (1) official changes in currency reserves; (2) changes in domestic price levels and in comes; (3) periodic realignment of parities; and (4) direct controls. O f these, periodic realignments are the most im portant fo r a judgment on the functioning o f the European Monetary System (EMS). I found a rem ark in M eltzer’s conclusion that “ several o f ...[the four] conjectures w ere cor rect.” These conjectures are such things as the positive political effects on the liberalization of trade. Regarding Friedman's (b) avoidance o f direct controls, M eltzer says the following: Direct controls on capital movements have been reduced since 1973 in all developed countries and in some developing countries. W ith respect to Friedman’s (c), M eltzer notes the following: It seems likely that rearmament (defense spend ing) would have provoked greater conflict about payments imbalances in the 1980s under fixed exchange rates than under the system that prevailed. Finally, there is the follow ing indirect treatment o f the possible evidence on Friedman’s (d), the harmonization o f internal monetary and fiscal policies: Flexible rates permitted countries to choose how much of the stimulus emanating from the United States they wished to absorb. Many countries, indeed most developed countries, both purchased dollar securities and appreciat ed their currency. that is, they both absorbed and sterilized the stimulus. This is ve ry little on a large part o f the con troversy about fixed vs. flexible exchange rates, but it w ill have to be a topic fo r a different paper because M eltzer prefers to concentrate on a previous problem that Friedman dealt with implicitly, though at length. Says M eltzer: The point that concerned later critics most, variability or instability, is dismissed early with the claim that exchange rate instability reflects instability in the economy and is not a property of a flexible or floating rate system . . Friedman appears to have anticipated this outcome. He devotes more space to refuting or dismissing the charge of instability than to making the positive case for the four benefits claimed for flexible rates. This is also what M eltzer does, in the belief that the question o f overshooting and o f destabilizing speculation has to be resolved before the politi cal cost-benefit analysis o f flexible rates can be addressed. The evidence M eltzer does present bears on five points that w ould clearly be important as preconditions fo r evaluating Friedman’s main political theses: (1) w hether the variability o f ex change rates in the main OECD countries in creased after the breakdow n o f Bretton W oods— especially w hether the EMS currencies showed less variability; (2) w hether and over w hat period can exchange rates be considered to m ove along a random walk; (3) connected w ith 2., w hether money supply and governm ent spending policies do affect real and m oney exchange rates signifi cantly; (4) connected w ith 3., w h y real and m onetary rates seem to m ove together; and (5) w hether exchange rates are congenitally unsta ble, w hether they show a tendency to return to mean values and if so, over what period. The results presented by M eltzer are most valuable and should becom e standard w ith the profession if confirm ed by rerunning them fo r MARCH/APRIL 1993 120 different periods and countries, and especially fo r the present episode o f instability in the EMS. REAL CAUSES And the yearly and quarterly movements o f exchange rates are not random because V A R IA B IL IT Y A N D ITS POSSIBLE W ELFAR E EFFECTS The figures presented fo r 1973 through to 1990 m oved over a relatively w ide range. This behavior o f the exchanges could, as Friedman said, turn out to reduce the variability in some real phenomena, such as output or employment. The variability could also derive from m ore acute and frequent real shocks in this period. This is as may be, but M eltzer concentrates on w hether the variability is m erely apparent and is robust under different measures used. If the said variability could be traced to the m ore frequent movements in a flexible regim e but left relative prices unmoved, the visible variabili ty could have only small real consequences. See table 1: R, the real exchange rate, and N, the nominal rate, varied much less in the EMS countries than in other OECD countries. (One w ould in any case want to see the variance af ter what happened on September 26.) The rela tive price changes betw een countries, how ever, w ere unaffected by the exchange regime. changes in money and changes in defense spending relative to GDP have considerable ef fect. For example, a 0.1 percentage point change in the share of defense spending changes the real exchange rate between 1.4 percentage points and 2.4 percentage points. ... The 1982 increase in defense spending alone appreciated the dollar by 8.7 percentage points. Not only governm ent expenditures, but also increases in real GDP and changes in real money balances seem to have significant effects on the real rate o f exchange. W ithin the black box, w e could surmise that increases in expenditures w ill contribute to raising interest rates and at tracting foreign capital and that increases in GDP w ill also lead to higher rates through the same mechanism and through the increased de mand fo r money. On the other hand, a fall in real money supply w ill also push up the real ex change rate. Deficits, on the other hand, seem to have no significant effect o f the real (and money) rates. REAL A N D M ONEY EXCHANGE RATES At the end o f his paper, M eltzer has a section on the possible limitations o f his empirical find ings. One is possible simultaneity, which he cor rects by lagging and fo r which he finds no evidence o f relevance. Another is that the results are not derived from a structural model, but from a simple model o f permanent and transitory changes w hich is, by the way, a traditional Friedmanite approach. M eltzer maintains that the w hole empirical exercise “nests the effects o f money and defense spending within a popular statistical model, the random walk,” so signifi cant departure from the null hypothesis would precisely be a most striking refutation o f the random walk theory o f foreign exchanges. David Ricardo in 1817, in the passages o f his Principles w h ere he discussed the distribution o f precious metals in the world, under the gold standard, saw that advances in productivity in a country led first to a fall in costs and real prices, then to an accumulation o f reserves and an in crease in m oney prices, and finally fo r a time to an overvaluation o f the m oney exchange rate until domestic prices fell to an equilibrium. The process w ould be the inverse fo r a country fall ing back in productivity. Hence the tendency o f real and m oney exchanges to fluctuate constant ly in separate directions turned out to be characteristic o f a fixed exchange rate regime. (Of course, it is contrary to the rules o f a fixed interest regime, especially o f the gold standard fo r the bank to sterilize foreign funds.) I can summarize the results w ith a quotation: "transitory random variation increased under flexible rates, but the increase is much smaller than is commonly alleged”; and a m ove from an nual to quarterly data does not significantly change the results. Daily data would, how ever, probably show much m ore randomness. In contrast, it was therefore expected that, under a flexible exchange regim e, because the inflow o f money from increased productivity and exports does not go into reserves but into foreign exchanges, the index o f m oney prices would be governed much m ore directly by the prices o f tradeable goods. THE R A N D O M W A L K M ODEL FEDERAL RESERVE BANK OF ST. LOUIS 121 M eltzer gives an additional reason fo r the joint m ovem ent o f real and nominal rates. He has shown reductions in real balances to be a pow erful cause o f the increase o f real exchange rates: and real balances also govern m oney ex change rates. NON-ST A T IO N A R IT Y One last element in the description o f a flexi ble exchange rate regim e is the rejection o f non-stationarity by M eltzer. That there are ob servably persistent departures from a random walk led some authors to think that speculation could be permanently destabilizing. H ow ever, if one uses a longer span o f years it becomes clear that exchanges are subject to both persis tent and transitory changes. I take it that the persistent changes are responsible fo r the time illusion o f non-stationarity. If the average real life o f a shock is 3 years, the period fo r return to the mean rate o f exchanges can be long: but return to the mean they do (I should add, if there are no capital or trade controls). Though it may be subject to correction from further empirical research, the picture o f the exchange w orld given by M eltzer’s empirical research is striking, both fo r the central banker and the investor. I read these provisional con clusions w ith some trepidation, but hope to be corrected by the audience before I becom e a central bank govern or or a large investor. First, real and money exchanges m ove together. Price indexes w ill m ove up or dow n w ith the real exchange rate and w ill be governed by the real causes o f real exchange appreciation or depreciation. In an open economy, therefore, a central banker can aspire to a steady or zero in flation rate only as an average over a long period—perhaps a three-year half-life. Second, there is money to be made in foreign currency (at least until everybody starts reading M eltzer) because o f long-term systematic and predictable forces in the foreign currency m ar ket. Government expenditures, GDP grow th and reductions in real balances portend o f revalua tions to come (as long as people do not expect the Government to inflate the accumulated debt away). To put it in another way, a good long term investment in a country blessed w ith a central bank that does not panic can discount exchange fluctuations if it has enough capital or is not subject to quarterly scrutiny at the stock market. Third, the variability observed w hen exchanges float does not seem to spill over into the goods and services market because it does not affect relative prices. Hence the decision to float or to fix w ill have to be taken on sociopolitical grounds and cannot be settled on evidence o f persistent overshooting. N ow given all this, the arguments w ith which the m onetary part o f the Maastricht Treaty is being defended begin to sound less convincing. The follow ing pros and cons are usually presented. The reduction in transaction costs from hav ing to deal in a single currency is a benefit. Cecchini has calculated a once-and-for-all gain equivalent to 4 percent o f European GDP. This may be exaggerated and is much low er than the recurrent gain from the single market. Another benefit is the control o f the central bankers o f the constituent states by a European Central Bank (ECB) w ith the express duty to de fend the value o f the single currency. This has the follow ing tw o drawbacks, how ever. The first is the suspicion that the states’ central bankers do not want to reduce their sovereignty, but want rather to increase it w ith their seat on the ECB’s executive committee because the mar kets themselves have made state central banks lose much o f their freedom ; the other drawback is that the new ECB w ill have quite a task being independent and refraining from playing with the exchanges, as can be guessed by the pres sures recently put on the Bundesbank. Th e solution to the difficulties posed by a m onetary union among w idely differing coun tries is problematic. W e have already seen the points that Friedman foresaw w ould plague such a m onetary union: policy disharmonies and the possible pressure fo r a central governm ent (a sure cause o f friction in Europe); a temptation to impose exchange or capital and trade con trols; and indifference to implementing cushions to prevent unemployment in the less productive parts o f the union. The question is then, w hy do I sometimes ad vocate a currency board fo r small countries, which is a strict form o f m onetary union, and w h y are the m onetary unions made up o f rather large countries and sometimes o f a large econo m y such as Germany and its close surrounding trade partners? MARCH/APRIL 1993 122 Let us imagine a w orld o f competing monies that float against each other. Th eir market share w ill be decided as in any other oligopolistic in dustry, the producers obtaining a seignorage or markup over marginal cost, a markup limited by potential entry; and demand being fo r the well-known services that m oney provides. These services are: fo r transactions (of which a part is coinage fo r small change subject to the metal content being o f less than face value); fo r pricing goods, services and savings; and fo r holding a real cash balance. The picture that em erges from this is not only that o f a w orld divided horizontally in zones, but also a w orld subject to a division in layers, w here different currencies may be used for different purposes: fo r example, deutsche marks or dollars fo r trade, Swiss francs fo r pensions, and pesetas fo r local payments. Apparently it w ould be ideal fo r consumers o f money, especially those that need it to produce goods and services, if all dealings could be in only one currency. This in fact is not necessarily so, fo r all the reasons w e have noted in this commentary. The union could, how ever, be ap proximated by the market, and the study o f the non-stationarity hypothesis w ill confirm that over long periods currencies tend to stay around FEDERAL RESERVE BANK OF ST. LOUIS their historical rates. I f there w ere m onetary competition, I doubt that there w ould be more than three currencies circulating in the United States. The smaller nations around Germany that trade intensely w ith her and that have simi lar economic structures to her—fo r example, the Benelux nations, Austria and S w itzerla n d w ili find it in their interest to stick to the deutsche mark. Only competition w ill tell how big m onetary zones must be. Avoiding hyperinflation and enjoying the serv ices o f a currency that is reasonably stable fo r purposes o f valuing goods, services and savings may lead some people to ask that the issuer o f m oney be separated from the creator o f the budget deficit as they have recently been in the Baltic States and the Ukraine. In other places, such as Hong Kong, a currency board that pegs the local m oney to the dollar may inspire confi dence in a highly volatile situation. In questions o f currency w e live very much in a second-best world. REFERENCES Friedman, Milton. “ The Case for Flexible Exchange Rates,” in Friedman, Milton, ed., Essays in Positive Economics (University of Chicago Press, 1953). 123 M ich a el D. B o r d o Michael D. Bordo is a professor of economics at Rutgers University and a research associate of the National Bureau of Economic Research. For excellent research assistance I would like to thank Jakob Koenes. For helpful comments and suggestions I am grateful to Barry Eichengreen, Allan Meltzer, Leslie Presnell, Hugh Rockoff and Anna Schwartz. The Gold Standard, Bretton Woods and Other Monetary Regimes: A Historical Appraisal IN TR O D U CTIO N Tw o Questions W hich international m onetary regim e is best fo r economic performance? One based on fixed exchange rates, including the gold standard and its variants? Adjustable peg regimes such as the Bretton W oods system and the European M one tary System (EMS)? Or one based on floating exchange rates? This question has been debated since Nurkse's classic indictment o f flexible rates and Friedman’s classic defense.1 W hy have some m onetary regimes been more successful than others? Specifically, w hy did the classical gold standard last fo r almost a century (at least fo r Great Britain) and w hy did Bretton W oods endure fo r only 25 years (or less)? W hy was the EMS successful fo r only a fe w years? This paper attempts to answer these questions. To answer the first question, I examine em piri cal evidence on the perform ance o f three m one tary regimes: the classical gold standard, Bretton Woods, and the current float. As a backdrop, I examine the mixed regime interwar period. I answer the second question by linking regim e success to the presence o f credible commitment mechan isms, that is, to the incentive compatibility fea tures o f the regime. Successful fixed-rate regimes, in addition to being based on simple transparent rules, contained features that encouraged a center country to enforce the rules and other coun tries to comply. The Issues These questions touch on a number o f im por tant issues raised in economic literature. The first is the effect o f the exchange rate regime on welfare. The key advantage o f fixed exchange rates is that they reduce the transac tions costs o f exchange. The key disadvantage is that in a w orld o f w age and price stickiness the benefits o f reduced transactions costs may be outweighed by the costs o f m ore volatile output and employment. Helpman and Razin (1979), Helpman (1981) and others have raised the w elfare issue. This theoretical literature concludes that it is difficult to provide an unambiguous ranking o f exchange rate arrangements.2 M eltzer (1990) argues the need fo r empirical 1See Nurkse (1944) and Friedman (1953). 2See DeKock and Grilli (1989 and 1992). MARCH/APRIL 1993 124 measures o f the excess burdens associated with flexible and fixed exchange rates—the costs o f increased volatility on the one hand compared with the output costs o f sticky prices on the other hand. His comparison o f EMS and nonEMS countries in the postwar period, however, does not yield clear-cut results. Earlier literature comparing the m acroeco nomic perform ance o f the classical gold stand ard, Bretton W oods and the current float also yielded mixed results. Bordo (1981) and Cooper (1982) showed that the classical gold standard was associated w ith greater price level and real output volatility than post-W orld W ar II arrange ments fo r the United States and United Kingdom. On the other hand, Klein (1975) and Schwartz (1986) presented evidence that the gold stand ard provided greater long-term price stability than did the post-W orld W ar II arrangements.3 Bordo (1993) compared the means and stand ard deviations o f nine variables fo r the Group o f Seven countries under the three regimes, as w ell as the interw ar period.4 According to these measures, the Bretton W oods convertible period from 1959 to 1970 was the most stable regim e fo r the majority o f countries and variables examined. Eichengreen (1992a) measured volatil ity applying tw o filters (the first difference of logarithms and a linear trend).5 Comparing Bret ton W oods and the float fo r a sample o f 10 countries, he found no clear-cut connection betw een the volatility o f real grow th and the exchange rate regime. He also found no signifi cant difference in the correlation o f output vola tility across countries between the tw o regimes. A second issue is w hether the exchange rate regim e provides insulation from shocks and m onetary policy independence. Under fixed rates, coordinated m onetary policy may provide effective insulation from common supply shocks, but not from country-specific shocks. Under 3This result is disputed by Meltzer and Robinson (1989). 4The Group of Seven countries are Canada, France, Ger many, Italy, Japan, the United Kingdom and the United States. 5Eichengreen followed the methodology of Baxter and Stockman (1989). 6Similarly a monetary union such as the proposed Euro pean Monetary Union could provide effective insulation from common supply shocks for its members. However, giving up monetary independence imposes additional bur dens in the case of member-specific (regional) shocks, Feldstein (1992). 7Addressing the issue of the optimum currency area, Bay oumi and Eichengreen (1992b, 1992c and 1992d) also FEDERAL RESERVE BANK OF ST. LOUIS flexible rates, country-specific shocks can be o ff set by independent m onetary policy.6 The evidence on this issue is limited. Bayoumi and Eichengreen (1992a) applied the BlanchardQuah approach to show that both supply (per manent) and demand (tem porary) shocks, fo r a sample o f five countries, w ere considerably greater under the gold standard than under post-W orld W a r II regimes. H ow ever, they found little dif ference in the incidence o f shocks betw een Bretton W oods and the floating exchange rate regime. Th eir results also showed that the dis persion o f shocks across countries was higher under the gold standard than under the tw o m ore recent regimes and slightly higher under Bretton W oods than under the floating exchange rate regime. They attributed the ability o f the gold standard to withstand greater shocks to evi dence o f a faster speed o f adjustment o f both prices and output, as measured by impulse response functions.7 A third issue is the case fo r rules vs. discre tion. A fixed exchange rate may be view ed as a commitment mechanism or rule. It binds the hands o f policymakers to prevent them from follow ing inflationary discretionary policies.8 Th e m onetary authority, in a closed econom y or under flexible rates, might be tempted to engi neer an inflation surprise to raise revenue.9 The outcome is higher inflation because the public, assuming rational expectations, w ill anticipate the policy. W e re some credible mechanism, such as a m onetary rule, in place the expansion ary policy w ould not be implemented. Alterna tively, a commitment to a fixed exchange rate through a pledge to maintain gold convertibility, fo r example, could achieve the same results, but because it is more transparent, it w ould possibly cost less.1 Such binding commitments may, h ow 0 ever, be undesirable in the presence o f extrem e emergencies such as major wars, supply shocks or financial crises.1 Under such circumstances 1 apply this methodology to examine the incidence of shocks within Western Europe and within regions of North America. 8See Kydland and Prescott (1977), Barro and Gordon (1983), and Persson and Tabellini (1990). A lternatively the monetary authority may create an inflation surprise to offset a labor market distortion that raises the unemployment rate above some desired level. ,0See Giavazzi and Pagano (1988). "S e e Rogoff (1985a) and Fischer (1990). 125 a contingent rule, or one with escape clauses that allow m em ber countries to suspend parity (convertibility) tem porarily, may be optimal.1 2 The rule constrains the governm ent to adhere to the fixed exchange rate except in the case o f a wellunderstood em ergency, when it can suspend parity (convertibility under the gold standard) and issue fiat money. Once the em ergency has passed, with allowance fo r a suitable delay, the authority is expected to return to the rule—that is, to the fixed rate at the original parity. I f the public believes in the governm ent's commitment to return to the rule, the governm ent w ill be able to raise m ore revenue than it could with no credi bility. The inflation rate during the em ergency w ould be higher than under the rule (when p re sumably it would be zero) but less than in the case o f pure discretion. The pattern o f alternat ing fixed and floating exchange rate regimes over the past 200 years may be w ell explained by adherence to a rule with an escape clause.1 3 On the other hand, in a regim e o f floating exchange rates the inflationary bias o f discre tionary policy may be overcom e by instituting credible m onetary rules or other commitment mechanisms, such as an independent conserva tive central bank.1 Such mechanisms may pre 4 vent the perceived disadvantage o f sacrificing national sovereignty to the supernational dic tates o f a fixed exchange rate. A fourth issue is that o f international coopera tion and policy coordination. Recent game theory literature has demonstrated that coordination o f policies (by fixing exchange rates) can offset spill over effects from uncoordinated policy actions.1 5 Cooperative fixed exchange rate arrangements, how ever, unless enforced by a supernational au thority whose pow er exceeds national sovereignty, tend to break down as individual members de value. Cooperation is m ore likely, without a supernational authority, in a w orld o f repeated games because the benefits o f reputation can offset the advantages to each country o f cheat ing.1 But even in this case, cooperation betw een 6 nations may produce an inflationary bias when no credible commitment mechanism is present to prevent governm ents from follow ing discre 12See Bordo and Kydland (1992), Flood and Isard (1989a and 1989b), and DeKock and Grilli (1989 and 1992). 13See DeKock and Grilli (1989) and Giovannini (1992). 14See Rogoff (1985a). tionary policies.1 Thus fo r an international m one 7 tary arrangement to be effective both betw een countries and within them, a consistent credible com mitment mechanism is required. Such a mechanism likely prevailed under the gold standard but was less evident under Bretton Woods. A fifth and final issue is the case fo r international m onetary reform . Several, prominent proposals have been made to reform the present managed floating exchange rate regim e and move it back tow ard one o f greater fixity. These proposals in part derive from a perception, based on the historical record, that fixed exchange rates are preferable to the current float. These proposals include McKinnon's case fo r a gold standard w ith out gold, Mundell’s proposal to target the real price o f gold and the case fo r target exchange rate zones presented by Williamson and Bergsten.1 Even m ore immediate is the m ove to con 8 vert the adjustable peg o f the EMS to a unified currency area with irrevocably fixed exchange rates. O verview The paper accomplishes a number o f tasks. The next section answers the first question, which international monetary regim e is best fo r economic perform ance, by presenting a compila tion o f statistical evidence on different aspects o f the perform ance o f alternative m onetary regimes. The measures cover the stability o f several macroeconomic variables; the dispersion o f macroeconomic variables across countries; the persistence o f inflation; forecast errors in inflation and growth; the incidence o f supply (permanent) and demand (temporary) shocks; the dispersion o f shocks betw een countries; and the mean response o f prices and output to sup ply and demand shocks. The third section stresses the importance o f adhering to credible rules in a historical examination o f three inter national m onetary regimes: the classical gold standard, Bretton W oods and the EMS. The final section answers the question w hy some regimes endured longer than others. It con cludes by discussing w hy even a regional exchange rate arrangem ent—the EMS—has con siderable difficulty surviving. 16See Dominguez (1993). 17See Rogoff (1985b). ,aSee McKinnon (1988), Mundell (1992), Williamson (1985a) and Bergsten (1992). 15See Hamada (1979) and Canzoneri and Henderson (1988 and 1991). MARCH/APRIL 1993 126 The statistical evidence on perform ance o f alter native monetary regimes in the next section makes it clear that the performance o f regimes in the postW orld W ar II era is superior to that o f the regimes in the preceding half century. The key exception is the classical gold standard, which exhibits the lowest inflation persistence and a relatively high degree o f financial market integration. The Bretton W oods convertible regim e from 1959 to 1970 per form ed the best by far on virtually all criteria. The greater durability o f the gold standard com pared w ith Bretton W oods cannot be explained by a low er incidence o f shocks. The key expla nation fo r its success lies with the credibility o f the commitment to the gold standard rule of convertibility by England and the other core countries and its near universal acceptance. As a contingent rule, it was flexible enough to w ith stand the major shocks that buffeted it. The Bretton W oods adjustable peg was in some re spects similar to the gold standard contingent rule, but it invited speculative attack hence weakening the escape clause. Unlike England, the leading country before W orld W ar I, the United States, the dominant country under Bret ton Woods, maintained a credible commitment to a noninflationary policy fo r only a fe w years. The world, faced with im ported inflation in the late 1960s, lost the incentive to follow its leader ship, and the system collapsed in 1971. Th e longevity o f general floating exchange rate regimes since 1973 suggests that the lessons o f Bretton W oods have been w ell learned. Coun tries are willing to subject their domestic policy autonomy neither to that o f another country whose commitment they cannot be sure o f in a stochastic world, nor to a supernational m onetary authority they cannot control. Even the recent experience o f the EMS—a regional exchange rate arrangement betw een countries supposedly pur suing common goals—revealed differing national priorities in the face o f asymmetric shocks that placed intolerable strains on the system. 19l also examined the period (1946-73) which includes the three years of transition from the Bretton Woods adjusta ble peg to the present floating regime. The results are similar to those of the 1946-70 period. 20The common world price level under the gold standard, however, exhibited secular periods of deflation and infla tion reflecting shocks to the demand for and supply of gold. See Bordo (1981) and Rockoff (1984). A welldesigned monetary rule, it is argued, could have prevented the long-run swings that characterized the price level under the gold standard. See Cagan (1984). 21See Giovannani (1992). 22See Bordo and Schwartz (1989a). FEDERAL RESERVE BANK OF ST. LOUIS THE PERFO RM ANCE OF A L T E R N ATIV E M O NETAR Y REGIMES T o make the case fo r one m onetary regim e over another, empirical and historical evidence on their perform ance is crucial. In this section I present some evidence on different aspects o f the macroeconomic perform ance o f alternative international m onetary regimes over the past 110 years. The comparison fo r the seven largest (non-Communist) industrialized countries (the Group o f Seven countries) is based on annual data fo r the classical gold standard (1881-1913), the interw ar period (1919-39), Bretton W oods (1946-70), and the present floating exchange rates regim e (1971-89). The Bretton W oods period (1946-70) is divided into tw o subperiods: the preconvertible phase (1946-58) and the con vertible phase (1959-70).1 9 The comparison relates to the theoretical issues raised by the debate over fixed vs. flexible ex change rates. According to the traditional view, adherence to a (commodity-based) fixed exchange rate regime, such as the gold standard, ensured long-run price stability fo r the w orld as a w hole because the fixed price o f gold provided a nomi nal anchor to the w orld money supply. Individ ual nations, by pegging their currencies to gold, fixed their price levels to that o f the w orld .2 A 0 fixed-rate system based on fiat money, how ever, may not provide a stable nominal anchor unless a credible commitment mechanism constrains the grow th o f the w orld ’s money supply.2 The 1 disadvantage o f fixed rates is that individual nations are exposed to both monetary and real shocks transmitted from the rest o f the w orld through the balance o f payments and other channels o f transmission.2 The advantage o f 2 floating exchange rates is to provide insulation from foreign shocks. The disadvantage is the absence o f the discipline o f the fixed-exchangerate rule because m onetary authorities might adopt inflationary policies. 127 Theoretical developments in recent years have complicated the simple distinction betw een fixed and floating exchange rates. In the presence o f capital mobility, currency substitution, policy reactions and policy interdependence, floating rates no longer necessarily provide insulation from either real or m onetary shocks.2 M oreover, 3 according to recent real business cycle approaches, there may be no relationship betw een the inter national monetary regim e and the transmission o f real shocks.2 Nevertheless, the comparison 4 betw een regimes may shed light on these issues. One important caveat is that the historical regimes presented here do not represent clear examples o f fixed and floating exchange rate regimes. The interw ar period comprises three regimes: a general floating rate system from 1919 to 1925, the gold exchange standard from 1926 to 1931 and a managed float to 1939.2 5 Th e Bretton W oods regim e cannot be character ized as a fixed exchange rate regim e throughout its history: The preconvertibility period was close to the adjustable peg envisioned by its architects, and the convertible period was close to a de fa cto fixed dollar standard.2 Finally, although the 6 period since 1973 has been characterized as a floating exchange rate regime, at various times it has experienced varying degrees o f management. Stability and Convergence Table 1 presents descriptive statistics on nine macroeconomic variables fo r each Group of Seven country, with the data fo r each variable converted to a continuous annual series from 1880 to 1989. The nine variables are the rate o f inflation; real per capita growth; money growth; short-term nominal interest rates; long-term nominal interest rates; short-term real interest rates; long-term real interest rates; and the absolute rates o f change o f nominal and real exchange rates. The definition o f the variable 23See Bordo and Schwartz (1989a). 24See Baxter and Stockman (1989). 25To be more exact, the United States stayed on the gold standard until 1933, and France stayed until 1936. For a detailed comparison of the performances of these three regimes in the interwar period, see Eichengreen (1991a). 26Within the sample of seven countries, Canada floated from 1950 to 1961. used, fo r example, M l vs. M2, was dictated by the availability o f data over the entire period. For each variable and each country I present tw o summary statistics: the mean and standard deviation. For the countries taken as a group, I show tw o summary statistics: the grand mean and a simple measure o f convergence defined as the mean o f the absolute differences betw een each country's summary statistic and the grand means o f the group o f countries.2 I comment 7 on the statistical results fo r each variable. In fla tio n . Countries using the classical gold standard had the lowest rate o f inflation and displayed mild deflation during the interw ar period. The rate o f inflation during the Bretton W oods period was on average and fo r every country except Japan low er than during the subsequent floating exchange rate period. The average rate o f inflation in the tw o Bretton W oods subperiods was virtually the same. This comparison, how ever, conceals the importance o f tw o periods o f rapid inflation in the 1940s and 1950s and in the late 1960s. See figure l . 2 8 Thus the evidence based on country and period averages o f very low inflation in the gold stand ard period and o f a low er inflation rate during Bretton W oods than the subsequent floating period is consistent with the traditional view on price behavior under fixed (commodity-based) and flexible exchange rates. In addition, the inflation rates show the highest degree o f convergence betw een countries during the classical gold standard and to a lesser extent during the Bretton W oods convertible subperiod compared with the floating rate period and the mixed interw ar regime. This evidence also is consistent with the traditional view o f the oper ation o f the classical price specie flo w mechanism and commodity arbitrage under fixed rates and insulation and greater m onetary independence under floating rates.2 9 series around the G-7 aggregate. I calculated this alterna tive measure of convergence for the data in table 1. The results are very close to those reported here for virtually every variable. 2aThe data sources for figure 1 and all subsequent figures are listed in the Data Appendix to Bordo (1993). 29For similar evidence see Bordo (1981), Darby, Lothian et.al. (1983) and Darby and Lothian (1989). 27This is a very crude measure of convergence or diver gence between the different countries’ summary statistics. Because it is based on the average for the whole period, it suppresses unusual movements within particular sub periods. Bayoumi and Eichengreen (1992d) presented an alternative measure of convergence or dispersion—the GDP-weighted standard deviation of the individual country MARCH/APRIL 1993 FEDERAL RESERVE Table 1 Descriptive Statistics of Selected Open Economy Macroeconomic Variables, the Group of Seven Countries 1881-1989 Annual Data: Mean, Standard Deviation BANK O ST. LOUIS F Gold Standard (1881-1913) Interwar (1919-1938) Bretton Woods (Total) (1946-1970) Bretton Woods (Preconvertible) (1946-1958) Bretton Woods (Convertible) (1959-1970) Standard Deviation Floating Exchange (1974-1989) Standard Deviation Mean Standard Deviation Mean 2.4 3.7 2.7 5.6 4.5 2.7 3.8 2.6 2.2 4.0 4.1 4.6 3.0 11.5 2.8 4.6 2.1 5.6 4.7 3.9 5.8 3.5 2.5 6.2 5.1 7.3 3.9 16.0 2.6 3.4 3.2 5.5 5.1 2.9 3.8 1.5 1.5 1.8 3.6 1.3 1.5 2.1 5.6 9.4 3.3 8.8 2.6 7.3 12.9 2.4 6.1 1.2 3.2 2.4 2.6 4.6 7.8 1.5 3.6 0.9 4.6 2.0 4.2 1.1 6.4 3.0 3.9 0.9 1.9 0.5 7.1 2.8 3.2 1.2 0.0 1.2 2.6 1.3 2.0 0.2 0.9 8.1 4.5 8.5 7.2 6.1 8.8 4.7 2.0 2.1 5.0 3.9 8.1 2.5 5.6 2.8 1.8 3.3 2.1 2.7 2.6 3.3 1.8 2.1 7.3 4.6 5.7 2.4 5.2 3.4 2.2 3.9 2.7 1.1 3.3 4.4 2.9 2.3 3.5 3.9 8.9 3.5 5.8 1.9 1.4 2.6 1.3 2.5 1.7 1.9 2.1 1.5 2.1 1.7 3.5 1.3 2.5 2.7 4.2 1.9 1.5 1.1 2.4 2.2 3.7 1.0 1.2 0.7 6.8 1.5 4.2 1.7 2.7 0.4 4.2 1.7 3.0 0.9 4.4 1.6 1.9 0.4 2.1 0.5 2.3 0.7 6.1 2.1 5.7 2.1 7.2 7.4 3.2 4.1 1.7 4.7 4.7 14.5 5.3 3.1 0.6 0.8 1.3 6.4 0.5 1.1 3.6 8.6 4.7 10.1 8.5 9.7 4.7 6.2 6.3 3.2 12.8 11.5 16.2 6.0 13.3 5.8 3.2 5.9 7.5 16.2 4.0 7.8 6.4 1.7 17.6 14.7 20.1 5.1 15.9 8.3 2.9 5.6 7.2 21.8 3.2 10.5 7.0 5.5 10.9 8.6 14.9 8.3 12.4 1.5 2.9 4.7 6.5 7.0 3.9 2.0 8.6 13.5 5.7 8.8 5.8 10.6 13.4 2.4 5.5 4.5 3.4 6.3 3.9 4.9 4.8 2.1 5.4 2.4 2.0 1.7 7.5 2.0 9.9 4.1 7.2 3.7 11.6 6.2 8.5 4.4 9.7 2.6 4.1 1.7 9.5 2.1 4.4 1.0 Standard Deviation 3.1 3.1 2.6 4.9 5.5 1.4 3.2 - 1 .8 - 1 .5 -2 .1 2.2 - 1 .7 - 1 .9 -1 .1 7.6 7.8 4.7 9.1 7.3 6.1 11.7 1.0 0.9 3.4 1.0 -1 .1 1.0 United States United Kingdom Germany France Japan Canada Italy 1.8 1.1 1.7 1.5 1.4 2.3 1.0 5.1 2.4 2.9 4.6 3.8 2.8 4.0 mean convergence 1.5 0.3 United States United Kingdom Germany France Japan Canada Italy mean convergence Mean United States United Kingdom Germany France Japan Canada Italy 0.3 0.3 0.6 0.0 4.6 0.4 0.6 mean convergence Standard Deviation Mean Mean Real per capita growth1 Money Growth1 128 Standard Deviation Mean Inflation PGNP1 Gold Standard (1881-1913) Mean Standard Deviation Interwar (1919-1938) Mean Standard Deviation Bretton Woods (Total) (1946-1970) Mean Standard Deviation Bretton Woods (Preconvertible) (1946-1958) Mean Standard Deviation Bretton Woods (Convertible) (1959-1970) Mean Standard Deviation Floating Exchange (1974-1989) Mean Standard Deviation Short-term interest rate United States United Kingdom Germany France Japan Canada Italy 4.8 2.8 3.2 2.5 2.4 n.a. n.a. 0.9 0.8 0.9 0.6 0.5 3.5 3.0 4.6 3.1 2.0 0.9 n.a. 2.0 1.8 1.6 1.4 0.5 0.4 3.4 4.0 4.0 4.1 6.5 2.9 n.a. 1.9 2.5 1.5 1.9 0.8 2.0 2.0 2.3 4.1 3.2 6.9 1.4 n.a. 0.9 1.8 1.1 1.5 1.0 1.1 4.8 5.8 4.0 5.1 6.3 4.6 n.a. 1.6 1.6 1.7 1.9 0.6 1.3 8.9 11.2 5.9 10.3 5.2 10.3 n.a. 2.6 2.1 2.4 2.6 2.0 2.8 3.1 0.7 0.7 0.2 2.9 0.9 1.3 0.6 4.2 0.8 1.8 0.4 3.3 1.5 1.2 0.3 5.1 0.6 1.4 0.4 8.6 2.1 2.4 0.3 United States United Kingdom Germany France Japan Canada Italy 3.8 2.9 3.7 3.2 n.a. 3.5 4.2 0.3 0.2 0.2 0.3 0.6 0.7 1.6 0.8 0.8 0.6 3.9 5.2 6.3 5.7 7.0 4.5 6.0 1.3 1.8 0.7 0.8 0.1 1.5 0.7 3.0 3.9 5.9 5.8 n.a. 3.3 6.3 0.4 0.8 0.5 0.5 0.4 0.5 4.2 4.1 6.8 4.6 n.a. 4.7 5.9 0.5 0.4 5.0 6.6 6.7 5.7 7.0 5.8 5.7 1.1 1.3 0.7 1.0 0.1 1.1 0.7 10.4 12.1 7.8 10.9 7.1 11.0 13.7 2.1 2.8 1.5 2.4 1.8 2.0 3.3 mean convergence 3.6 0.4 0.3 0.1 5.1 0.9 0.9 0.3 5.5 0.8 1.0 0.5 4.7 1.3 0.5 0.1 6.1 0.6 0.9 0.3 10.4 1.7 2.3 0.5 4.8 2.9 2.4 2.8 -1 .5 n.a. n.a. 2.0 2.3 2.3 6.4 5.5 3.8 4.2 5.1 1.2 1.4 - 0 .8 n.a. 6.7 7.1 5.2 14.7 8.8 1.3 0.3 -0 .1 2.2 - 0 .9 1.9 - 0 .3 n.a. 3.9 3.4 2.6 5.2 2.5 4.2 -1 .6 - 2 .4 3.0 - 3 .3 3.4 - 2 .6 n.a. 4.7 3.2 3.6 6.9 3.5 4.8 2.4 2.3 1.6 1.2 1.1 2.2 n.a. 0.4 1.1 1.5 1.4 1.6 0.8 2.5 1.3 2.5 2.1 1.4 3.1 n.a. 2.8 5.1 1.9 2.8 3.5 3.2 2.3 1.5 3.7 1.8 2.5 1.9 7.3 3.0 0.5 1.0 3.6 0.8 - 0 .6 2.5 4.5 1.0 1.8 0.5 1.1 0.4 2.2 0.6 3.2 0.7 mean convergence Long term interest rate MARCH/APRIL United States United Kingdom Germany France Japan Canada Italy mean convergence 1993 129 Real short term interest rate2 FEDERAL RESERVE BANK Table 1 (continued) Descriptive Statistics of Selected Open Economy Macroeconomic Variables, the Group of Seven Countries 1881-1989 Annual Data: Mean, Standard Deviation_____________________________________________________ Gold Standard (1881-1913) O ST. LOUIS F Mean Standard Deviation Interwar (1919-1938) Mean Standard Deviation Bretton Woods (Total) (1946-1970) Mean Standard Deviation Bretton Woods (Preconvertible) (1946-1958) Mean Standard Deviation Bretton Woods (Convertible) (1959-1970) Mean Standard Deviation Floating Exchange (1974-1989) Mean Standard Deviation Real long term interest rate2 3.9 2.2 4.4 2.7 2.0 3.8 1.4 3.8 3.7 0.9 3.1 4.2 3.1 4.8 -0 .1 1.5 5.8 2.5 2.9 0.8 1.1 0.4 2.9 1.0 3.5 0.7 0.8 6.2 3.8 7.7 37.2 1.9 20.6 3.3 2.4 4.4 30.5 1.8 14.1 7.9 5.3 11.3 47.9 1.9 27.4 0.7 1.5 1.3 1.1 0.2 1.4 0.2 0.8 4.0 2.1 3.3 0.2 2.0 0.2 10.0 9.3 10.7 8.8 3.7 10.9 4.7 8.2 7.8 9.5 2.4 9.0 11.2 10.1 8.1 7.7 14.5 12.9 0.9 0.6 1.8 1.2 8.9 1.8 6.9 2.3 6.5 9.3 - 0 .7 -0 .8 4.3 -1 .2 n.a. - 0 .7 -1 .5 4.5 1.5 8.5 2.5 1.4 0.9 4.0 1.6 0.2 0.1 0.0 4.5 0.0 1.5 6.8 3.9 18.4 6.8 2.7 13.6 7.9 9.5 15.6 9.0 3.4 20.1 0.7 2.4 1.8 2.5 15.9 1.6 7.4 0.9 1.3 7.5 4.5 9.4 4.2 4.6 4.0 4.1 2.4 3.4 0.4 3.4 1.3 United States4 United Kingdom Germany France Japan Canada Italy 0.2 0.2 0.0 2.9 0.0 1.4 mean convergence 0.7 0.9 3.7 3.0 2.9 3.5 n.a. 2.9 4.2 2.2 2.5 2.4 6.5 4.4 2.6 4.4 6.2 Nominal exchange rate5 130 0.7 1.0 0.9 1.0 1.3 0.7 2.2 3.6 2.8 2.8 4.4 1.3 3.8 9.4 6.8 7.1 6.0 15.1 mean convergence 4.4 12.9 2.5 3.2 4.3 1.8 1.7 3.4 2.3 0.8 1.1 4.3 0.4 1.7 1.3 0.3 4.6 5.4 6.9 1.0 n.a. 5.4 3.4 United States United Kingdom Germany France Japan Canada Italy Bretton W oods Bretton W oods Bretton Woods Gold Standard Interw ar (Total) (Preconvertible) (Convertible) Floating Exchange (1881-1913)_____________(1919-1938)_____________(1946-1970)_____________(1946-1958)_____________(1959-1970)_____________(1974-1989) Mean Standard Deviation Mean Standard Deviation Mean Standard Deviation Mean Standard Deviation Mean Standard Deviation Mean Standai Deviatic Real exchange rate3,5 United States4 United Kingdom G erm any France Japan Canada Italy 1.7 2.4 4.3 6.6 2.6 2.1 1.5 1.2 5.0 5.5 2.2 1.7 8.5 5.8 9.0 7.8 3.2 13.3 10.0 9.2 6.9 7.2 2.8 16.9 1.7 4.2 2.8 4.1 3.5 2.4 8.0 1.0 6.5 5.1 5.6 2.9 2.3 18.7 5.9 3.9 6.2 4.8 3.3 13.1 8.2 7.3 7.7 4.1 2.5 25.2 1.7 2.5 1.9 2.5 2.7 1.5 2.4 1.0 3.9 1.8 2.9 1.5 1.7 1.6 12.3 8.8 9.2 9.6 3.8 8.6 6.1 8.2 7.7 8.9 2.0 7.8 mean convergence 3.3 1.5 2.9 1.6 7.9 2.3 8.8 3.2 3.8 1.3 6.0 3.8 6.2 2.3 9.2 5.4 2.2 0.5 2.1 0.8 8.7 1.4 6.8 1.8 Data Sources: See Data Appendix to Bordo (1993). MARCH/APRIL 1993 131 'M e a n grow th rate calculated as the tim e coefficient from a regression of the natural logarithm of the variable on a constant and a tim e trend. C a lc u la te d as the nom inal interest rate minus the annual rate of change of the consum er price index (CPI). 3Absolute rates of change. 4T rade-w eighted nom inal and real exchange rate starting in 1960. C a lc u la te d as the nom inal exchange rate divided by the ratio of foreign CPI to the U.S. CPI. 132 Figure 1 Inflation Rates, 1880-1989, G7 Countries Percent 4 0 -i------------------------------------------------------ 30- ----------1 ----------1 ----------1 ----------1 ---------- 1 ----------1 ----------1 ----------1--------- 1 ----------1 ---------1880 1890 1900 1910 U.S. ------ 1920 1930 U.K. 1940 ------ 1950 Germany 1960 1970 ------ 1980 Canada Percent France FEDERAL RESERVE BANK OF ST. LOUIS ------ Italy ------ Japan 1990 133 The Bretton Woods convertible subperiod had the most stable inflation rate o f any regime as judged by the standard deviation. By contrast, the preconvertible Bretton Woods period exhibited greater inflation variability than either the gold standard period or the recent floating exchange rate period. The evidence of a high degree of price stability in the convertible phase of Bretton Woods is also consistent with the traditional view that fixed rate (commodity-based) regimes provide a stable nominal anchor; however, the remarkable price stability during this period may also reflect the absence of major shocks. Real p e r capita GNP. Generally, the Bretton Woods period, especially the convertible period, exhibited the most rapid output growth of any monetary regime, and not surprisingly the interwar period the lowest (see figure 2). Output variability was also lowest in the convertible subperiod o f Bretton Woods, but because of higher variability in the preconvertible period, the Bretton Woods system as a whole was more variable than the floating exchange rate period. Both pre-W orld W ar II regimes exhibit higher variability than their post-World W ar II coun terparts.3 The divergence of output variability 0 between countries was also lowest during the Bretton Woods regime, with the interwar regime showing the highest divergence.3 The greater 1 convergence of output variability under Bretton Woods may reflect conformity between countries’ business fluctuations, created by the operation of the fixed exchange rate regime.3 2 M oney g ro w th (M2). Money grew consider ably more rapidly across all countries after World W ar II than before the war (see figure 3). There is not much difference between Bretton Woods and the subsequent floating exchange rate regime. Within the Bretton Woods regime, money grew more rapidly in the preconvertibility period than in the convertibility period. Money growth rates showed the least divergence between 30Baxter and Stockm an (1989) and Eichengreen (1992a) use residuals from a linear trend to the logarithm of real output as a detrending filte r rather than the logarithm ic first d iffe r ence used here. A ccording to th e ir results, real output va riab ility is not greater in the floating than in the fixed period. 31However, using th e ir alternative m easure of convergence— the G DP-weighted standard deviation of the individual country series around the G-7 aggregate— Bayoum i and Eichengreen (1992a) report th a t the lowest degree of dis persion of real GDP growth was in the floating rate period, followed by the Bretton W oods convertible period. Sim ilar results hold for the real GNP per capita data in table 1. For Bayoum i and Eichengreen (1992a) the decline in the countries during the fixed-exchange-rate gold standard and the convertible Bretton Woods regime, with the greatest divergence in the preconvertible Bretton Woods period and the interwar period. Like inflation and real output variability, money growth variability was lowest in the convertible Bretton Woods period. This, however, was not the case for the preconvertible period, which was the most variable of any regime. It also exhibited the greatest divergence in variability between countries. To the extent that one of the properties o f adherence to a fixed-exchangerate regime is conformity o f monetary growth rates between countries, these results are sym pathetic to the view that the Bretton Woods system really began in 1959. Short-term and long-term interest rates. The underlying data for short-term and long term interest rates are seen in figures 4 and 5. As in other nominal series, the degree o f con vergence of mean short-term interest rates is highest in the convertible Bretton Woods period. Long-term rates are most closely related in the classical gold standard regime, with the convertible Bretton Woods period not far behind. McKinnon (1988) has similar findings. He views them as evidence of capital market integration under fixed exchange rates. The lack o f conver gence in the preconvertibility Bretton Woods period reflects the presence of pervasive capital controls. Convergence of nominal interest rates would not be expected under floating exchange rates. Convergence of standard deviations is also highest in the gold standard period followed by Bretton Woods. Long-term rates were most stable and least divergent under the classical gold stand ard, followed by the two Bretton Woods subperiods, with floating exchange rates the least stable. The evidence that nominal interest rates are more stable and convergent between countries under fixed exchange rate (commodity-based) regimes is consistent with the traditional view. dispersion of real growth and the rise in the dispersion of inflation rates between the Bretton W oods convertible period and the float have the follow ing explanations: the move to fle xib le rates allowed countries to stabilize th e ir relative growth rates in the face of asym m etric supply shocks at the expense o f th e ir relative inflation rates. They also report that, when they apply the linear trend filte r of Baxter and Stockm an (1989), evidence of a rise in the cross country correlation between o utput m ovem ents after 1970 is considerably reduced. 32See Bordo and Schw artz (1989a) and Darby and Lothian (1989). M ARCH/APRIL 1993 134 Figure 2 Per Capita Income Growth Rates, 1880-1989, G7 Countries Percent ------ U.S. ------ U.K. ------ Germany ------ Canada Percent ------ France FEDERAL RESERVE BANK OF ST. LOUIS ------ Italy ------ Japan 135 Figure 3 Money Growth Rates, 1880-1989, G7 Countries Percent ------ U.S. ------ U.K. ------ Germany ------ Canada Percent ------ France ------ Italy ------ Japan M ARCH/APRIL 1993 136 Figure 4 Short-Term Interest Rates, 1880-1989, G7 Countries Percent ------ U.S. ------- U.K. ------ Germany ------- Percent France FEDERAL RESERVE BANK OF ST. LOUIS ------ Japan Canada 137 Figure 5 Long-Term Interest Rates, 1880-1989, G7 Countries Percent ------ U.S. ------- U.K. ------ Germany ------- Canada Percent ------- France ------- Italy ------ Japan MARCH/APRIL 1993 138 Real short-term and real long-term interest rates. For the underlying real short term and real long-term interest rate data, see figures 6 and 7. The real interest rates are e^c post rates calculated using the rate of change of a consumer price index.3 Unlike the nominal 3 series, the degree of convergence in means between real short-term interest rates is lowest in the floating exchange rate period, next lowest in the Bretton Woods convertible period and highest in the preconvertible period. For long term real rates, as in the case of nominal rates, convergence is highest under the gold standard followed by the Bretton Woods convertible regime. It is lowest under preconvertible Bretton Woods. The real short-term interest rate is most stable across countries during the Bretton Woods con vertible period. It also shows the least amount of divergence in standard deviations. The same holds for real long-term interest rates. The behavior of real interest rates across regimes is consistent with McKinnon’s explana tion.3 He argued that fixed exchange rates 4 encourage capital market integration by elimina ting devaluation risk. This reduces variability in short-term real interest rates. Similarly, real long-term interest rates are stabilized by pooling across markets, which reduces capital market risk. N om inal and real exchange rates. The lowest mean rate of change of the nominal ex change rate and the least divergence between rates of change occurred during the Bretton Woods convertible and gold standard periods, with the former exhibiting the lowest degree of divergence. Exchange rates during the precon vertibility Bretton Woods regime changed almost as much as during the floating period. This mainly reflected the major devaluations of 1949 (see figure 8).3 Nominal exchange rates were 5 least variable in the gold standard and converti ble Bretton Woods periods and the most varia ble and most divergent in the Bretton Woods preconvertible period. As with the nominal exchange rate, the lowest mean rate of change in the real exchange rate 33Define the real interest rate as rt = it - Alog Pt; w here it is the nom inal interest rate and Alog Pt = log Pt - log Pt _ n is the percentage change in the consum er price index. 34See M cKinnon (1988). 35See Bordo (1993) table 2. 36Also see D ornbusch (1976). 37Stockm an (1983 and 1988) argues that greater variability FEDERAL RESERVE BANK OF ST. LOUIS across countries and the least divergence between countries was in the Bretton Woods convertible period, with the divergence in gold standard period next smallest (see figure 9). The highest rate of change was in the floating exchange rate period. Similarly data from the Bretton Woods convertible period had the lowest standard deviation across countries and the least divergence between standard devia tions, with the gold standard again next in these rankings. The other regimes were characterized by much greater variability and divergence. These results shed light on the relationship between the nominal exchange rate regime and the behavior of real exchange rates. Mussa (1986) presented evidence for 16 industrial countries in the post-World W ar II period showing the similarity between nominal and real exchange rate variability under floating rates. His explanation for greater real exchange rate variability under floating rates than under fixed rates is nominal price rigidity.3 The expla 6 nation may, however, be questioned. For exam ple, a fixed nominal exchange rates may produce greater trade stability that will be reflected in the real exchange rate, as is evident for both the Bretton Woods and gold standard periods. Yet as Eichengreen (1991b) points out and as can be seen in table 4, these results could be explained by the fact that both periods were characterized by few shocks.3 7 Finally, based on monthly data between 1880 and 1986 for the United Kingdom and the United States, Grilli and Kaminsky show that, with the exception of the post-World W ar II period, no clear connection exists between the nominal exchange rate regime and the variability of real exchange rates.3 My results for the Group of 8 Seven countries show a clear correlation between nominal exchange rate rigidity and lower real exchange rate variability for the gold standard and Bretton Woods convertible regime. For the preconvertible Bretton Woods periodde ju r e a type of fixed exchange rate regime— the correlation is not evident. I do not distin guish between fixed and flexible periods in the interwar segment as do Grilli and Kaminsky, in real exchange rates under floating rates than under fixed rates reflects the response of real exchange rates to productivity shocks, with changes in the real exchange rate producing nom inal exchange rate volatility. This vo la tility is offset under fixed rates by exchange m arket in te r vention. 38See G rilli and K am insky (1991). 139 Figure 6 Real Short-Term Interest Rates, 1880-1989, G7 Countries Percent ----- - U.S. ------ U.K. ------ Germany ------ Canada Percent Japan ------ France M ARCH/APRIL 1993 140 Figure 7 Real Long-Term Interest Rates, 1880-1989, G7 Countries Percent ------ U.S. ------ U.K. ------ Germany ------ Canada Percent France FEDERAL RESERVE BANK OF ST. LOUIS ------ Italy ------ Japan 141 Figure 8 Absolute Change in Nominal Exchange Rates, 1880-1989, G7 Countries Percent ------ U.K. ------ Germany ------ Canada Percent 140— i— 1880 1890 1900 France 1910 1920 1930 ------ 1940 Italy 1950 1960 1970 ------ 1980 1990 Japan M ARCH/APRIL 1993 142 Figure 9 Absolute Change in Real Exchange Rates, 1880-1989, G7 Countries 1880 1890 1900 ---------1 --------- 1 -------- ---------- 1 _l--------- (-----— ,--------- 1 1910 1920 1930 1940 1950 1960 1970 1980 Germany U.K. 1990 Canada i’ * ~ -i — — r—------ 1-------- 1--------1 --------- 1-------- 1-------- 1-------- r 1880 1890 1900 1910 France FEDERAL RESERVE BANK OF ST. LOUIS 1920 1930 ------ 1940 Italy 1950 1960 1970 ------ 1980 Japan 1990 143 hence that period cannot be used in the com parison.3 9 In summary, the Bretton Woods regime exhibited the best overall macroeconomic performance of any regime. This is especially so for the con vertible period (1959-70).4 As the summary 0 statistics in table 1 show, both nominal and real variables were most stable in this period. The floating exchange rate regime, on most criteria, was not far behind the Bretton Woods converti ble regime, whereas the classical gold standard exhibited the most stability and the closest con vergence of financial variables. The preconvertible Bretton Woods period (194658) was considerably less stable for the average of all countries for both nominal and real variables than other regimes. Also both nominal and real variables did not vary nearly as closely together. These differences likely reflect the presence of pervasive exchange and capital controls before 1958 and, related to these, more variable and more rapid monetary growth. These data, how ever, are limited. Although they show excellent performance for the convertible Bretton Woods regime, they do not tell us why it did well— whether it reflected a set of favorable circum stances, whether it reflected the absence of aggravating shocks, whether it reflected stable monetary policy by the key country of the sys tem, the United States, or whether it masked underlying strains to the system. Inflation Persistence A second piece of evidence is persistent infla tion. Evidence of persistence in the inflation rate suggests that market agents expect the monetary authorities to continually follow an inflationary policy; its absence would be consis tent with the belief that the authorities are fol lowing a stable monetary rule, such as the gold standard's convertibility rule. Barsky (1987) pre sented evidence for the United Kingdom and United States based on both autocorrelations and time series models that inflation under the 39M eltzer (1990) in a com parison of EMS and non-EMS contries in the floating rate period also finds a strong correla tion between changes in nom inal and real exchange rates. 40M cKinnon (1992) treats the period 1950 to 1970 as the de facto dollar standard. He view s this period rather than 1959 to 1971 as the appropriate one for m aking the type of regim e com parisons undertaken here. I m ade the same calculations as those shown in table 1 for the period 1950 to 1971. V irtu a lly every variable for each country exhibited greater instability than in the 1959 to 1970 period. This reinforces my choice o f dates. gold standard was very nearly a white-noise process, whereas in the post-World War II period, the inflation rate exhibited considerable persistence. Alogoskoufis and Smith (1991) also show, based on AR(1) regressions of the infla tion rate, that inflation persistence in the two countries increased between the classical gold standard period and the interwar period and between the interwar period and the post-World W ar II period.4 1 Table 2 presents the inflation-rate coefficient from the type of AR(1) regressions on consumer price index inflation estimated by Alogoskoufis and Smith, for the Group of Seven countries over successive regimes since 1880, as well as the standard errors and the Dickey-Fuller tests for a unit root.4 2 The results, as in Alogoskoufis and Smith, show an increase in inflation persistence for most countries between the classical gold standard and the interwar period, and also between the interwar period and the post-World War II period as a whole. Within the post-World War II period, inflation persistence is generally lower, with the exceptions of France and Japan, in the preconvertible Bretton Woods than the convertible period. This suggests that though the immediate post-World W ar II period was characterized by rapid inflation, market agents may have expected a return to a stable price regime. The higher degree of persistence in the convertible regime suggests that this expectation lost credence. Finally, the evidence that persistence was generally highest during the floating exchange rate regime may imply that the public realized that there was no longer a stable nominal anchor. Forecast Errors in Inflation and Growth A third piece of evidence relates to the forecast errors of inflation and real output growth. Accord ing to Meltzer and Robinson (1989), “ a welfare maximizing monetary rule would reduce varia bility to the minimum inherent in nature and 41Also see Alogoskoufis (1992), who attributes the increase in persistence to the accom m odation by the m onetary authorities of shocks. This evidence is also consistent with the results of Klein (1975). 42Eichengreen (1992b) also presents these statistics fo r four of the countries. MARCH/APRIL 1993 Table 2 Persistence of CPI Inflation: Group of Seven Countries 1880-1989 Annual Data: Coefficient of AR1 Regression; (Standard error); t-s ta tis tic for unit root test United States AR1 Standard Error Coefficient Gold Standard Interw ar Bretton W oods Total) Bretton W oods (Preconvertible) Bretton W oods (Convertible) Floating Exchange Post W orld W ar II United Kingdom tstatistic AR1 Standard Error Coefficient tstatistic AR1 Standard Error Coefficient tstatistic (0.18) (0.17) 4.05 3.18 0.30 0.35 (0.17) (0.19) 4.03 3.37 0.51 0.51 (0.16) (0.21) 3.06 2.33 -0 .2 2 0.42 (0.18) (0.24) 6.78 2.42 0.49 (0.19) 2.68 0.33 (0.20) 3.35 - 0 .0 3 (0.21) 4.90 0.56 (0.16) 2.75 0.41 (0.27) 2.19 0.15 (0.29) 2.93 - 0 .0 7 (0.31) 3.00 0.60 (0.27) 1.48 1.07 0.68 0.65 (0.20) (0.18) (0.12) - 0 .3 5 1.78 2.92 0.57 0.69 0.75 (0.34) 1.26 1.63 2.50 0.44 0.83 0.31 (0.31) (0.14) (0.15) 1.81 1.21 4.60 0.12 0.85 0.69 (0.14) (0.16) 6.29 0.94 2.82 (0.19) (0 1 0 ) Italy Canada Japan tstatistic AR1 Standard Coefficient Error tstatistic AR1 Standard Coefficient Error tstatistic 0.22 0.70 (0.18) (0.25) 4.331 1.20 0.08 0.35 (0.18) (0.20) 5.11 3.25 0.28 0.28 (0.14) (0.18) 5.14 4.00 0.52 (0.18) 2.67 0.39 (0.19) 3.21 0.21 (0.12) 6.58 0.47 (0.27) 1.96 0.32 (0.28) 2.43 0.18 (0.18) 4.56 0.18 0.70 0.54 (0.31) (0.19) (0.13) 2.64 1.58 3.54 0.81 0.75 0.65 (0.20) (0.17) (0.11) 0.95 1.47 3.18 0.38 0.75 0.28 (0.29) (0.17) (0.10) 2.14 1.47 7.20 For data sources see ta b le 1. The 5 percent sig n ifica n ce level for unit root test with 25 observations is 3.00. 1G N P d eflator was used because o f unavailability of CPI data. AR1 Standard Error Coefficient 0.27 0.45 AR1 Standard Error C oefficient Gold Standard Interw ar Bretton W oods (Total) Bretton W oods (Preconvertible) Bretton W oods (Convertible) Floating Exchange Post W orld W ar II ________________ Germ any_________________________ France tstatistic (0.11) 145 institutional arrangements.” They measure varia bility by the mean absolute error (MAE) of a oneperiod forecast based on the univariate multistate Kalman Filter (MSKF). Following their approach, table 3 presents the MAEs for inflation and real growth for the Group of Seven countries over suc cessive regimes. The MSKF forecasts incorporate both transitory and permanent shocks to the rateof-change series.4 3 absence of shocks to the underlying environment. One way to shed light on this issue, following earlier work by Bayoumi and Eichengreen, is to identify underlying shocks to aggregate supply and demand.4 According to them, aggregate 4 supply shocks reflect shocks to the environment and are independent of the regime, but aggregate demand shocks likely reflect policy actions and are specific to the regime. The smallest forecast errors for inflation on aver age were for the Bretton Woods convertible period, followed by the gold standard and the floating rate periods. The largest errors were for the interwar period, followed by the preconvertible Bretton Woods period. The most notable exception to the pattern was the United Kingdom, where the floating rate period exhibited the largest varia bility. The approach used to calculate aggregate sup ply and demand shocks is an extension of the bivariate structural vector autoregression (VAR) methodology developed by Blanchard and Quah.4 5 Following Bayoumi and Eichengreen (1992a), I estimated a two-variable VAR on the rate of change of the price level and output.4 Restric 6 tions on the VAR identify an aggregate demand disturbance, which is assumed to have only a temporary effect on output and a permanent effect on the price level, and an aggregate sup ply disturbance, which is assumed to have a permanent effect on both prices and output.4 7 Overidentifying restrictions, namely, restrictions that demand shocks are positively correlated and supply shocks are negatively correlated with prices, can be tested by examining the impulse response functions to the shocks. For real growth, as for the inflation rate, the lowest MAE, on average, occurred in the con vertible Bretton Woods period. Another excep tion to this pattern was Japan. The highest MAE was again in the interwar and the preconverti ble Bretton Woods period. The floating exchange rate period, though more variable than the con vertible Bretton Woods period, was slightly less variable than the gold standard regime. These results are quite consistent with those o f table 1. The Bretton Woods convertible period was the most stable both in an ep post and ex ante sense. The performance o f the gold standard and the float, however, are not much worse, at least for real growth for the float and inflation for the gold standard. Dem and and Supply Disturbances An important issue is the extent to which the performance of alternative monetary regimes, as revealed by the data in the preceeding tables, reflects the operation of the monetary regime in constraining policy actions or the presence or 43M eltzer and Robinson (1989) present th e ir results for levels, grow th rates, and perm anent growth rates of the series. I present only growth rates to m ake the results com parable to those in table 1. 44See Bayoum i and Eichengreen (1992a, 1992b, 1992c and 1992d). 45See Blanchard and Q uah (1989). The methodology has important limitations that suggest that the results should be viewed with caution. The key limitation is that one can easily imagine frameworks in which demand shocks have permanent effects on output, where as supply shocks have only temporary effects.4 8 I estimated supply (permanent) and demand (temporary) shocks, using annual data for each of the Group o f Seven countries, over alterna tive regimes in the period 1880-1989. The VARs are based on three separate sets of data—1880-1913, 1919-39 and 1946-89—with the war years omit ted because complete data on them were avail able for only four o f the countries.4 The VARs 9 have two lags. I also did the estimation for supply shocks are orthogonal; the fo u rth is that dem and shocks have only tem porary effects on output, that is, that the cum ulative effect of dem and shocks on the rate of change in output m ust be zero. 48See Keating and Nye (1991). 49For results using the com plete data set for these four countries, see appendix table 1 and appendix fig u re 1. 46Both variables w ere rendered stationary by first differencing. 47S pecifically, four restrictions are placed on the m atrix of the shocks: two are sim ple norm alizations, which define the variances of the shocks to aggregate dem and and aggregate supply; the third assum es that dem and and M ARCH/APRIL 1993 FEDERAL Table 3 RESERVE Forecast Errors in Inflation and Real Growth: Group of Seven Countries 1880-1989 Annual Data: Mean Absolute Errors Using the Multistate Kalman Filter____________ BANK Bretton W oods Bretton W oods Bretton W oods Gold Standard Interw ar (Total) (Preconvertible) (Convertible) Floating Exchange (1880-1913)___________ (1919-1939)____________(1946-1970)____________(1946-1958)____________(1959-1970)____________(1973-1989) O ST. LO UIS F G rowth Inflation Growth Inflation G rowth Inflation Growth Inflation Growth Inflation G rowth Inflation United States United Kingdom G erm any France Japan Canada Italy 2.04 1.42 1.69 2.25 1.69 1.58 2.08 1.59 2.10 1.69 2.47 3.95 0.80 2.01 4.79 4.11 8.07 5.54 2.71 7.16 2.08 4.61 4.41 3.97 7.07 6.83 3.99 7.58 3.22 1.41 2.77 1.51 1.59 2.19 4.21 2.32 1.47 2.59 3.13 2.63 1.92 6.35 5.02 1.82 3.24 2.28 0.98 3.06 7.04 3.44 1.76 4.99 4.62 5.48 2.96 10.53 1.28 0.97 2.49 0.87 1.60 1.24 1.15 1.12 1.15 1.18 2.14 0.96 0.79 1.81 1.65 2.83 1.49 1.47 1.34 1.72 1.14 2.46 4.66 1.30 2.74 2.95 2.29 3.84 Average 1.82 2.09 4.92 5.50 2.41 2.92 3.35 4.82 1.37 1.31 1.66 2.89 146 Note: For data sources, see table 1. 147 aggregated price and output data for the Group of Seven countries.5 0 The overidentifying restrictions that demand shocks be positively correlated and supply shocks be negatively correlated with the price level are satisfied for all countries for the two post-World W ar II regimes. But for the period before World W ar II, for a number of countries, including the United States, United Kingdom, and France, they are not. Supply shocks were positively correlated with prices. This can be seen in the impulse response functions displayed in figure 10. Fig ure 10 shows the impulse responses, to one standard deviation shocks in aggregate supply and aggregate demand, on output and prices for the Group of Seven countries aggregate by regime.5 1 Keating and Nye (1991) attempted to explain this result by possible hysteresis effects. Bayoumi and Eichengreen (1992a) argued that the perverse impulse response patterns for the classical gold standard and interwar periods reflected the interaction of a positive aggregate demand curve with a very steep aggregate supply curve. They explain the positively sloped aggregate demand curve as reflecting the effects of gold discover ies induced by the supply shock of agricultural settlements in the United States and Australia. These results may also reflect a limitation of the Blanchard-Quah methodology. Table 4 presents the standard deviations of supply and demand shocks for the Group of Seven countries and the Group of Seven coun tries taken as a whole (Group of Seven aggregate) by regime. I also show, following Bayoumi and Eichengreen, the weighted average of the indi vidual country shocks.5 Figures 11 and 12 show 2 the shocks for the Group of Seven aggregate and for each of the seven countries. 50The G roup of Seven aggregate incom e growth and infla tion rate are a w eighted average of the rates in the d iffe r ent countries. The weights for each year are the share of each c o u n try’s nom inal national incom e in the total incom e in tlfe G roup of Seven countries, w here the national incom e data are converted to U.S. dollars using the actual exchange rates. 51The im pulse response functions were calculated from VARs run for the separate regim e periods. Because the num ber of observations was lim ited, the Bretton W oods regim e could not be split into the two subperiods shown in preceding tables. Table 4 shows for the Group of Seven aggre gate that the convertible Bretton Woods regime was the most tranquil of all the regimes—neither supply nor demand shocks dominated. It was not, however, that much less turbulent than the suc ceeding float. The interwar period, unsurpris ingly, shows the largest supply and demand shocks.5 Sizeable supply and demand shocks 3 that are two or three times greater than the post-World W ar II period also characterize the classical gold standard.5 4 For individual countries, the Bretton Woods convertible period was the most stable in four countries and the flexible exchange rate period was the most stable in three. The difference between the convertible Bretton Woods period and the floating exchange rate period, however, was not great in any country. The interwar period as expected was the most volatile. Both types of shocks were the largest in every coun try except the United Kingdom. Finally, in the majority of countries, with the principal excep tions being the United Kingdom and Germany, both supply and demand shocks were consider ably greater in the gold standard period than in the post-World W ar II period. The dispersion o f demand shocks across coun tries, as measured by the GNP-weighted stand ard deviation of the individual country shocks around the Group of Seven aggregate, reveals very little difference between the gold standard and the post-World War II regimes, with the convertible Bretton Woods regime displaying the highest degree of convergence. Dispersion is much greater in the interwar period. The dispersion of supply shocks is considerably greater during the gold standard and the inter war periods than in any of the post-World W ar II regimes. • 1920s and early 1930s reveal a m ajor negative dem and shock consistent with Friedm an and S chw artz’s (1963) attribution of the onset of the G reat Depression to m onetary forces. A fter 1931, negative supply shocks predom inate, consistent w ith B ernanke’s (1983) and Bernanke and Jam es (1991) explanation for the severity of the G reat Depression that stresses the collapse of the financial system. 54Though the shocks are sm aller, the rankings by regim e for the weighted average of individual country shocks are sim ilar to the G-7 aggregate. 52See Bayoum i and Eichengreen (1992a). 53The results for the G roup of Seven in the interw ar period (figures 11 and 12) as well as those for four countries (appendix figure 1) are sim ila r to those reported for the United States by C ecchetti and Karras (1992), who esti m ate a three-variable VAR with m onthly data. The late M ARCH/APRIL 1993 148 Figure 10 Impulse Response Functions of Demand and Supply Shocks on Prices (Dotted Lines) and Output (Solid Lines), G7 Aggregate by Regimes, Annual Data, 1881-1989 Effects of Aggregate Demand Shocks, 1881-1913 Effects of Aggregate Supply Shocks, 1881-1913 FEDERAL RESERVE BANK OF ST. LOUIS 149 Figure 10 (continued) Impulse Response Functions of Demand and Supply Shocks on Prices (Dotted Lines) and Output (Solid Lines), G7 Aggregate by Regimes, Annual Data, 1881-1989 Effects of Aggregate Demand Shocks, 1919-1939 Effects of Aggregate Supply Shocks, 1919-1939 M ARCH/APRIL 1993 150 Figure 10 (continued) Impulse Response Functions of Demand and Supply Shocks on Prices (Dotted Lines) and Output (Solid Lines), G7 Aggregate by Regimes, Annual Data, 1881-1989 Effects of Aggregate Demand Shocks, 1946-1970 Effects of Aggregate Supply Shocks, 1946-1970 FEDERAL RESERVE BANK OF ST. LOUIS 151 Figure 10 (continued) Impulse Response Functions of Demand and Supply Shocks on Prices (Dotted Lines) and Output (Solid Lines), G7 Aggregate by Regimes, Annual Data, 1881-1989 Effects of Aggregate Demand Shocks, 1946-1989 Effects of Aggregate Supply Shocks, 1946-1989 M ARCH/APRIL 1993 152 Figure 10 (continued) Impulse Response Functions of Demand and Supply Shocks on Prices (Dotted Lines) and Output (Solid Lines), G7 Aggregate by Regimes, Annual Data, 1881-1989 Effects of Aggregate Demand Shocks, 1971-1989 Effects of Aggregate Supply Shocks, 1971-1989 FEDERAL RESERVE BANK OF ST. LOUIS Table 4 Supply (Permanent) and Demand (Temporary) Shocks: 1880-1989 Annual Data: Standard Deviations of Shocks (percent); Dispersion of shocks across countries (percent) Gold Standard 1883-1913 Interwar 1921-1939 Bretton Woods (Total) 1948-1970 Bretton W oods (Preconvertible) 1948-1958 Bretton W oods (Convertible) 1959-1970 Floating Exchange 1973-1989 Post WW II 1948-1989 Demand Supply Demand Supply Demand Supply Demand Supply DemandI Supply Demand Supply Demand Supply United States United Kingdom G erm any F rance Japan C anada Italy 2,03 2.66 2.37 4.58 4.85 0.93 3.16 3.81 2.16 2.32 3.75 3.39 2.75 3.13 4.45 1.93 4.47 7.17 6.28 4.01 7.40 6.73 3.52 3.13 5.19 5.36 8.61 4.14 2.33 2.62 2.88 3.50 3.18 2.42 2.76 1.54 1.95 2.65 1.75 2.56 2.60 1.75 3.11 3.06 1.85 3.23 4.05 3.09 2.93 1.98 2.61 2.87 1.61 1.97 3.70 1.85 1.37 2.26 3.34 3.77 2.83 1.74 2.56 1.07 1.14 2.66 1.84 2.78 0.98 1.73 1.72 3.57 1.66 1.93 2.39 2.66 3.58 1.94 4.31 1.39 1.52 2.50 2.10 1.91 2.07 3.03 2.36 2.84 2.75 2.48 3.30 1.68 3.10 2.20 1.70 2.69 2.47 2.47 G7 G 7* D ispersion 1.56 1.49 3.94 2.21 2.08 5.03 3.09 2.99 8.25 4.12 4.61 8.58 1.24 1.11 3.93 0.86 0.81 2.38 1.99 1.64 3.44 1.00 1.01 1.89 0.75 0.75 4.09 0.80 0.71 2.54 1.50 1.40 3.53 0.96 1.60 2.74 1.40 1.28 3.69 0.91 1.32 2.63 G 7 *: W eighted average of individual country shocks; the weights are calculated as the share of each co u n try’s National Income in the Total Incom e in the G7 countries, w here the G NP/G D P data are converted to dollars using the actual exchange rate. D ispersion = D (w e ig ht^sh o ck-E w e ig h ty shockj)2)0-5 for i = United States, United Kingdom , G erm any, France, Japan, C anada, Italy MARCH/APRIL 1993 153 G7: G 7-aggregate data 154 Figure 11 Supply and Demand Shocks: G7 Aggregate, 1880-1989, Annual Data Percent Figure 12 Supply and Demand Shocks: 1880-1989, United States Percent -16 1880 1890 1900 1910 FEDERAL RESERVE BANK OF ST. LOUIS 1920 1930 1940 1950 1960 1970 1980 1990 155 Figure 12 (continued) Supply and Demand Shocks: 1880-1989, United Kingdom Percent 10 -10 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 1950 1960 1970 1980 1990 Supply and Demand Shocks: 1880-1989, Germany Percent 10 -10 1880 1890 1900 1910 1920 1930 1940 M ARCH/APRIL 1993 156 Figure 12 (continued) Supply and Demand Shocks: 1880-1989, France Percent -16 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 Supply and Demand Shocks: 1880-1989, Japan Percent 18—---------------------------------------------------------------------------------------------- ---------------------------------1 I Supply 14- -10 K Demand -14-|-----------1 -----------1 ----------- 1 -----------1 -----------1 ----------- 1 -----------1 ----------- ,-----------1 ----------- 1 ----------t --------- 1 --------- 1 --------- 1 --------- 1 --------- 1 --------- 1 --------- 1 --------- 1 --------- r 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 FEDERAL RESERVE BANK OF ST. LOUIS 157 Figure 12 (Continued) Supply and Demand Shocks: 1880-1989, Canada Percent 16 -16 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 1940 1950 1960 1970 1980 1990 Supply and Demand Shocks: 1880-1989, Italy Percent 20 -20 1880 1890 1900 1910 1920 1930 M ARCH/APRIL 1993 158 In sum, the evidence on supply and demand shocks is quite similar to the measures of vola tility drawn from the forecast errors using the MSKF. The gold standard regime, as well as the interwar period, emerges as a relatively unstable period stressed by widely dispersed supply shocks. By contrast, the Bretton Woods convertible period is the most stable, with the floating exchange rate period not far behind. These results raise the interesting question, why in the past century was the classical gold standard so durable in the face of substantial shocks, whereas Bretton Woods was so fragile in the face of the mildest shocks? Responsiveness to Shocks The final piece of evidence to be calculated in the comparison of regime performance is the response of the price level and output to the aggregate supply (permanent) and aggregate demand (temporary) shocks. Evidence of a more rapid adjustment of prices and output to shocks may help explain why one regime may have been more durable than another. A measure of speed of response can be gleaned from the impulse response functions derived from the bivariate VABs. In addition, as a crude measure o f response speed, which allowed easy comparison of all seven countries during the four regimes, I calculated the mean absolute lag of the response functions.5 Table 5 presents 5 these measures. The response o f output to both demand and supply shocks for the Group o f Seven aggregate and for most of the individual countries was markedly more rapid under the gold standard regime than under the postwar regimes (an ex ception is the U.S. response to demand shocks) and within the postwar regimes was slightly more rapid under the Bretton Woods regime than the floating exchange rate regime. The response of prices to both demand and supply shocks was considerably more rapid during the gold standard (and the interwar) regime than the postwar regimes for the Group o f Seven and most countries.5 Within the postwar 6 period, it was considerably more rapid under 55T he form ula was S i | A c, | / S | Aq | for i = 1 to 40, w here i is the year and A C the im pulse response, cal ; cu la tin g absolute changes because of the presence of both positive and negative responses. T his m easure is only a rough approxim ation because it is not possible to calculate the standard errors. FEDERAL RESERVE BANK OF ST. LOUIS Bretton Woods than under the floating exchange rate period. Perhaps the gold standard was able to endure the greater shocks that it faced because o f both greater price flexibility and greater factor mobil ity before W orld W ar I. Alternatively, the gold standard was more durable than Bretton Woods because before W orld W ar I, suffrage was limited, central banks were often privately owned and, before Keynes, there was less understanding of the link between monetary policy and the level of economic activity. Hence there was less of an incentive for the monetary authorities to pursue full employment policies, which would threaten adherence to convertibility. In addition, the Bretton Woods regime was both more stable and seemingly more flexible than the floating exchange rate regime and yet more fragile. This suggests that its collapse is attribut able less to outside shocks to the environment or the structure of the Group of Seven economy and more to flaws in the design of the regime. Summary The performance of alternative international monetary regimes suggests that the Bretton Woods convertible regime (1959-70) was the most stable, followed by the floating exchange rate and the classical gold standard regimes. The stability of forecast errors to both inflation and growth paralleled that of the e \ post data. Limited inflation persistence—evidence for credibility o f the nominal anchor—was lowest during the classical gold standard. Though con siderably higher than under the gold standard, persistence was less under Bretton Woods than under the float. Under Bretton Woods the nomi nal anchor of the U. S. commitment to peg the price of its currency to gold was apparently still effective. Finally, supply shocks were greater and less symmetric, and demand shocks were greater under the classical gold standard than under the post-World W ar II regimes. A more rapid response of both prices and output to these shocks also occurred under the gold standard. The question still remains why some fixed exchange rate regimes endured longer than others 56As m entioned above, according to the overidentifying restrictions of the Bayoum i-E ichengreen-Blanchard-Q uah approach, supply shocks should have produced a negative response in prices, not the positive one shown here for the pre-W orld W ar II periods. 159 Table 5 The Mean Lag of Adjustment to Demand and Supply Shocks, G7 Countries 1880-19891 Gold Standard Interw ar Post World W ar II Bretton W oods Floating Exchange United States a) b) c) d) 2.77 1.65 1.61 2.24 3.79 2.89 1.65 2.31 2.16 2.68 2.21 2.55 2.29 1.85 2.12 3.06 3.65 6.48 4.64 5.52 United Kingdom a) b) c) d) 2.52 2.13 1.88 3.14 2.01 1.97 1.48 3.06 3.38 4.93 1.96 3.04 4.24 3.06 2.39 2.23 3.43 4.46 2.62 2.44 G erm any a) b) c) d) 2.74 2.23 1.87 2.03 3.08 2.86 3.09 3.50 2.70 2.67 1.86 1.92 3.48 3.13 1.80 2.85 3.53 6.06 2.22 4.14 Canada a) b) d) 1.76 1.75 1.20 2.51 2.53 1.87 1.36 2.62 2.55 2.75 2.32 2.87 3.12 1.79 1.52 2.98 2.40 2.67 2.26 2.98 France a) b) c) d) 1.79 1.70 1.85 2.85 3.84 2.60 2.05 4.80 2.19 1.83 2.54 1.92 3.50 3.06 2.19 3.50 4.97 5.53 3.79 6.02 Japan a) b) d) 2.67 1.85 1.96 1.52 4.16 3.87 3.41 3.13 2.94 2.70 2.97 4.69 3.14 1.99 1.96 2.81 4.55 9.42 3.66 8.34 Italy a) b) c) d) 2.13 1.71 2.43 1.66 2.50 1.60 1.57 2.19 3.51 4.20 6.20 6.68 1.86 1.43 1.63 1.66 2.99 5.57 1.94 3.02 Average a) b) 2.34 1.86 1.83 2.27 3.13 2.52 2.09 3.09 2.77 3.11 2.87 3.38 3.09 2.33 1.94 2.73 3.63 5.74 3.02 4.63 2.19 1.38 1.78 2.10 3.61 2.68 2.63 3.35 3.27 6.00 4.18 4.74 4.59 5.14 6.03 5.11 5.16 11.83 6.14 11.26 o) c) c) d) G 7 aggregate a) b) o) d) a) b) c) d) Effect Effect Effect Effect of of of of tem porary shock on output tem porary shock on prices perm anent shock on output perm anent shock on prices 1the mean lag of adjustm ent is calculated as: D i|ACj| / 2 |Ac,| for i = 1 to 40 w here c, is th e value of the im pulse response function in period i. MARCH/APRIL 1993 160 or why the world periodically shifted between fixed and flexible rates. The durability of the gold standard may be due to greater price flexi bility and factor mobility before W orld W ar I that allowed the world economy to respond to shocks more rapidly. It also may be due to the absence of discretionary monetary policies dedi cated to maintaining full employment. But the fragility of the most stable regime, Bretton Woods, in the face o f mild shocks, suggests that an understanding of its demise requires a closer look at the history, institutions and rules of behavior of alternative monetary regimes. THE GOLD STA ND AR D , BR ETT O N W O O D S, A N D THE EMS AS COM M ITM ENT MECHANISMS Perhaps the answer to the foregoing questions concerning regime performance and durability may be linked to the commitment technology of the regime. In this section I argue that the gold standard rule of convertibility was a credible commitment mechanism that was crucial to its success and that the absence of such a mech anism underlies the failure of the Bretton Woods variant. The EMS, though not anchored to gold convertibility, may have been successful for sev eral years because it embodied a commitment technology reminiscent of the gold standard. However, like Bretton Woods, it was subject in September 1992 to intolerable strains because the commitment mechanism proved to be not credible for many of the members. Under the classical gold standard, the monetary authorities committed themselves to fix the prices of their currencies in terms of a fixed weight of gold and to buy and sell gold freely in unlimited amounts. The pledge to fix the price of a coun try’s currency in terms of gold represents the basic rule of the gold standard. The fixed price of domestic currency in terms of gold provided a nominal anchor to the international monetary system. Under the Bretton Woods system only the United States fixed the price of its currency in terms of gold. All other convertible currencies were pegged to the dollar. Also, under Bretton Woods, free convertibility of gold into dollars was limited. Thus Bretton Woods was a weak variant of the gold standard. Although the Bretton Woods system in its convertible phase (1959-71) was the most stable monetary regime of the past 57See Kydland and Prescott (1977) and Barro and Gordon (1983). FEDERAL RESERVE BANK OF ST. LOUIS century, it was short lived. It collapsed both because of fatal flaws in its design (the adjusta ble peg in the face of improved capital mobility and the confidence problem associated with the gold dollar standard) and the lack of commit ment by the United States to the gold standard convertibilty rule. The EMS, although not based on gold, incor porated many of the features of the Bretton Woods adjustable peg system. Its success in promoting the convergence of inflation rates among its members in the 1980s has been linked to the presence of an effective commit ment mechanism—the adherence by the German central bank to price stability and the willing ness o f other members to tie their currencies to the German mark. However, like Bretton Woods, it suffered serious stress in September 1992 in the face of massive shocks because of the basic incompatibility o f pegged exchange rates, capital mobility and policy autonomy. Both the center country and the members were unwilling to commit to a common policy. An overview of the three regimes as embodying the operation of credible monetary rules follows. The Gold Standard as a Commitment Mechanism In the recent literature on the time inconsistency of optimal government policy, the absence o f a credible commitment mechanism leads govern ments, in pursuing stabilization policies, to pro duce an inflationary outcome.5 In a closed 7 economy environment, once the monetary authority has announced a given rate of monetary growth, which the public expects it to validate, the au thority then has an incentive to create a mone tary surprise to either reduce unemployment or capture seigniorage revenue. The public, with rational expectations, will come to anticipate the authorities’ perfidy, leading to an inflationary equilibrium. A credible precommitment mech anism, by preventing the government from cheating, can preserve long-run price stability. The gold standard rule of maintaining a fixed price of gold can be viewed as such a mechanism. The gold standard rule can be viewed as a form of contingent rule or a rule with escape clauses.5 8 The monetary authority maintains the standard— that is, keeps the price of the currency in terms of gold fixed—except in the event of a well-understood 58See Grossm an and Van H uyck (1988), DeKock and G rilli (1989), and Bordo and Kydland (1992). 161 emergency, such as a major war or a financial cri sis. In wartime it may suspend gold convertibility and issue paper money to finance its expenditures, and it can sell debt issues in terms of the nominal value of its currency on the understanding that debt will eventually be paid o ff in gold. The rule is con tingent in the sense that the public understands that the suspension will last only for the duration of the wartime emergency plus some period of adjustment. It assumes that afterward the government will follow the deflationary policies necessary to resume payments at the original parity. Following such a rule will also allow the government to smooth its revenue from different sources o f finance, such as taxation, borrowing and seigniorage.5 9 historical evolution itself of the gold standard. Gold was accepted as money because o f its intrinsic value and desirable properties. Paper claims, developed to economize on the scarce resources tied up in a commodity money, became acceptable only because they were convertible into gold. An alternative commitment mechanism was to guarantee gold convertibility in the con stitution. This was the case for example in Sweden before 1914, when laws pertaining to the gold standard could be changed only by two identical parliamentary decisions with an elec tion in between.6 Convertibility was also 2 enshrined in the laws of a number of gold standard central banks.6 3 According to Bordo and Kydland (1992), the gold standard contingent rule worked success fully for three core countries of the classical gold standard: the United Kingdom, the United States, and France. In all these countries the monetary authorities adhered faithfully to the fixed price of gold except during major wars. During the Napoleonic W ar and W orld W ar I for England, the Civil W ar for the United States, and the Franco-Prussian War for France, specie payments were suspended, and paper money and debt were issued. But in each case, after the wartime emergency had passed, policies leading to resumption were adopted.6 Indeed, 0 successful adherence to the rule may have ena bled the belligerents to obtain access to debt finance more easily in subsequent wars. Other countries, such as Italy, which did not continu ously maintain gold convertibility, nevertheless adopted policies consistent with long-run con vertibility.6 1 The gold standard originally evolved as a domes tic commitment mechanism, but its enduring fame is as an international rule. The classical gold standard emerged as a true international standard by 1880 following the switch by the majority of countries from bimetallism, silver monometalism and paper to gold as the basis of their currencies.6 As an international standard, 4 the key rule was maintenance o f gold converti bility at the established par. Maintenance of a fixed price of gold by its adherents in turn ensured fixed exchange rates. Recent evidence suggests that, indeed, exchange rates through out the 1880-1914 period were characterized by a high degree o f fixity in the principal coun tries. Although exchange rates frequently deviated from par, violations of the gold points and de valuations were rare.6 5 The gold standard rule may also have been enforced by reputational considerations. Longrun adherence to the rule was based on the 59See Lucas and Stokey (1983) and M ankiw (1987). 60A case study com paring British and French finances d u r ing the N apoleonic W ars shows that Britain was able to finance its w artim e expenditures by a com bination of taxes, debt and paper m oney issue— to sm ooth revenue; whereas France had to rely prim arily on taxation. France had to rely on a less e fficient m ix of finance than Britain because she had used up her cred ib ility by defaulting on outstanding debt at the end of the Am erican R evolutionary W ar and by hyperinflating during the Revolution. Napoleon ultim ately returned France to the bim etallic standard in 1803 as part of a policy to restore fiscal probity, but because of the previous loss of reputation France was unable to take advantage of the co n tin ge n t aspect of the b im etallic standard rule. See Bordo and W hite (1991). According to game theory literature, for an international monetary arrangement to be effec tive both between countries and within them, a time-consistent credible commitment mechanism is required. Adherence to the gold convertibility rule provided such a mechanism. In addition to the who present evidence of expected appreciation of the greenback d u ring the Am erican C ivil W ar based on a negative interest diffe re n tia l between bonds that were paid in greenbacks and those paid in gold. G iovannini (1992) finds that the variation of both exchange rates and short-term interest rates varied w ithin the lim its set by the gold points in the 1899-1909 period consistent with m arket ag e nts’ expectations of a credible com m itm ent by the four " c o re ” countries to the gold standard rule in the sense of this paper. 62See Jonung (1984). 63See G iovannini (1993). 64See Eichengreen (1985). 65See O fficer (1986) and Eichengreen (1985). 61The behavior of asset prices (exchange rates and interest rates) suggests that m arket agents viewed the com m itm ent to gold as credible. See Roll (1972) and C alom iris (1988), M ARCH/APRIL 1993 162 reputation of the domestic gold standard and con stitutional provisions that ensured domestic com mitment, adherence to the international goldstandard rule may have been enforced by other mechanisms. These include improved access to international capital markets, the operation of the rules o f the game, and the hegemonic power of England. Support for the international gold standard likely grew because it provided improved access to the international capital markets of the core countries. Countries were eager to adhere to the standard because they believed that gold convertibility would be a signal to creditors of sound government finance and the future abil ity to service debt.6 6 This was the case both for developing countries seeking access to long-term capital, such as Austria-Hungary and Latin America, and for countries seeking short-term loans, such as Japan, which financed the Russo-Japanese war of 1905-06 with foreign loans seven years after joining the gold standard.6 Once on the gold 7 standard, these countries feared the consequences of suspension.6 That England, the most success 8 ful country of the nineteenth century, as well as other progressive countries were on the gold standard was probably a powerful argument for joining.6 9 The operation of the rules of the game, whereby the monetary authorities were sup posed to alter the discount rate to speed up the adjustment to a change in external balance, may also have been an important part o f the com mitment mechanism to the international gold standard rule. To the extent the rules were fol lowed and adjustment facilitated, the commit ment to convertibility was strengthened and conditions conducive to abandonment were lessened. Evidence on the operation of the rules of the game questions their validity. Bloomfield (1959), in a classic study, showed that, with the principal exception of England, the rules were frequently violated in the sense that discount rates were 66A case study of C anada during the G reat Depression pro vides evidence for the im portance of the credible com m it m ent m echanism of adherence to gold. C anada suspended the gold standard in 1929 but did not allow the C anadian dollar to depreciate nor the price level to rise for tw o years. C anada did not take advantage of the suspen sion to em erge from the depression because of concern for its cred ib ility with foreign lenders. See Bordo and Red ish (1990). 67See Y eager (1984), Fishlow (1989) and Hayashi (1989). FEDERAL RESERVE BANK OF ST. LOUIS not always changed in the required direction (or by sufficient amounts) and in the sense that changes in domestic credit were often nega tively correlated with changes in gold reserves. In addition, a number of countries used gold devices—practices to prevent gold outflows. For the major countries, however (at least before 1914) such policies w ere not used exten sively enough to threaten the convertibility to gold—evidence o f commitment to the rule.7 0 Moreover, as McKinnon (1992) argues, to the extent that monetary authorities followed Bagehot’s rule and prevented a financial crisis while seemingly violating the rules of the game, the commitment to the gold standard in the long run may have been strengthened. An additional enforcement mechanism for the international gold standard rule may have been the hegemonic power of England, the most important gold standard country.7 A persistent 1 theme in the literature on the international gold standard is that the classical gold standard of 1880 to 1914 was a British-managed standard.7 2 Because London was the center for the world's principal gold, commodities and capital markets, because of the extensive outstanding sterlingdenominated assets, and because many countries substituted sterling for gold as an international reserve currency, some argue that the Bank of England, by manipulating its bank rate, could attract whatever gold it needed and, further more, that other central banks would adjust their discount rates accordingly. Thus the Bank of England could exert a powerful influence on the money supplies and price levels of other gold standard countries. The evidence suggests that the Bank did have some influence on other European central banks.7 3 Eichengreen (1987) treats the Bank of England as one engaged in a leadership role in a Stackelberg strategic game with other central banks as fol lowers. The other central banks accepted a passive role because they benefited from using sterling as a reserve asset. According to this interpretation, the gold standard rule may 68See Eichengreen (1989a) and Fishlow (1987 and 1989). 69See Friedm an (1990) and G allarotti (1991). 70See Schw artz (1984). 71See Eichengreen (1989b). 72See Bordo (1984). 73See Lindert (1969). 163 have been enforced by the Bank of England.7 4 Thus the monetary authorities of many coun tries may have been constrained from following independent discretionary policies that would have threatened adherence to the gold standard rule. Indeed, according to Giovannini (1989), the gold standard was an asymmetric system. England was the center country. It used its monetary policy (bank rate) to maintain gold converti bility. Other countries accepted the dictates of fixed parities and allowed their money sup plies to respond passively. His regressions support this view—the French and German central banks adapted their domestic policies to external conditions, whereas the British did not. The benefits to England as leader o f the gold standard—from seigniorage earned on foreignheld sterling balances to returns to financial institutions generated by its central position in the gold standard and to access to international capital markets in wartime—were substantial enough to make the costs of not following the rule extremely high. The classical gold standard ended in the face of the massive shocks of World W ar I.7 The 5 gold exchange standard, which prevailed for only a few years from the mid-1920s to the Great Depression, was an attempt to restore the beneficial features of the classical gold standard while allowing a greater role for domestic stabil ization policy. This in turn created a growing conflict between adherence to the rule and dis cretion. It also attempted to economize on gold reserves by restricting its use to central banks and by encouraging the use of foreign exchange as a substitute. As is well known, the gold ex change standard suffered from a number of fatal flaws.7 These include the use of two 6 reserve currencies (the pound and the dollar), the absence of leadership by a hegemonic power, the failure o f cooperation between the 74A ccording to Eichengreen (1989a), the Bank of E n g la n d ’s ability to ensure co n ve rtib ility was aided by the coopera tion of other central banks. In addition, as m entioned above, belief based on past perform ance that England attached highest p riority to co n ve rtib ility encouraged stabilizing private capital m ovem ents in tim es of threats to convertibility, such as in 1890 and 1907. key members (England, France and the United States), and the unwillingness of its two strongest members, the United States and France, to fol low the rules of the game. Instead they exerted deflationary pressure on the rest o f the world by persistent sterilization of balance-of-payment surpluses. The gold exchange standard collapsed, but according to Friedman and Schwartz, Temin, and Eichengreen, not before transmitting defla tion and depression across the world.7 7 The Bretton W oods International Monetary System The planning that led to Bretton Woods aimed to prevent the chaos of the interwar period.7 8 The perceived ills to be prevented included (1) floating exchange rates that were condemned as subject to destabilizing speculation; (2) a gold exchange standard that was vulnerable to prob lems of adjustment, liquidity and confidence, which enforced the international transmission of deflation in the early 1930s; and (3) the resort to beggar-thy-neighbor devaluations, trade restrictions, exchange controls and bilater alism after 1933. To prevent these ills, the case for an adjustable peg system was made by Keynes, White, Nurkse and others.7 The new 9 system would combine the favorable features of the fixed exchange rate gold standard—stability of exchange rates—and of the flexible exchange rate standard—monetary and fiscal independence. Both Keynes, leading the British negotiating team at Bretton Woods, and White, leading the Ameri can team at Bretton Woods, planned an adjustable peg system to be coordinated by an international monetary agency. The Keynes plan gave the Inter national Currency Union substantially more reserves and power than the United Nations Stabili zation Fund proposed by White, but both institu tions would have had considerable control over the domestic financial policy of the members. The British plan contained more domestic policy autonomy than did the U.S. plan, whereas the 76See K indleberger (1973), Tem in (1989), and Eichengreen (1992c). 77See Friedm an and Schwartz (1963), Tem in (1989) and Eichengreen (1992c). 78T his section draw s heavily on Bordo (1992). 79See Bordo (1993). 75The standard deviations of both supply and dem and shocks during W orld W ar I for the countries for which we have continuous data were tw o to three tim es as great as du ring the classical gold standard. See appendix table 1 and figure 1. M ARCH/APRIL 1993 164 American plan put more emphasis on exchange rate stability. Neither architect was in favor o f a rule-based system.8 The British were most con 0 cerned with preventing the deflation of the 1930s, which they attributed to the constraint of the gold standard rule and to deflationary U.S. monetary pol icies. Thus they wanted an expansionary system. The American plan was closer to the gold standard rule in that it stressed the fixity of exchange rates. It did not explicitly mention the importance of rules as a credible commitment mechanism, but there were to be strict regula tions on the linkage between UNIT AS (the pro posed international reserve account) and gold. Members, in the event o f a fundamental dis equilibrium, could change their parities only with approval from a three-quarters majority of all members of the fund.8 1 The Articles of Agreement of the International Monetary Fund incorporated elements of both the Keynes and White plans, although in the end, U.S. concerns predominated.8 The main points of the 2 articles were: the creation of the par value system; multilateral payments; the use of the fund's resources; its powers; and its organization. The Par Value System Article IV defined the numeraire o f the interna tional monetary system as either gold or the U.S. dollar of the weight and fineness on July 1, 1944. All members were urged to declare a par value and maintain it within a 1 percent margin on either side of parity. Parity could be changed in the event of a fundamental payments disequili brium at the decision o f the member, after con sultation with other fund members. The fund would not, however, reject the change if it was not more than 10 percent; if the change was more than 10 percent, the fund would decide 80ln the sense of a com m itm ent m echanism to prevent the tim e consistency problem . A ccording to M eltzer (1988) and M oggridge (1986), Keynes had a strong preference for rules over discretion, interpreting rules in the traditional sense. 8,See G iovannini (1993). 82At the same tim e as the A rticles of A greem ent for the International M onetary Fund were signed, the International Bank for R econstruction and D evelopm ent (the W orld Bank) was established. The C harter of the International Trade O rganization (ITO) was drafted and signed in 1947 but never ratified. It was succeeded by the General A gree m ent for T ariffs and Trade (GATT) o rigina lly negotiated in G eneva in 1947 as an interim institution until the ITO cam e into force. FEDERAL RESERVE BANK OF ST. LOUIS within 72 hours. Unauthorized changes in the exchange rate could make members ineligible to use the fund’s resources, and if a member continued to make unauthorized changes, it could be expelled from the fund. A uniform change in par value of all currencies (in terms o f gold) required approval by a majority of all voting fund members and also had to be approved by every member with 10 percent or more of the total quota. Multilateral Payments Members were supposed to make their curren cies convertible for current account transactions (Art. VIII), but capital controls were permitted (Art. VI.3). They were also to avoid discrimina tory currency and multiple curency arrange ments. Countries could avoid declaring their currencies convertible, however, by invoking Art. XIV, which allowed a three-year transition period after establishment of the fund. During the transition period, existing exchange controls could be maintained.8 3 The Fund’s Resources As under the White plan, members could obtain resources from the fund to help finance short- or medium-term payments disequilibria. The total fund, contributed by members quotas (25 percent in gold and 75 percent in currencies) was set at $8.8 billion. It could be raised every five years if the majority of members wanted to do so. The fund set a number of conditions on the use of its resources by deficit countries to prevent it from accumulating soft currencies and from depleting its holdings o f harder cur rencies.8 It also established requirements and 4 conditions for repurchase (repayment o f a loan), including giving the fund the right to decide the currency in which repurchase would be made. 83Under Art. XIV, three years after M arch 1, 1947, the IMF w ould begin reporting on the countries with existing con trols, two years later it would begin co n sulting w ith in d ivid ual m em bers and advising them on policies to restore paym ents e q uilibrium and convertibility. C ountries that did not m ake satisfactory progress w ould be censured and ultim ately asked to leave the fund. In fact, the fund always accepted the m em b e r’s reason for rem aining under Art. XIV. 84M em bers could draw on th e ir quotas w ithout condition. Beyond that, referred to later as the cred it tranches, although not spelled out in the articles, increasingly more exacting conditions were required. 165 In the case o f countries prone to running large surpluses, the scarce currency clause (Art. VII) would come into play. If the fund’s holdings of a currency were insufficient to satisfy the demand for it by other members, it could declare it scarce and then urge members to ration its use by dis criminatory exchange controls. the fixed price o f gold at $35 per ounce, which the U.S. was to maintain, represented the nomi nal anchor of the system. Members were required to maintain parity o f their exchange rates in terms of dollars (or gold). Also, like the gold standard, it was a rule with an escape clause. Members at their initiative could alter their par ities in the event of a fundamental disequilibrium. The P ow ers o f the Fund The architects never spelled out exactly how the system was supposed to work. Subsequent writers, however, have suggested a number of salient features.8 First, currencies were treated 6 as equal in the articles. This meant that in theory each country was required to maintain its par value by intervening in the currency o f every other country—a practice that would have worked at cross purposes. In fact, because the United States was the only country that pegged its cur rency in terms o f gold (bought and sold gold), all other countries would fix their parities in terms of dollars and would intervene to monitor their exchange rates within 1 percent of parity with the dollar. The fund had considerably less discretionary power over the domestic policies o f its members than either of the architects wanted, but it still had power to influence the international mone tary system strongly. These powers included its authority to approve or reject changes in parity; the use of multiple exchange rates and other discriminatory practices; the conditionality implicit in members’ access to the credit tranches of their quotas, which was made explicit by 1952; its authority to declare currencies scarce; its authority to declare members ineligible to use its resources (used against France in 1948 following an unap proved devaluation); and its ultimate authority to expel members.8 The fund also had consider 5 able power as the premier international mone tary organization in consulting and cooperating with national and other international monetary authorities. Organization The fund was to be governed by a board of governors appointed by the members. The board would make the major policy decisions, such as approving a change in parity. Operations of the fund were to be directed by executive directors appointed by the members and a managing direc tor selected by the executive directors. Major changes such as a uniform change in the par value of all currencies or the second amend ment to the articles, which created the special drawing right (SDR), would require a majority vote by the members. The number o f votes in turn was tied to the size o f each member’s quota, which was determined by its economic size. Though the articles could not be interpreted strictly as a return to the gold standard rule of the fixed price o f gold with free convertibility, 85See Diz (1984) for a discussion of co n d itio na lity im p licit in m em bers’ access to the cred it tranches of th e ir quotas. 86See, for exam ple, Tew (1988), Scam m ell (1976) and Y eager (1976). For W illiam son (1985b) it was a com pre hensive set o f rules for assigning m acroeconom ic pol- Second, countries would use their international reserves or draw resources from the fund to finance payments deficits. In the case o f surpluses, coun tries would temporarily build up reserves or re purchase their currencies from the fund. In the event of medium-term disequilibria, they would use monetary and fiscal policy to alter aggregate demand. In the event of a fundamental disequil ibrium, which was never defined but presumably reflected either some permanent structural shock or sustained inflation, a member was supposed to alter parity by an amount sufficient to restore exter nal equilibrium. Third, capital controls were permitted to prevent destabilizing speculation from forcing members to alter their parities prematurely or unintentionally. THE H ISTO R Y OF BR ETT O N W O O D S: PR E -C O N VE R T IB ILIT Y 1946-58 The international monetary system that began after W orld W ar II was far different from the system that the architects of Bretton Woods envisioned. The transition period from war to icies— exchange rates to m edium -run external balance, m onetary and fiscal policy to short-run internal balance and international reserves—to provide a b u ffer to allow short-run departures from external balance. M ARCH/APRIL 1993 166 peace was much longer and more painful than anticipated. Full convertibility of the major industrial countries was not achieved until the end of 1958, although the system had started functioning normally by 1955. Tw o interrelated problems dominated the first postwar decade: bilateralism and the dollar shortage. Bilateralism The legacy of W orld W ar II for virtually every country except the United States was one of pervasive exchange controls and controls on trade. No major currency except the dollar was convertible.8 Under Art. XIV of the Bretton 7 Woods agreement, countries could continue to use exchange controls for an indefinite transi tion period after the establishment of the Inter national Monetary Fund (IMF) on March 1, 1947. In conjunction with exchange controls, every country negotiated a series of bilateral pay ments agreements with each of its trading part ners. The rationale for the continued use of controls and bilateralism was a shortage of international reserves. After the war, the econo mies of Europe and Asia were devastated. To produce the exports needed to generate foreign exchange, industries required new and improved capital. There was an acute shortage of key imports, both foodstuffs to maintain living standards and raw materials and capital equip ment. Controls allocated the scarce reserves. The Dollar Shortage By the end of World W ar II, the United States held two-thirds of the world’s monetary gold stock (see figure 13). The gold avalanche in the United States in the 1930s was the consequence of both the dollar devaluation in 1934, when the Roosevelt administration raised the price of gold from $20.67 per ounce to $35.00, and capi tal flight from Europe. During the war, gold 87U nder the classical gold standard, co nvertibility meant the ab ility of a private individual to freely convert a unit of any national currency into gold at the o fficia l fixed price. A suspension of co n ve rtib ility m eant that the exchange rate between gold and national currency becam e fle xib le but the individual could still freely transact in either asset. See T riffin (1960). By the eve of W orld W ar II, co nvertibility referred to the ability of a private individual to freely make and receive paym ents in international transactions in term s of the currency of another country. Under Bretton W oods, c o n ve rtib ility m eant the freedom for individuals to engage in cu rre n t account transactions w ithout being subject to exchange controls. Tew (1988) defines this as m arket con ve rtib ility and d istinguishes it from official co nvertibility w hereby the m onetary authorities of each country freely FEDERAL RESERVE BANK OF ST. LOUIS inflows continued to finance wartime expendi tures by the allies. At the end of W orld W ar II, gold and dollar reserves in Europe and Japan were depleted. Europe ran a massive current account deficit, reflecting the demand for essen tial imports and the reduced capacity o f the export industries. The Organization for Euro pean Economic Cooperation (OEEC) deficit, aggravated by the bad winter of 1946-47, reached a high of $9 billion in 1947. The OEEC deficit equaled the amount of the U.S. current account surplus, which was large because the United States, as the only major industrial coun try operating at full capacity, supplied the needed imports. The dollar shortage was likely aggravated by overvalued official parities set by the major European industrial countries at the end of 1946.8 8 By the mid-1950s both problems had been solved. The currencies of western Europe were virtually convertible by 1955 and their current accounts were generally in surplus. The key developments in this progress were the Mar shall Plan and the European Payments Union. The Marshall Plan The Marshall Plan funneled approximately $13 billion in aid (grants and loans) to western Europe between 1948 and 1952.8 The plan required 9 the recipients to cooperate in the liberalization of trade and payments. Consequently, the OEEC was established in April 1948. It presided over the allocation of aid to members based on the size of their current account deficits. U.S. aid was to pay for essential imports and to provide international reserves. Each recipient govern ment provided matching funds in local currency to be used for investment in the productive capacity o f industry, agriculture and infrastruc ture. Each country also had a delegation of U.S. administrators that advised the host govern ment on the spending o f its counterpart funds. buy and sell foreign exchange (prim arily dollars) to keep the parity fixed (within the 1 percent m argin) and the United States freely buys and sells gold to m aintain the fixed price of $35 an ounce (within the 1 percent m argin). He refers to both m arket and official co n ve rtib ility as B ret ton W oods convertibility. See also M cKinnon (1979) and Black (1987). 88See T riffin (1957). 89See M ilward (1984) and H offm an and M aier (1984). 167 Figure 13 Monetary Gold and Dollar Holdings: the United States and the Rest of the World, 1945-1971 Billions of U.S. dollars The plan encouraged the liberalization o f intraEuropean trade and payments by granting aid to countries that extended bilateral credits to other members. Finally, the European Payments Union (EPU) was established in 1950, under the auspices of the OEEC, to simplify bilateral clear ing and pave the way to multilateralism. By 1952, in part thanks to the Marshall Plan, the OEEC countries had achieved a 39 percent increase in industrial production, a doubling of exports, an increase in imports by one-third and a current account surplus.9 0 The European Payments Union and the Return to Convertibility It took 12 years from the declaration o f offi cial par values by 32 nations in December 1946 to achieve convertibility for current transactions by the major industrial countries, as specified by the Bretton Woods Articles. The Western European nations tried several schemes to facili tate the payments process before establishing the EPU in 1950.9' The EPU, established September 19, 1950, by the OEEC countries, initially was to run for two years, renewable thereafter on a yearly basis. It followed the basic principle of a commercial 90Solom on (1976). bank clearinghouse. At the end of each month, each member would clear its net debit or credit position (against all other members) with the EPU (the BIS acting as its agent). The unit of account for these clearings was the U.S. dollar. The EPU also provided extensive credit lines. The EPU was highly successful in reducing the volume o f payments transactions and provided the background for the gradual liberalization of payments so that by 1953 commercial banks were able to engage in multicurrency arbitrage.9 On 2 December 27, 1958, eight countries declared their currencies convertible for current account transactions. The movement to convertibility was aided by the devaluations of 1949. Following a specula tive run on the pound in the summer of 1949, the British, 24 hours after informing the IMF, devalued the pound by 30.5 percent. Shortly thereafter, 23 countries reduced their parities by similar magnitudes in most cases. The devaluations of 1949 were important for the Bretton Woods system for two reasons. First, they, along with the Marshall Plan aid, helped move the European countries from a current account deficit to a surplus, a movement impor tant to the eventual restoration of convertibility. Second, they revealed a basic weakness of the 92Tew (1988) and Yeager (1976). 9 Kaplan and Schleim inger (1989). 1 M ARCH/APRIL 1993 168 adjustable peg arrangement—the one-way option of speculation against parity. By allowing changes in parity only in the event of a fun damental disequilibrium, the Bretton Woods sys tem encouraged the monetary authorities to delay adjustment until they were sure it was necessary. By that time, speculators also would be sure and they would take a position from which they could not lose. If the currency is devalued, they win and if it is not, they just lose the interest (if any) on the speculative funds.9 3 The crisis associated with the 1949 sterling devaluation in turn created further resistance by monetary authorities to changes in parity, which ultimately changed the nature of the international monetary system from the adjusta ble peg intended by the Bretton Woods Articles to a fixed rate regime. Other developments in the preconvertiblity period included the decline of sterling as a reserve asset and the reduced prestige of the IMF. The IMF by intention was not equipped to deal with the postwar reconstruction problem. Although some limited drawings occurred before 1952, most of the structural balance of payments assistance in this period was provided by the Marshall Plan and other U.S. assistance, including the Anglo-American Loan of 1946. The consequence of this development is that other institutions such as the BIS, the agent for the EPU, emerged as competing sources of interna tional monetary authority.9 4 The fund’s prestige was dealt a severe blow by three events in the preconvertibility period. The first event was the French devaluation of January 1948, which created a multiple exchange rate system. The fund censured France for creating broken cross rates between the dollar and the pound. France was denied access to the fund's resources until 1952. France ended the broken cross rates in October 1948 and adopted a unified rate in the devaluation of 1949. Since France had access to the Marshall Plan, the fund’s actions had little effect. The second event was the sterling devaluation of September 1949, when the fund, instead of being actively involved in consultation, was given 24 hours perfunctory notice. The third event was the decision by Canada to float its currency in September 1950. Though the fund was highly critical of the action, it was unable to prevent it. The Canadian dollar floated successfully until 1961. 93See Friedm an (1953). 94See M undell (1969). FEDERAL RESERVE BANK OF ST. LOUIS Finally, the fund’s resources were inadequate to solve the emerging liquidity problem o f the 1960s. The difference between the required growth of international reserves (to finance the growth of real output and trade and to avoid deflation) and the growth in the world’s mone tary gold stock was met largely by an increase in official holdings of U.S. dollars resulting from growing U.S. balance-of-payments deficits. By the time full convertibility was achieved, the U.S. dollar was serving the buffer function intended by the Bretton Woods Articles for the fund’s resources.9 5 THE H IST O R Y OF B R ET T O N W O O D S: THE H EYD AY OF BR ETT O N W O O D S 1959-1967 With current account convertibility established by the western European industrial nations at the end of December 1958, the full-blown Bretton Woods system was in operation. Each member intervened in the foreign exchange market, either buying or selling dollars, to maintain its parity within the prescribed 1 percent margins. The U.S. Treasury in turn pegged the price of the dollar at $35 per ounce by freely buying and selling gold. Thus each currency was anchored to the dollar and indirectly to gold. Triangular arbitrage kept all cross rates within a band of 2 percent on either side of parity. Through much of this period, capital controls prevailed in most countries except the United States in one form or another, although by the mid-1960s their use declined while increasing in the United States. The system that operated in the next decade turned out to be quite different from what the architects had in mind. First, instead of a system of equal currencies, it evolved into a variant of the gold exchange standard—the gold-dollar sys tem. Initially, it was a gold exchange standard with two key currencies, the dollar and the pound. But the role of the pound as key cur rency declined steadily throughout the 1960s. Concurrently with the decline of sterling was the rise in the dollar as a key currency. Use of the dollar as both a private and official interna tional money increased dramatically in the 1950s and continued into the 1960s. With full convertibility, the dollar’s fundamental role as 95See Mundell (1969). 169 intervention currency led to its use as interna tional reserves. This was aided by stable and low monetary growth and relatively low infla tion (before 1965). See figure 1 and table 1. The gold exchange standard evolved in the post-World War II period for the same reasons it did in the 1920s—to economize on non-interestbearing gold reserves. By the late 1950s, the growth of the world's monetary gold stock was insuffi cient to finance the growth of world real output and trade.9 The other intended source of inter 6 national liquidity—the resources of the fund— was also insufficient. The second important difference between the convertible Bretton Woods system and the inten tions of the Bretton Woods Articles was the evo lution of the adjustable peg system into a virtual fixed exchange rate system. Between 1949 and 1967, very few changes in parities of the Group of Ten countries occurred.9 The only excep 7 tions were the Canadian float in 1950, devalua tions by France in 1957 and 1958, and minor revaluations by Germany and the Netherlands in 1961. The adjustable peg system became less adjustable because the monetary authorities, based on the 1949 experience, were unwilling to accept the risks associated with discrete changes in parities—loss of prestige, the likeli hood that others would follow and the pressure o f speculative capital flows if even a hint of a change in parity were present. As the system evolved into a fixed exchange rate gold dollar standard, the three key prob lems of the interwar system reemerged: adjust ment, liquidity and confidence. These problems dominated academic and policy discussions dur ing the period. The Adjustment Problem The adjustment issue focused on how to achieve it in a world with capital controls, fixed exchange rates and domestic policy autonomy. Various pol icy measures were proposed to aid adjustment, including income policies, rescue packages, capi tal and trade controls, a mix o f monetary and fiscal policy, and the injection of new liquidity. Of particular interest during the period was asymmetry in adjustment between deficit countries like the United Kingdom and surplus countries like Germany and between the United States as the reserve currency country and rest of the world. Both the United Kingdom and Germany ran the gauntlet between concern over external convertibility and domestic stability. The United Kingdom alternated between expansionary pol icy that led to balance-of-payments deficits and austerity. Germany alternated between a balanceof-payments surplus that led to inflation and austerity. The United States had an official settlements balance-of-payments deficit in 1958 that per sisted, with the notable exception o f 1968-69, until the end o f Bretton Woods. See figure 14. With the exception of 1959, however, the United States had a current account surplus until 1970. The balance-of-payments deficit under Bretton Woods arose because capital out flows exceeded the current account surplus. In the early postwar years, the outflow consisted largely of foreign aid. By the end of the 1950s, private long-term investment abroad (mainly direct investment) exceeded military expendi tures abroad and other official transfers.9 8 The balance-of-payments deficit was perceived as a problem by the U.S. monetary authorities because of its effect on confidence. As official dollar liabilities held abroad mounted with suc cessive deficits, the likelihood increased that these dollars would be converted into gold and eventually the U.S. monetary gold stock would reach a point low enough to trigger a run. Indeed the U.S. monetary gold stock by 1959 equalled total external dollar liabilities and the rest of the world's monetary gold stock exceeded that of the United States. See figure 13. By 1964 official dollar liabilities held by foreign mone tary authorities exceeded the U.S. monetary gold stock. A second reason the balance-of-payments deficit was perceived as a problem was the dol lar’s role in providing liquidity to the rest of the world. Elimination of the U.S. deficit would cre ate a worldwide liquidity shortage. 96See T riffin (1960) and G ilbert (1968). 97The G roup of Ten countries were Belgium , Canada, France, W est G erm any, Italy, Japan, the N etherlands, Sweden, United Kingdom , and the United States. Sw itzer land was an associate member. 98See Eichengreen (1991c). M ARCH/APRIL 1993 170 Figure 14 Balance of Payments: United States, 1950-1971 Millions of U.S. dollars For the Europeans, the U.S. balance-ofpayments deficit was a problem for different reasons. First, as the reserve currency country, the United States did not have to adjust its domestic economy to the balance of payments. As a matter of routine, the Federal Reserve automatically sterilized dollar outflows. The asymmetry in adjustment was resented. The Germans viewed the situation as the United States exporting inflation to surplus countries through its deficits. Their remedy was for the United States (and the United Kingdom) to pur sue contractionary monetary and fiscal policy." In fact, U.S. inflation was less (on a GNPweighted average basis) than that of the rest of the Group of Seven countries before 1968. See figures 1 and 15. The French resented U.S. financial dominance and the seigniorage they believed the United States earned on its out standing liabilities. Acting on this perception, the French in 1965 began to systematically con vert outstanding dollar liabilities into gold. The French solution to the dollar problem was to double the price of gold—the amount by which " S e e E m m inger (1967). ’ “ See Rueff (1967). 101See Despres, K indleberger and Salant (1966). FEDERAL RESERVE BANK OF ST. LOUIS the real price of gold had declined since 1934. The capital gains earned on the revaluation of the world's monetary gold reserves would be sufficient to retire the outstanding dollar (and sterling) balances. Once the United States returned to balance-of-payments equilibrium, the world could return to a fully functioning classical gold standard.1 0 0 Some economists argued that the U.S. balanceof-payments deficit was not really a problem. The rest of the world held dollars voluntarily because o f their valuable service flow; the deficit was demand determined.1 1 0 The policy response of the U.S. monetary authorities was fourfold: to impose controls on capital exports; to institute measures to improve the balance of trade; to alter the monetary fis cal policy mix; and to employ measures to stem the conversion of outstanding dollars into gold. During this period, various solutions to the U.S. adjustment problem were proposed: provi sion of an alternative international reserve 171 F ig u re 15 Inflation Rates in the United States, G7 and G7 Excluding the United States, 1951-1973 Percent media to increase world liquidity; an increase in the price of gold, either unilaterally, which would devalue the dollar against other currencies, or by a uniform change in all parities as under Art. IV; and increased exchange rate flexibility.1 2 0 The U.S. balance-of-payments policies were in the main ineffective.1 3 As long as the United 0 States maintained relatively stable prices, as it did before 1965, the system could be preserved for a number of years. The real problem was that of the gold exchange standard—a converti bility crisis was ultimately inevitable. The twin solutions advocated at the time of an increase in the price o f gold and an increase in world liquidity by creation of an artificial reserve asset would not have permanently eradicated the problem.1 4 0 ,02T he official view, w hich was stron g ly opposed to increased exchange rate fle xib ility, is in m arked contrast to the aca dem ic view, which by the end of the decade was solidly in favor of increased fle xib ility, as evident at the fam ous Burq enstock C onference. See Halm (1970) and also Johnson (1972a). 103See M eltzer (1991). 104Even at a higher gold price, w orld gold production w ould e ventually be inadequate to produce long-run price sta b il ity. In the long-run, w hen account is taken of gold as a durable exhaustible resource, deflation is inevitable. See The Liquidity P ro b lem The liquidity problem, posed by Robert Triffin and others, evolved from a shortfall of mone tary gold beginning in the late 1950s. The real price o f gold had been falling since W orld War I I and would eventually reduce world gold production. This happened in the early 1950s but was offset by new sources o f production. Gold production declined again in 1966. More over, the falling real price would stimulate pri vate demand for gold and it seemed unlikely that Russian gold sales would make up much of the shortfall. The prospect o f the world mone tary gold stock growing enough to finance the growth o f world real output and the value of trade (without deflation) seemed dim. As can clearly be seen in figure 16 for the Group of Bordo and Ellson (1985). M oreover, an increase in w orld liq u id ity by an a rtificial reserve asset, if convertible into gold, would not rem ove the basic co n ve rtib ility problem . See M cKinnon (1988). Finally as Tow nsend (1977), Salant (1983) and Buiter (1989) point out, the gold exchange standard as a type of com m odity stabilization schem e is bound to collapse in the face of unforeseen shocks. See G arber (1993). A ccording to M eltzer (1991), however, a 50 percent gold revaluation w ould have succeeded in preserv ing the Bretton W oods system well into the 1970s had the United States not follow ed an inflationary policy in the late 1960s. M ARCH/APRIL 1993 172 Figure 16 The Growth of the Monetary Gold Stock, the Growth in International Reserves and the Growth of the Volume of Real Trade and Real Income, G 7 ,1951-1973 Percent opened in 1958 between the growth of output and the volume of trade and the growth of Group of Seven gold reserves. As can be seen in figure 17, the shortfall for the Group of Seven countries, excluding the U nited States, was made up by a drain on the U.S. monetary gold reserves until 1966. As Triffin (1960) pointed out, dollars supplied by the U.S. deficit could not be a permanent solution to the impending gold shortage because with continuous deficits, U.S. monetary gold reserves would decline both absolutely and rela tively to outstanding dollar liabilities until an eventual convertibility crisis. Should the U.S. monetary authorities close the deficit before such a crisis, however, it would create a mas sive shortage of international liquidity and the prospect of world deflation. New sources of liquidity were required, answered by the crea tion of the special drawing rights in 1967. By the time SDRs were injected into the system in 1970, however, they exacerbated worldwide inflation. 105A lthough according to M eltzer (1991), there is little evi dence in asset m arkets through the 1960s of a growing loss of confidence in the dollar. Real interest rates did not FEDERAL RESERVE BANK OF ST. LOUIS The perceived key problem of the convertible Bretton Woods period was the confidence crisis for the dollar.1 5 As argued by Triffin (1960), 0 Kenen (1960) and Gilbert (1968), the gold-dollar system that evolved after 1959 was bound to be dynamically unstable if the growth of the world monetary gold stock was insufficient to finance the growth of world output and trade and to prevent the U.S. monetary gold stock from declining relative to outstanding U.S. dollar lia bilities. The pressure on the U.S. monetary gold stock would continue, as growth o f the world monetary gold stock declined relative to the growth o f world output and trade and as the world substituted dollars for gold, until it trig gered a confidence crisis that led to the collapse of the system, as occurred in 1931. At the same time, however, as fears over U.S. gold converti bility threatened the dynamic stability o f the Bretton Woods system, gold still served two positive roles. Gold was the numeraire of the system; all rise sign ifica n tly relative to trade w eighted real interest rates. Nor did the gold and foreign exchange m arkets sug gest a flig h t from the dollar. 173 Figure 17 The Growth of the Monetary Gold Stock, the Growth in International Reserves and the Growth of the Volume of Real Trade and Real Income, G7 Minus the United States, 1951-1973 Percent currencies were anchored to its fixed price through the U.S. commitment to peg its price. Until 1968 gold still served as backing to the U.S. dollar with a 25 percent gold reserve requirement against Federal Reserve notes; the requirement may have served as a brake on U.S. monetary expansion. The first glimpse of a confidence crisis was the gold rush of October 1960 when speculators pushed the free market price of gold on the London market up from $35.20 (the U.S. Treas ury’s buying price) to $40. See figure 18. This first significant runup in gold prices since the London gold market was reopened in 1954 was supposedly triggered by concerns over a Democratic victory in the 1960 U.S. Presidential election. U.S. monetary authorities feared that private speculation in the gold market might spill into official demands for conversion. Consequently, remedial action was taken quickly. The Treas ury supplied the Bank of England sufficient gold to restore stability, and the monetary authori ties of the Group of Ten countries agreed to refrain from buying gold above $35.20. In suc ceeding months, the London Gold Pool, which agreed to buy or sell gold to peg the price at $35 an ounce, was formed between the United States and seven other countries. The pool became official in November 1961. For the next six years, it succeeded in stabilizing the price of gold but did not prevent a steady decline in the U.S. monetary gold stock. See figure 13. In fact, though the central banks in the seven other countries supplied 40 percent of the gold required to stabilize the price of gold, they replenished their monetary gold stocks outside the pool by converting outstanding dollar balances into gold at the U.S. Treasury.1 6 0 During the period 1961-67, the United States made a series o f arrangements to protect its monetary gold reserves. These included a net work of swap arrangements with other central banks, the issue of Roosa bonds, and moral sua sion. France, however, did not go along with these efforts and began its campaign against the dollar in February 1965. The period was marked by two sets o f under 106See Meltzer (1991). M ARCH/APRIL 1993 174 Figure 18 London Gold Price Dollars per ounce lying forces that would undermine the dollar’s relationship to gold—growing gold scarcity and a rise in U.S. inflation. W orld gold production leveled o ff in the mid-1960s and even declined in 1966, while at the same time private demand soared, precipitating a drop in the world mone tary gold stock after 1966. Indeed, beginning in 1966, ,the London Gold Pool became a net seller of gold. Also, U.S. money growth accelerated in 1965, in part to finance the Vietnam War, and inflation began to rise (figures 1 and 15). The current account surplus began to deteriorate in 1964 (figure 14), as did U.S. competitiveness, mirrored in a rise in the ratio o f U.S. unit labor costs relative to trade weighted unit labor costs.1 7 The balance-of-payments deficit w or 0 sened between 1964 and 1966 but was reversed in 1966 by capital inflows triggered by tight monetary policy. After the devaluation of sterling, which the United States tried unsuccessfully to prevent, pressure mounted against the dollar via the London Gold market. From December 1967 to 107See M eltzer (1991). 108See Solom on (1976). 109See G ilbert (1968) and Johnson (1968). FEDERAL RESERVE BANK OF ST. LOUIS March 1968 the Gold Pool lost $3 billion in gold, with the U.S. share at $2.2 billion.1 8 The 0 immediate concerns o f the speculators may have been fears o f a dollar devaluation, but according to Gilbert and Johnson, it reflected the underlying gold scarcity.1 9 In the face of 0 the pressure, the Gold Pool was disbanded on March 17, 1968, and a two-tier arrangement replaced it. Henceforth, the monetary authori ties o f the Gold Pool agreed neither to sell nor to buy gold from the market. They would trans act only among themselves at the official $35 price. In addition, on March 12, 1968, the United States removed the 25 percent gold reserve requirement against Federal Reserve notes. The key consequence of these new arrangements was that gold was demonetized at the margin. The link between gold production and other market sources of gold and official reserves was cut. Moreover, in the following years, the United States put considerable pres sure on other monetary authorities to refrain from converting their dollar holdings to gold. 175 By 1968 the international monetary system had evolved very far indeed from the model of the architects of the Articles of Agreement. In reaction to both developments in financial mar kets and the confidence problem, the system had evolved into a de fa c to dollar standard. Gold convertibility, however, still played a role. Though the major industrial countries tacitly agreed not to convert their outstanding dollar liabilities into U.S. monetary gold, the threat of their doing so was always present. At the same time, as the countries of continental Europe and Japan gained economic strength relative to the United States, they became more reluctant to absorb outstanding dollars. They also were reluctant to adjust their surpluses by revaluing their currencies, increasingly coming to believe that adjustment should be undertaken by the United States. The system had also developed into a de fa c to fixed exchange rate system. Unlike the classical gold standard, however, where the fixed exchange rate was the voluntary focal point for both internal and external equilibrium, in the Bretton Woods system exchange rates became fixed because members feared the consequences of allowing them to change. Nevertheless, because of increased capital mobility, the pressure for altering the parities of countries with persistent deficits and surpluses became harder to stop through the use of domestic policy tools and the aid of international rescue packages. Pressure increased from both academic and official sources for greater exchange rate flexibility. By 1968, the system had also evolved a form of international governance that was quite dif ferent from that envisioned at the beginning. Instead of a community o f equal currencies managed by the IMF, the system was managed by the United States in cooperation with the other members of the Group of Ten countries. In many respects, it was closer to the key cur rency system proposed by Williams.1 0 1 According to Dominguez (1993), the IMF was designed to facilitate international cooperation by serving as a commitment mechanism. It was to use its influence and its financial to enforce the par value system. It did not, however, have sufficient power to prevent devaluations by major countries and its financial resources were too limited to provide adequate adjustment assistance for them. The IMF still had an important role as a clearinghouse for different views on monetary reform, as a center of information, as the principal voice for the countries of the world other than the Group of Ten countries, as these countries' primary source of adjustment assistance and finally as an important partner in the major Group of Ten rescue packages. In sum, the problems of the interwar system that Bretton Woods was designed to prevent reemerged with a vengeance. The fundamental difference, however, was that the system was not likely to collapse into deflation as in 1931 but rather explode into inflation. THE CO LLAPSE OF BR ETTO N WOODS After the establishment of the two-tier arrange ment, the world monetary system was on a de fa c to dollar standard. The system became increasingly unstable until it collapsed with the closing of the gold window in August 1971. The collapse o f a system beset by the fatal flaws of the gold exchange standard and the adjustable peg was triggered by an acceleration in world inflation, in large part the consequence of an earlier acceleration of inflation in the United States. Before 1968, the U.S. inflation rate was below that of the GNP weighted inflation rate of the Group of Seven countries excluding the United States (see figure 15). It began accelerating in 1964, with a pause in 1966-67. The increase in inflation in the United States and the rest o f the world was closely related to an increase in money growth and in money growth relative to the growth of real output. (See figures 19 and 20.) Indeed, a prevalence o f excess demand shocks in the mid- and late 1960s is apparent for the United States and other Group of Seven countries in figures 11 and 12. Darby et al (1983) provided considerable evi dence on the transmission of inflation in the 1,0See W illiam s (1936 and 1943) and Johnson (1972b). MARCH/APRIL 1993 176 Figure 19 Money (M1) Growth Rates in the United States, G7 and G7 Excluding the United States, 1951-1973 Percent Figure 20 Money (M1) Less Real Output Growth in the United States, G7 and G7 Excluding the United States, 1951-1973 Percent FEDERAL RESERVE BANK OF ST. LOUIS 177 Bretton Woods system. Their regression anal yses led to a number of important conclusions. First, U.S. inflation was caused by lagged U.S. money growth. Second, U.S. money growth was independent of changes in international reserves—the balance of payments had no effect on the Federal Reserve’s reaction function. Third, U.S. money growth had strong and sig nificant effects on money growth in seven major countries. These lags were very long— up to four years—and reflected the fact that central banks in the seven countries sterilized reserve flows partially. Finally, money growth in the seven countries explained inflation in these countries with a significant lag.1 1 1 The key transmission mechanism of inflation was the classical price specie flow mechanism augmented by capital flows. Little evidence for other mechanisms including commodity market arbitrage was detected.1 2 According to these 1 authors, the Bretton Woods system collapsed because of the lagged effects of U.S. expansion ary monetary policy. As the dollar reserves of Germany, Japan and other countries accumu lated in the late 1960s and early 1970s, it became increasingly more difficult to sterilize them. This fostered domestic monetary expan sion and inflation. In addition, world inflation was aggravated by expansionary monetary and fiscal policies in the rest of the Group of Seven countries, as their governments adopted fullemployment stabilization policies. The only alternative to importing U.S. inflation was to float—the route taken by all countries in 1973.1 3 1 The crisis mounted from 1968 to 1971. The U.S. current account balance continued to deteriorate in 1968, but the overall balance of payments exhibited a surplus in 1968 and 1969 thanks to a large short-term capital inflow. The capital inflow was activated by events in the eurodollar market. In the face of tight monetary policy in 1968-69 and Regulation Q ceilings on time deposits, deposits shifted from U.S. banks to the eurodollar market. U.S. banks in turn borrowed in the eurodollar market, repatriating these funds. In 1970, as U.S. interest rates fell 111See Darby et al (1983) 1,2See Darby et al (1983). 113Except for the case of Japan, there is little evidence for the leading alternative explanation for the collapse— that it reflected grow ing m isalignm ent in real exchange rates between the U nited States and her p rincipal com petitors in the face of d ifferential pro d u ctivity trends. See M arston in response to rapid monetary expansion and Regulation Q was suspended for large certifi cates o f deposit, the borrowed funds returned abroad and the deficit grew to $9 billion, exploding to $30 billion by August 1971 (see fig ure 14). The dollar flood increased the reserves o f the surplus countries, auguring inflation. Ger man money growth doubled from 6.4 percent to 12 percent in 1971, and the German inflation rate increased from 1.8 percent in 1969 to 5.3 percent in 1971.1 4 Pressure mounted for a 1 revaluation of the mark. In April 1971 the dol lar inflow to Germany reached $3 billion. On May 5, 1971, the German central bank sus pended official operations in the foreign exchange market and allowed the deutsche mark to float. Similar action by Austria, Bel gium, the Netherlands and Switzerland fol lowed.1 5 1 In the following months, many began advocat ing ending the dollar’s link with gold. In April 1971, the U.S. balance of trade turned to deficit for the first time and influential voices began urging dollar devaluation. The decision to sus pend gold convertibility was triggered by French and British intentions in early August to convert dollars into gold. On August 15 at Camp David, President Nixon announced that he had directed Secretary Connolly "to suspend tem porarily the convertibility of the dollar into gold or other reserve assets.” The accompanying pol icy package included a 90-day wage-price freeze, a 10 percent import surcharge and a 10 percent investment tax credit.1 6 1 The U.S. decision to suspend gold convertibil ity ended a key aspect of the Bretton Woods system. The remaining part o f the system—the adjustable peg—disappeared 19 months later. The Bretton Woods system collapsed for three basic reasons. First, two major flaws under mined the system. One flaw was the gold exchange standard, which placed the United States under threat of a convertibility crisis. In reaction it pursued policies that in the end made adjustment more difficult. (1987) and Eichengreen (1992b). 114See M eltzer (1991). 115See Solom on (1976). 116See Solom on (1976). M ARCH/APRIL 1993 178 The second flaw was the adjustable peg. Because the costs o f discrete changes in parities were deemed high, in the face of growing capital mobility, the system evolved into a reluctant fixed exchange rate system without an effective adjustment mechanism. Finally, U.S. monetary policy was inappropri ate for a key currency. After 1965, the United States, by inflating, followed an inappropriate policy for a key currency country. Though the acceleration of inflation was low by the stand ards of the following decade, when superim posed on the cumulation of low inflation since World War II, it was sufficient to trigger a speculative attack on the world’s monetary gold stock in 1968, leading to the collapse of the Gold Pool.1 7 Once the regime had evolved into 1 a de fa c to dollar standard, the obligation of the United States was to maintain price stability. Instead, it conducted an inflationary policy, which ultimately destroyed the system. to Giovannini (1993), the Bretton Woods system was an asymmetric solution to Mundell’s n-1 currency problem.1 9 The United States as the 1 nth country, had to maintain the nominal anchor by following a stable monetary policy. In addition, it had to supply the dollars demanded by the rest of the world as reserves. The rest of the world had to accept, through its commitment to fixed parities, the price level set by the United States. But because o f the ad justable peg, it had the option to change parities. The rule was defective for the nonreserve cur rencies because the fundamental disequilibrium contingency was never spelled out and no con straint was placed on the extent to which domestic financial policy could stray from maintaining ex ternal balance. In addition, with growing capital mobility, the option to change parities became less viable. One can view the Bretton Woods system as a set of rules or commitment mechanisms.1 8 For 1 nonreserve-currency countries the rules were to maintain fixed parities, except in the contin gency of a fundamental disequilibrium in the balance of payments, and to use fiscal policy to smooth out short-run disturbances. The U.S. enforcement mechanism—access to its open cap ital markets—was presumably its dominant power because the IMF had little power. For the United States this rule suffered from a number o f fatal flaws. First, because of the fear of a confidence crisis, the gold convertibil ity requirement may have prevented the United States in the early 1960s from acting as a center country and willingly supplying the reserves demanded by the rest of the world. Second, as became evident in the later 1960s, this require ment was useless in preventing U.S. monetary authorities from pursuing an inflationary policy. Finally, although a mechanism was available for the United States to devalue the dollar, mone tary authorities were loath to use it for fear of undermining confidence. No effective enforce ment mechanism existed. Ultimately, the United States attached greater importance to domestic economic concerns than to its role as the center of the international monetary system. For the United States, the center country, the rule was to fix the gold price of the dollar at $35 per ounce and to maintain price stability. If a majority o f Bretton Woods members (and every member with 10 percent or more of the total quotas) agreed, however, the United States could change the dollar price of gold. There was no explicit enforcement mechanism other than reputation and the commitment to gold convertibility. According Thus although the Bretton Woods system can be interpreted as one based on rules, the system did not provide a credible commitment mech anism.1 0 The United States was unwilling to 2 subsume domestic considerations to the respon sibility of maintaining a nominal anchor. At the same time, other Group o f Seven countries became increasingly unwilling to follow the dic tates of the U.S.-imposed world inflation rate. D ID THE BR ETTO N W O O D S SYSTEM O PE R A TE AS A SYSTEM BASED ON CREDIBLE RULES? ” 7See G arber (1993). 118 See M cKinnon (1992) for his version of the rules of the Bretton W oods A rticles and the dollar standard. Also see G iovannini (1993) and O bstfeld (1993). ,19See M undell (1968). 120G io va n nini’s (1993) calculations show that d u ring the Bret ton W oods convertible period cred ib ility bounds on interest rates for the m ajor currencies, in contrast to the classical gold standard, were frequently violated. FEDERAL RESERVE BANK OF ST. LOUIS 179 The failure of the Bretton Woods rule sug gests a number of requirements for a welldesigned fixed exchange rate system. These include the following: • that the countries follow similar domes tic economic goals (underlying inflation rates); • that the rules be transparent; and • that some central monetary authority enforce them. The recent EMS system was quite successful for a number of years because it seemed to encom pass these three elements. Its recent crisis, how ever, reflected the emergence of some of the same problems that led to the breakdown of Bretton Woods. I discuss these issues below in the following subsection. PO ST BRETTON WOODS: MANAGED FLO A TIN G A N D THE EMS As a reaction to the flaws of the Bretton Woods system, the world turned to generalized floating exchange rates in March 1973. Though the early years of the floating exchange rate were often characterized as a dirty float, whereby mone tary authorities extensively intervened to affect both the levels of volatility and exchange rates, by the end of the 1970s it evolved into a system where exchange market intervention occurred primarily to smooth out fluctuations. Again in the 1980s exchange market intervention was used by the Group of Seven countries as part of a strategy o f policy coordination.1 1 In recent 2 years, floating exchange rates have been assailed from many quarters for excessive volatility in both nominal and real exchange rates, which in turn increase macroeconomic instability and raise the costs o f international transactions. The attack cites the favorable experience of the EMS from 1987 to 1991 in producing ex change rate and price stability as a recommen dation for a return to a global system of fixed exchange rates. It is argued that recent attempts at policy coordination can be formalized and ex tended to a more general managed system based on either close policy coordination (to keep ex change rates within specified target zones) or a renewed gold standard. In this paper I do not consider the merits or drawbacks o f policy coordination in general, but I examine the EMS briefly as a monetary regime similar to Bretton Woods.1 2 Of interest is whether lessons for the 2 international monetary system can be derived from its experience. The EMS, like the Bretton Woods system, represents an agreement among countries to set exchange rate parities, to manage intra-European Community exchange rates and to finance exc hange market intervention. Like Bretton Woods, it is an adjustable peg system. The origins of the EMS date back to the Bretton Woods period. The case for stable exchange rates within Europe was made in the context of the European Common Market (EEC). In addi tion to a strong dislike by Europeans for flexible exchange rates—based on their perception of inter war experience and their belief that exchange rate volatility reduces trade in highly open economies—the key motivation for extensive policy coordination to stabilize exchange rates was the common agricultural policy established in 1959.1 3 2 Food prices in the community are set in terms of a central unit o f account (the ECU) but quoted in local currency. Consequently, any changes in exchange rates lead to changes in local prices. During the Bretton Woods era, a system of sub sidies and taxes was worked out to insulate the local economy from policy realignments. This led to an asymmetric adjustment between hard currency countries reluctant to lower their agricultural prices and soft currency countries, which allowed their prices to rise. The result was overproduction of agricultural products and an ever-increasing fraction of the EEC budget allocated to subsidize agriculture. Early attempts to stabilize intra-European exchange rates during the Bretton Woods era were unsuccessful, as was the Snake in the Tunnel agreement o f the 1970s. The EMS, estab lished in 1979, was a formal attempt to overcome earlier obstacles to exchange rate stabilization. It was designed to prevent the defects o f the Bretton Woods system, including: the asymmetric adjustment mechanism, with the United States as the center, setting the tune for the rest of 121 See Bordo and Schwartz (1991). 122See Feldstein (1988) and Bordo and Schw artz (1989a). 123See Giavazzi and G iovannini (1989). M ARCH/APRIL 1993 180 the world; the problems associated with grow ing capital mobility; and the dramas of parity realignments. Instead, the EMS designed a set of intervention rules that would produce a sym metric system of adjustment; create a mechanism to finance exchange market interventions; and establish a code of conduct for realigning parities.1 4 2 Like Bretton Woods, the EMS was based on a set o f fixed parities called the exchange rate mechanism (ERM). Each country was to estab lish a central parity o f its currency in terms of ECU, the official unit of account. The ECU con sisted of a basket containing a set number of units of each currency. As the value of curren cies varied, the weights of each country in the basket would change. A parity grid of all bilateral rates could then be derived from the ratio of members’ central rates. Again, like Bret ton Woods, each currency was bounded by a set of margins o f 2.25 percent on either side of parity, creating a total band of 4.5 percent (for Italy, and later the United Kingdom, when it joined the ERM in 1990, the margin was set at 6 percent on either side of parity). The monetary authorities of both the depreciating and appre ciating countries were required to intervene when a currency hit one o f the margins. Coun tries were also allowed, but not required, to undertake intramarginal intervention. The indi cator of divergences, which measured each cur rency’s average deviation from the central parity, was devised as a signal for the monetary authorities to take policy actions to strengthen or weaken their currencies. It was supposed to work symmetrically. Intervention and adjustment was to be financed under a complicated set of arrange ments. These arrangements were designed to overcome the weaknesses of the IMF during Bretton Woods. The very short-term financing facility (VSTF) was to provide credibility to the bilateral parties by ensuring unlimited financing for marginal intervention. It provided automatic unlimited lines of credit from the creditor to the debtor members. The short-term monetary support (STMS) was designed to provide short term finance for temporary balance of pay 124See Giavazzi and G iovannini (1989). 125 A t its outset, there was considerable doubt that the EMS would be successful at w ithstanding the strains of greatly divergent m oney growth and inflation rates am ong its m em bers. See Fratianni (1980). See Giavazzi and G iovan nini (1989); Fratianni and von Hagen (1990 and 1992); and M eltzer (1990). FEDERAL RESERVE BANK OF ST. LOUIS ments disequilibration. The medium term finan cial assistance (MTFA) would provide longer term support. Unlike Bretton Woods, where members (other than the United States) could effectively decide to unilaterally alter their parities, changes in central parities were to be decided collectively. Finally, like Bretton Woods, members could (and did) impose capital controls. These have recently been phased out. The evidence on the performance of the EMS indicates that it was successful in the latter half of the 1980s at stabilizing both nominal and real exchange rates within Europe, at producing credi ble bilateral bands and at reducing divergence between members’ inflation rates.1 5 2 Giavazzi and Pagano (1988), Giavazzi and Giovannini (1989), and Giovannini (1989) make a strong case that the success of the EMS was largely due to the fact that it was an asymmet ric system with Germany acting as the center country. The other EMS members adapted their monetary policies to maintain fixed parities with Germany. Also, according to the aforementioned writers, the Bundesbank exhibited a strong credi ble commitment to low inflation and the other members of the ERM, by tying their currencies to the deutsche mark, used an exchange rate target as a commitment mechanism to success fully reduce their own rates of inflation. Evidence for the asymmetry hypothesis is based on the fact that the Bundesbank intervened only when bilateral exchange rates were breached, whereas the other countries engaged in intra-marginal intervention, and on evidence of asymmetrical behavior of interest rates in Germany and the other EMS countries. In the period preceding several EMS realignments, non-German EMS interest rates changed drastically, whereas no change was observed in their German counter part. Evidence that the Bundesbank’s reputation was responsible for the disinflation of the 1980s is based on an out-of-sample simulation of a VAR to predict the inflation rate. Downward shifts in the predicted values of inflation for a number o f countries after the advent o f the EMS makes the case. That inflation expectations 181 were significantly reduced only in France and Italy several years after the advent of the EMS (the argument goes) may reflect slow learning or alternatively that these countries used the EMS to justify following unpopular austerity policies. Fratianni and von Hagen (1990 and 1992) dispute both the asymmetry and the imported disinflation hypotheses.1 6 Evidence based on Granger causal 2 ity tests from a structural VAR suggests that the German monetary base was not insulated from other EMS base movements nor were non-German EMS monetary bases insulated by the German monetary base from external shocks. In this interpretation, the EMS is a coordinated mone tary policy system with all members playing a role. Finally, Fratianni and von Hagen (1990) pro vide evidence that the EMS has reduced intraEuropean exchange rate volatility; however, this reduction has been at the expense of increased volatility of non-EMS currencies. Thus they argue that the EMS is on net balance beneficial to welfare because intra-EMS trade exceeds external trade. They also show that although the advent of the EMS has not reduced inflation uncertainty relative to non-EMS countries, it has reduced the effects of foreign inflation shocks on the members. Despite its favorable performance since the mid-1980s, the EMS was recently subjected to the same kinds of stress that plagued Bretton Woods. September 1992 and November 1992 marked a series of exchange rate crises in Europe that paralleled the events of 1967 to 1971. Precipitated by concerns that French vot ers would reject the Maastricht Treaty on Euro pean Monetary Union in a referendum on September 20, speculators staged attacks early in the month on the Nordic currencies and then later on the Italian lira, the British pound, the French franc, and other weaker currencies. The crisis led to the disabling o f the ERM. Both Italy and the United Kingdom left it while Spain, Portugal and Ireland reimposed or strengthened existing capital controls: in November Sweden floated and Portugal and Spain devalued. The fundamental causes o f the crisis, like the crises that plagued Bretton Woods, lay in large part with the exchange rate system. The EMS, like Bretton Woods, is a pegged exchange rate system that requires member countries to follow similar domestic monetary and fiscal policies and hence have similar inflation rates. This is difficult to do in the face of both differ ing shocks across countries and differing national priorities. Under Bretton Woods, capital con trols and less integrated international capital markets allowed members to follow divergent policies for considerable periods. Under the EMS, the absence of controls (after 1990) and the presence o f extremely mobile capital meant that any movement o f domestic policies away from those consistent with maintaining parity would quickly precipitate a speculative attack. Also, just as under Bretton Woods, the adjusta ble peg in the face o f such capital mobility became unworkable. Thus the difference between the two regimes when faced with asymmetric shocks or differing national priori ties was the speed of reaction by world capital markets. Though the fundamental cause of the crisis was similar in the two regimes, the source of the problem differed. Under Bretton Woods, the shock that led to its collapse was an accelera tion o f inflation in the United States, ostensibly to finance the Vietnam War, as well as social policies, and to maintain full employment. Under the EMS, the shock was bond financed German reunification and the Bundesbank’s sub sequent deflationary policy. In each case, the system broke down because other countries were unwilling to go along with the policies of the center country. The commitments to price stability by both the center country and the other members were not shown as credible. Under Bretton Woods, Germany and other west ern European countries were reluctant to inflate or revalue and the United States was reluctant to devalue. Under the EMS, the United Kingdom, Italy, Spain, Portugal, Ireland and Sweden were unwilling to deflate and Germany was unwilling to revalue. As under Bretton Woods, although the EMS had the option for a 126Also, C ollins (1988) and Eichengreen (1992d) present evi dence that EMS m em bership m ay not have been responsi ble for reducing the inflation rates of EMS countries. Their cross-country regressions show that EMS m em bership had little e ffect on inflation perform ance. Changing p ublic a tti tudes tow ard inflation w ithin each country represent a more im portant determ inant. See Giavazzi and Collins (1992). M ARCH/APRIL 1993 182 general realignment, both improved capital mobility and the Maastricht commitment to a unified currency made it an unrealizable outcome. Thus the lesson from both the EMS and Bret ton Woods is that pegged exchange rate systems do not work for long no matter how well they are designed. Pegged exchange rates, capital mobility and policy autonomy just do not mix. During the heyday of Bretton Woods years ago, the case made for floating exchange rates for major countries still holds. This is not to say that if European countries were completely will ing to give up domestic policy autonomy, they could not eventually form a currency union with perfectly fixed exchange rates. In an uncertain world subject to diverse shocks, the costs for individual countries of doing so are apparently extremely high. CONCLUSION This paper examines statistical evidence on the performance of alternative monetary regimes over the past century. It also examines some aspects of the history of these regimes. Both statistical and historical evidence may help provide answers to the question why some regimes have been more successful than others. They also have implications for current issues in international monetary reform and the ongo ing debate over rules and discretion. The statistical evidence on performance of alternative monetary regimes in the second sec tion leads to the conclusion that the Bretton Woods convertible regime from 1959 to 1970 was by far the best on virtually all criteria, but that the recent floating regime is not much worse. Indeed, it is clear that the performance of the regimes in the post-World W ar II era is superior to the performance of regimes in the preceding half century. Finally, though the clas sical gold standard does relatively poorly in terms of the stability of real variables, it per formed best on inflation persistence and financial market integration—evidence for the successful operation of gold as a nominal anchor. This evidence leads to the following question: Why was Bretton Woods so stable yet so fragile and the classical gold standard so unstable and yet so durable? The answer may be due in part to the shocks the two regimes faced. This, how ever, seems unlikely because the gold standard was subject to both supply and demand shocks FEDERAL RESERVE BANK OF ST. LOUIS that were a multiple of those facing Bretton Woods. It could also be due to greater flexibility of wages and prices and greater factor mobility before W orld W ar I, which meant that adjust ing to the greater shocks did not have as serious consequences on real activity and employment as later in the twentieth century. Alternatively, political economy factors—such as a more limited suffrage; less concern over the main tenance of full employment; limited understand ing of the link between monetary policy and the level of economic activity; and hence loss o f an incentive for monetary authorities to pursue policies that would threaten adherence to convertibility—could be responsible. These hypotheses clearly need more investigation. It also could be due to regime design and especially the incentive compatibility features of the regime. The classical gold standard may have been so successful because of the credibility of the commitment to the gold standard rule of convertibility and because o f its near universal acceptance. In turn, the credibility of the gold standard may stem from the origins of gold as money and the importance of Great Britain, the most important commercial nation of the nine teenth century, in enforcing the rules. England’s commitment to convertibility in turn was aided by stabilizing private capital flows. The classical gold standard for the core coun tries worked as a contingent rule or a rule with escape clauses. As a consequence, it was flexible enough to withstand major shocks. It also ena bled governments to finance major wars flexi bly, by allowing them to leave the gold standard and temporarily use seigniorage to finance unusual government expenditures. The rule may have endured because the requisite defla tion required to restore convertibility after the emergency had passed may not have had severe effects on real variables. This may have been because wages and prices were highly flexible. Alternatively the deflation accompanying re sumption may have had significant real effects but no political constituency strong enough to oppose it existed. The classical gold standard collapsed under the unprecedented shocks of W orld W ar I. It was reinstated as the short-lived gold exchange standard. Its brief life reflected the fatal flaws made famous by the Triffin dilemma. But regard less o f the weakness of the gold exchange stan dard, it suffered from the absence o f an effec tive commitment mechanism. There was no cen 183 ter country to enforce the rule, just three rivals pulling in different directions. Also, it was the beginning o f an era when countries were less willing to go along with the gold convertibility rule because they attached greater weight to the objective of domestic economic stability. The Bretton Woods system was set up to pre vent the perceived flaws of the classical gold standard and the trauma of the interwar period. The Bretton Woods adjustable peg was in some respects similar to the gold standard contingent rule, but it invited speculative attacks, hence compromising its role as an escape clause. Bretton Woods evolved into a gold exchange standard fraught with the adjust ment, liquidity and confidence problems of the interwar period. Though the problems of the gold exchange standard could possibly have been corrected by raising the price of gold, as it turned out, it evolved into an asymmetric dollar stand ard. The United States maintained the credible commitment to a noninflationary policy for only a few years. By the mid-1960s it shifted to an inflationary policy to further its domestic inter ests. The rest of the world, faced with imported inflation, soon lost the incentive to follow U.S. leadership and the system collapsed in 1971. The advent o f the general floating exchange rate system in 1973 and its longevity suggests that the lessons of Bretton Woods have been learned well. Countries are not willing to sub ject their domestic policy autonomy to that of another country of whose commitment they cannot be sure in a stochastic world nor to a supernational monetary authority they cannot control. The key advantage of floating exchange rates stressed a generation ago by Milton Fried man and Harry Johnson—the freedom to pursue an independent monetary policy—still holds today. Major countries can design domestic monetary policy rules to achieve domestic price stability without the costs o f giving up their policy autonomy. The experience of the EMS reveals that coun tries that have similar goals and face similar shocks can establish a regional exchange rate regime. This regime requires both a credible commitment mechanism and the willingness of member countries to give up sovereignty for a higher purpose. As attested to by the events of September and November 1992, however, the durability o f such an arrangement seems doubt ful, as was the case for Bretton Woods, in an uncertain world subject to diverse shocks where national priorities can change and com mitments can be broken. Some have argued that the EMS can be preserved only by precom mitment to price stability and fixed exchange rates by independent central banks.1 7 Others 2 argue that the only solution is rapid movement to a unified currency enforced by a European central bank.1 8 As Feldstein (1992) points out, 2 however, full-fledged monetary union com pletely precludes the use of domestic monetary policy. To the extent that country-specific shocks dominate common shocks and labor is relatively immobile between European coun tries, the benefits o f permanently fixed exchange rates may not outweigh the cost of increased economic dislocation.1 9 2 Finally, proposals for monetary reform, such as exchange rate target zones or targeting the real price of gold, though possibly of scientific merit, would work only if nations are willing to give up domestic autonomy and follow credible commitments. The history of international monetary regimes casts doubt on the likelihood that the nations of the world will do so in the forseeable future. REFERENCES A logoskoufis, G eorge S. and Ronald Sm ith. “ The Phillips Curve, the Persistence of Inflation, and the Lucas Critique: Evidence from Exchange-Rate R egim es,” A m e ric a n E c o n o m ic R e v ie w (Decem ber 1991), pp. 1254-75. Alogoskoufis, G eorge S. “ M onetary A ccom odation, Exchange R a te R e g im e s a n d In fla tio n P e rs is te n c e ,” E c o n o m ic J o u rn a l (M ay 1992), pp. 4 6 1-80. Barro, Robert J., and David B. G ordon. “ Rules, Discretion and R eputation in a M odel of M onetary P olicy,” J o u rn a l o f M o n e ta ry E c o n o m ic s (July 1983), pp. 101-21. Barsky, Robert B. “ The Fisher Hypothesis and the Forecastability and Persistence of In fla tion ,” J o u rn a l o f M o n e ta ry E c o n o m ic s (January 1987), pp. 3 -2 4 . Baxter, M arianne, and Alan C. S tockm an. “ Business Cycles and the Exchange-Rate Regim e: Some International Evidence,” J o u rn a l o f M o n e ta ry E c o n o m ic s (May 1989), pp. 377-400. Bayoum i, Tam in, and Barry E ichengreen. E c o n o m ic P e rfo rm a n c e U n d e r A lte rn a tiv e E x c h a n g e R a te R e g im e s: S om e H is to ric a l E vid e n ce (U niversity of C alifornia at Berkeley, 1992a). 127See Neum ann (1991). 128See Delors (1989). 129See Eichengreen (1992d). M ARCH/APRIL 1993 184 ________ , a n d ________ . S h o c k in g A s p e c ts o f E u ro pean M o n e ta ry U n ific a tio n , NBER W orking Paper No. 3949 (January 1992b). Buiter, W illem H. “ A Viable Gold Standard Requires Flexible M onetary and Fiscal P o licy,” R e v ie w o f E c o n o m ic S tu d ie s (January 1989), pp. 101-18. ________ , a n d ________ Is There a C o n flic t B e tw e e n EC E n la rg e m e n t a n d E u ro p e a n M o n e ta ry U n ific a tio n ? NBER W orking Paper No. 3950 (January 1992c). Cagan, Philip. "O n the Report of the Gold C om m ission (1982),” C a rn e g ie R o c h e s te r C o n fe re n ce S e rie s on P u b lic P olicy, (Autum n 1984), pp. 2 4 7-67. ________ , a n d ________ . M o n e ta ry a n d E x c h a n g e R ate A rra n g e m e n ts fo r N A FTA (U niversity of C alifornia at Ber keley, 1992d). Calom iris, C harles W. “ Price and Exchange Rate Determ ina tion During the G reenback S u sp e n sio n ,” O x fo rd E c o n o m ic P a p e rs (Decem ber 1988), pp. 7 1 9-50. Bergsten, C. Fred. “ The Collapse of Bretton W oods: Im plica tions for International M onetary R eform ,” in M ichael D. Bordo and Barry Eichengreen, eds. A R e tro s p e c tiv e on the B re tto n W ood s In te rn a tio n a l M o n e ta ry S yste m , (U niversity of Chicago Press, 1993), pp. 5 8 7-93. C anzoneri, M atthew B., and Dale W. H enderson. “ Is Sover eign Policym aking B ad?” C a rn e g ie R o c h e s te r C o n fe re n ce S e rie s o n P u b lic P o lic y (Spring 1988), pp. 9 3 -140. Bernanke, Benjam in. “ Nonm onetary Effects of the Financial Crisis in the Propagation of the Great D epression,” A m e ri c a n E c o n o m ic R e v ie w (June 1983), pp. 2 5 7-76. ________ and Harold Jam es. “ The Gold Standard, Deflation and Financial C risis in the Great Depression: An Interna tional C o m pa riso n ,” in R. Glenn Hubbard, ed. F in a n c ia l M a rk e ts a n d F in a n c ia l C risis (U niversity of Chicago Press, 1991), pp. 3 3 -6 8 . Blanchard, O livier, and Danny Quah. “ The Dynam ic Effects of A ggregate Dem and and Aggregate Supply D istur ba n ces,” A m e ric a n E c o n o m ic R e v ie w (Septem ber 1989), pp. 6 5 5-73. Black, Stanley W. “ International M onetary In stitu tion s,” N e w P a lg ra v e D ic tio n a ry o f E c o n o m ic s (M acM illan, 1987), pp. 9 1 7-20. Bloom field, A rthur. M o n e ta ry P o lic y U n d e r th e In te rn a tio n a l G o ld S ta n d a rd , 1 8 0 0 -1 9 1 4 (Federal Reserve Bank of New York, 1959). Bordo, M ichael D. “ The Classical Gold Standard: Some Lessons for T o d a y,” this R e v ie w (May 1981), pp. 2 -1 7 . ________ . “ The G old Standard: The T raditional A p p ro a ch ,” in M ichael D. Bordo and Anna J. Schwartz, eds. A R e tro s p e c tiv e on the C la s s ic a l G o ld S ta n d a rd , 1 8 2 1 -1 9 3 1 (U niversity of Chicago Press, 1984), pp. 2 3 -120. ________ . “ The Bretton W oods International M onetary System : An Historical O verview ,” in M ichael D. Bordo and Barry E ichangreen, eds. A R e tro s p e c tiv e o n th e B re tto n W ood s S y s e m (U niversity of Chicago Press, 1993). __________ , a n d ________ . M o n e ta ry P o lic y in In te rd e p e n d e n t E co n o m ie s (M assachusetts Institute of T echnology Press, 1991). C ecchetti, Stephen G., and G eorgios Karras. S o u rce s o f O u tp u t F lu c tu a tio n s D u rin g th e In te rw a r P e rio d : F u rth e r E vid e n ce on the C a u se s o f the G re a t D e p re ssio n , NBER W orking Paper No. 4049 (April 1992). C ollins, Susan. “ Inflation and the European M onetary Sys te m ,” in Francesco Giavazzi, Stefano M icossi and M arcus M iller, eds. The E u ro p e a n M o n e ta ry S y s te m (Cam bridge University Press, 1988), pp. 112-36. ________ , and Francesco Giavazzi. “ A ttitudes Toward Infla tion and the V iability of Fixed Exchange Rates: Evidence from E M S ,” in M ichael Bordo and Barry E ichengreen, eds. A R e tro s p e c tiv e on th e B re tto n W ood s S yste m , (U niversity of Chicago Press, 1993), pp. 5 4 7-77. Cooper, R ichard. “ The Gold Standard: Historical Facts and Future P rospects,” B ro o k in g s P a p e rs o n E c o n o m ic A c tiv ity (vol. 1, 1982), pp. 1-45. Darby, M ichael R., Jam es R. Lothian et. al. The In te rn a tio n a l T ransm issio n o f In fla tio n . (U niversity of Chicago Press, 1983). ________ , a n d ________ . “ The International Transm ission of Inflation A flo a t,” in M ichael D. Bordo, ed. M o ney, H istory, a n d In te rn a tio n a l F in a n ce : E ssa ys in H o n o r o f A n n a J. S c h w a rtz (U niversity of Chicago Press, 1989), pp. 2 0 3-36. DeKock, G abriel, and V ittorio G rilli. E n d o g e n o u s E xch a n g e R a te R e g im e S w itc h e s , NBER W orking Paper No. 3066 (August 1989). ________ , and R ichard E. Ellson. “ A M odel of the Classical G old Standard W ith D e pletion,” J o u rn a l o f M o n e ta ry E c o n o m ic s (July 1985), pp. 109-20. ________ , a n d _________. F is c a l P o lic ie s a n d the C h o ic e o f E xch a n g e R a te R egim e, CEPR D iscussion Paper No. 631 ________ , and Finn E. Kydland. The G o ld S ta n d a rd A s a R ule, Federal Reserve Bank of Cleveland W orking Paper No. 9205, (M arch 1992). Delors, Jacques et al. R e p o rt o n E c o n o m ic a n d M o n e ta ry U nion in the E u ro p e a n C o m m u n ity (European C om m ission, 1989). ________ , and Angela Redish. “ Credible C om m itm ent and Exchange Rate Stability: C anada’s Interwar E xp e rie n ce,” C a n a d ia n J o u rn a l o f E c o n o m ic s (May 1990), pp. 3 5 7-80. Despres, Emile, C harles K indleberger and W illiam Salant. “ The Dollar and W orld Liquidity: A M inority V ie w ,” The E c o n o m is t (February 1966), pp. 5 2 6-29. ________ , and A nna J. Schwartz. “ T ransm ission of Real and M onetary Disturbances U nder Fixed and Floating R ates,” in Jam es A. Dorn and W illiam A. Niskanen, eds. D ollars, D e fic its a n d T rade (Kluwer A cadem ic Publishers, 1989a), pp. 2 3 7 -5 8 . Diz, Adolfo C. “ The C onditions Attached to A djustm ent Financing: Evolution of the IMF P ra ctice ,” in The In te rn a ________ , a n d ________ . “ The ECU— An Im aginary or E m bryonic Form of M oney: W hat Can We Learn From H istory?” in Paul DeGrauwe and Theo Peeters, eds. The E C U a n d E u ro p e a n M o n e ta ry In te g ra tio n M acM illan, 1989b), pp. 1-21. D om inguez, Kathryn. “ The Role of International O rganiza tions in the Bretton W oods S ystem ,” in M ichael D. Bordo and Barry Eichengreen, eds. A R e tro s p e c tiv e o n th e B re t ton Woods S yste m (U niversity of Chicago Press, 1993), pp. 357-97. ________ , a n d ________ . “ W hat Has Foreign Exchange M arket Intervention Since the Plaza A greem ent A ccom plish e d ? ” O pe n E c o n o m ic s R e v ie w (vol. 2, no. 1, 1991). Dornbusch, R udiger. “ Expectations and Exchange Rate D ynam ics,” J o u rn a l o f P o litic a l E c o n o m y (Decem ber 1976), pp. 1161-76. ________ , and Eugene N. W hite. “ A Tale of Two Currencies: British and French Finance During the N apoleonic W a r,” J o u rn a l o f E c o n o m ic H is to ry (June 1991), pp. 3 0 3-16. Eichengreen, Barry. “ E d ito r’s In tro d u ctio n ,” in Barry E ichen green, ed. The G o ld S ta n d a rd in Theory a n d H is to ry (M ethuen, 1985), pp. 1-36. FEDERAL RESERVE BANK OF ST. LOUIS (M arch 1992). tio n a l M o n e ta ry S yste m : F o rty Years A fte r B re tto n W oods, Federal Reserve Bank of Boston C onference Series No. 28 (May 1984), pp. 214-325. 185 ________ “ C onducting the International O rchestra: Bank of England Leadership Under the Classical Gold S ta n d a rd ,” J o u rn a l o f In te rn a tio n a l M o n e y a n d F in a n c e (M arch 1987), pp. 5 -2 9 . ________. The G o ld S ta n d a rd S in c e A le c F o rd. (U niversity of C alifornia at Berkeley, 1989a). ________. “ H egem onic S tability T he o rie s,” in Richard Cooper et al, eds. C an N a tio n s A g re e : Is s u e s in In te rn a tio n a l E c o n o m ic C o o p e ra tio n (Brookings Institution, 1989b), pp. 2 55-98. ________ . “ The C om parative Perform ance of Fixed and Flexi ble Exchange Rate Regim es: Inter-war E vidence,” in Niel Thygesen et al, eds. B u s in e s s C y c le s : Theories, E v id e n c e a n d A n a ly s is (M acm illan, 1991a), pp. 2 2 9-72. ________ “ T rends and C ycles in Foreign Le n ding ,” in Horst Siebert, ed. C a p ita l F lo w s in the W orld E c o n o m y (World Institute of W orld Econom ics, 1991b), pp. 3 -2 8 . ________ . U.S. F o re ig n R e la tio n s in the 2 0 th C e n tu ry (U niver sity of C alifornia Berkeley, 1991c). ________ “ H istory of the International M onetary System: Im plications for Research in International M acroeconom ics and F inance,” in Frederick van der Ploeg, ed. H a n d b o o k o f In te rn a tio n a l M a c ro e c o n o m ic s . (Blackwell, 1993a). ________ . “ Epilogue: Three Perspectives on the Bretton W oods S ystem ,” in M ichael D. Bordo and Barry Eichen green, eds. A R e tro s p e c tiv e o n the B re tto n W oods S ystem (U niversity of C hicago Press, 1993b), pp. 6 2 1-57. ________ . G o ld e n F e tte rs : The G o ld S ta n d a rd a n d th e G re a t D epression , 1 9 1 9 -1 9 3 9 , (O xford University, 1992a). ________ S h o u ld the M a a s tric h t Treaty b e S a v e d ? University of C alifornia at Berkeley Political Econom y of European Integration Research G roup W orking Paper 1.10 (Septem ber 1992b). E m m inger, Otmar. “ Practical Aspects of the Problem of Balance-of-Paym ents A d ju stm e nt,” J o u rn a l o f P o litic a l E c o n o m y (August 1967), pp. 5 12-22. Feldstein, M artin. In te rn a tio n a l E c o n o m ic C o o p e ra tio n (U niver sity of C hicago Press, 1988). Feldstein, M artin. “ The Case Against E M U ,” The E c o n o m is t (June 13, 1992), pp. 19-22. Fischer, Stanley. “ Rules Versus Discretion in M onetary Pol icy,” in Benjam in M. Friedm an and Frank H. Hahn, eds. H a n d b o o k o f M o n e ta ry E co n o m ics, Vol. 2 (North Holland, 1990). ________ , a n d ________ The E u ro p e a n M o n e ta ry S yste m a n d E u ro p e a n M o n e ta ry U nion (W estview Press, 1992). Friedm an, M ilton. “ The Case for Flexible Exchange R ates,” in E ssa ys in P o sitive E c o n o m ic s (U niversity of Chicago Press, 1953), pp. 157-203. ________. “ Bim etallism R evisited,” J o u rn a l o f E c o n o m ic P er s p e c tiv e s (Fall 1990), pp. 8 5-104. ________ , and Schwartz, A nna J. A M o n e ta ry H is to ry o f the U n ite d S tate s, 1 8 6 7 -1 9 6 0 (Princeton U niversity Press, 1963). G allarotti, G uilio. “ Centralized Versus D ecentralized Interna tional M onetary System s: T he Lessons of the Classical G old S ta n d a rd ,” in Jam es Dorn, ed. A lte rn a tiv e s to G o v e rn m e n t F ia t M o n e y [Kluw er A cadem ic Publishers, 1991 (forthcom ing)]. G arber, Peter. “ The C ollapse of the Bretton W oods Fixed Exchange Rate S yste m ,” in M ichael D. Bordo and Barry Eichengreen, eds. A R e tro s p e c tiv e o n the B re tto n W oods S yste m (U niversity of Chicago Press, 1993), pp. 4 6 1-85. Giavazzi, Francesco, and A lberto G iovannini. L im itin g E xch a n g e R a te F le x ib ility : The E u ro p e a n M o n e ta ry S ystem (M assachusetts Institute of T echnology Press, 1989). ________ , and M arc Pagano. “ The Advantage of Tying O n e ’s Hands: EMS D iscipline and C entral Bank C re d ib ility,” E u ro p e a n E c o n o m ic R e v ie w (June 1988), pp. 1055-75. Gilbert, Milton. The G o ld -D o lla r S yste m : C o n d itio n s o f E q u ilibriu m a n d the P rice o f G o ld (Princeton Departm ent of Econom ics, 1968). G iovannini, Alberto. “ How Do Fixed-Exchange-Rate Regim es W ork Evidence From the Gold S tandard, Bretton W oods and the E M S ,” in M arvin M iller, Barry Eichengreen and Richard Portes, eds. B lu e p rin ts fo r E x c h a n g e R a te M a n a g e m e n t (Center for Econom ic Policy Research, 1989), pp. 13-41. ________ . “ Bretton W oods and Its Precursors: Rules Versus D iscretion in the H istory of International M onetary R egim es,” in M ichael D. Bordo and Barry Eichengreen, eds. A R e tro s p e c tiv e o n the B re tto n W ood s S yste m (U niversity of Chicago Press, 1993), pp. 109-55. G rilli, Vittorio, and G raciella Kam insky. “ Nom inal Exchange Rate Regim es and the Real Exchange Rate: Evidence from the United States and Great Britain, 1 8 8 5-1 9 8 6 .” J o u rn a l o f M o n e ta ry E c o n o m ic s (April 1991), pp. 191-212. Grossm an, Herschel I. and John B. Van Huyck. “ Sovereign Debt as a C ontingent Claim : Excusable Default, Repudia tion, and R ep u tatio n ,” A m e ric a n E c o n o m ic R e v ie w (Decem ber 1988), pp. 1088-97. Fishlow, Albert. M a rk e t F o rc e s o r G ro u p In te re s ts : In c o n v e rti b le C u rre n c y in P re-1914 L a tin A m e ric a (U niversity of C alifornia at Berkeley, 1987). Halm, George. A p p ro a c h e s to G re a te r F le x ib ility o f E xch a n g e R a te s: The B u rg e n s to c k P a p e rs (Princeton University Press, 1970). ________. “ C o n ditionality and W illingness to Pay: Som e Par allels from the 1890s,” in Barry Eichengreen and Peter Lindert, eds. The In te rn a tio n a l D e b t C risis in H is to ric a l P er s p e c tiv e (M assachusetts Institute of T echnology Press, 1989), pp. 8 6-105. Ham ada, Koichi. “ M acroeconom ic Strategy C oordination Under A lternative Exchange Rates,” in R udiger D ornbusch and Jacob Frankel, eds. In te rn a tio n a l E c o n o m ic P o lic y (The Johns Hopkins U niversity Press, 1979), pp. 2 9 2-324. Flood, Robert P., and Peter Isard. S im p le R ules, D is c re tio n a n d M o n e ta ry P o licy, NBER W orking Paper No. 2934 (April 1989a). ________ , a n d ________ “ M onetary Policy S tra te gie s,” IM F S ta ff P a p e rs (Septem ber 1989b), pp. 612-32. Fratianni, M ichele. “ The European M onetary System : A Return to an A djustable-P eg A rran g e m e nt,” C a rn e g ie R o c h e s te r C o n fe re n c e S eries on P u b lic P o lic y (Autumn 1980), pp. 9 3 -140. ________, and Juergen von Hagen. “ The European M onetary System Ten Years A fte r,” C a rn e g ie R o c h e s te r C o n fe re n c e S eries on P u b lic P o lic y (Spring 1990), pp. 173-241. Hayashi, Fumio. J a p a n ’s S a v in g R ate : N e w D a ta a n d R e fle c tio n s, NBER W orking Paper No. 3205 (Decem ber 1989). Helpm an, Elhanan. “ An Exploration in the T heory of Exchange Rate R egim es,” J o u rn a l o f P o litic a l E co n o m y (O ctober 1981), pp. 8 6 5-90. ________ , and Assaf Razin. “ Towards a Consistent C om pari son of Alternative Exchange Rate S ystem s,” C an a d ia n J o u rn a l o f E c o n o m ic s (August 1979), pp. 3 9 4-409. H offm an, Stanley, and C harles Maier, eds. The M a rs h a ll P lan: A R e tro s p e c tiv e (W estview Press, 1984). Johnson, H arry G. “ The Sterling Crisis of 1967 and the Gold Rush of 1968,” N e b ra ska J o u rn a l o f E c o n o m ic s a n d B u s i n e ss (Autum n 1968), pp. 3 -1 7 . M ARCH/APRIL 1993 186 ________ “ The Case For Flexible Exchange Rates, 1969,” in F u rth e r E ssa ys in M o n e ta ry E c o n o m ic s (Harvard U niversity Press, 1976), pp. 198-227. ________ ‘ ‘The Bretton W oods System , Key C urrencies, and the ‘D ollar C ris is ’ of 1971,” Three B a n k s R e v ie w (June 1972), pp. 3 -2 2 . Jonung, Lars. ‘ ‘Sw edish Experience Under the Classical Gold Standard, 1 8 7 3-1 9 1 4 ,” in M ichael D. Bordo and Anna J. Schwartz, eds. A R e tro s p e c tiv e o n the C la s s ic a l G o ld S ta n d a rd , 1 8 2 1 -1 9 3 1 (U niversity of Chicago Press, 1984), pp. 3 6 4-404. Kaplan, Jacob J., and G unther Schleim inger. The E u ro pean P a y m e n ts U nion : F in a n c ia l D ip lo m a c y in the 1950s (C laren don Press, 1989). Keating, John W., and John V. Nye. P e rm a n e n t a n d Transi to ry S h o c k s in R e a l O u tp u t: E s tim a te s fro m N in e te e n th C e n tu ry a n d P o s tw a r E c o n o m ic s , W ashington U niversity W orking Paper No. 160 (July 1991). Kenen, Peter B. “ International Liquidity and the Balance of Paym ents of a Reserve-Currency C o u n try,” Q u a rte rly J o u r n a l o f E c o n o m ic s (Novem ber 1960), pp. 5 7 2-86. Kindleberger, Charles. The W o rld in D epression , 1 9 2 9 -1 9 3 9 (U niversity of C alifornia Press, 1973). Klein, Benjam in. “ O ur New M onetary Standard: The M easure m ent and Effects of Price Uncertainty, 1 8 8 0-1 9 7 3 ,” E c o n o m ic In q u iry (Decem ber 1975), pp. 4 6 1-84. M oggridge, Donald E. “ Keynes and the International M one tary System , 1 9 0 9 -4 6 ,” in John Cohen and G regory C. Harcourt, eds. In te rn a tio n a l M o n e ta ry P ro b le m s a n d S u p p ly S id e E c o n o m ic s (M acM illan, 1986), pp. 5 6 -8 3 . M undell, Robert A. In te rn a tio n a l E c o n o m ic s (M acM illian, 1968). ______ _ “ The International M onetary F u n d ,” J o u rn a l o f W o rld Trade L a w (1969) pp. 4 5 5 -9 7 . ________ Untitled, in M ichael D. Bordo and Barry Eichengreen, eds. A R e tro s p e c tiv e o n th e B re tto n W oods In te rn a tio n a l M o n e ta ry S y s te m (U niversity o f C hicago Press, 1993), pp. 6 0 4-12. Mussa, M ichael. “ Nom inal Exchange Rates and the Behavior of Real Exchange Rates: Evidence and Im plicatio n s,” C a rn e g ie R o c h e s te r C o n fe re n ce S e rie s on P u b lic P o lic y (Autum n 1986) pp. 117-214. Neum ann, M anfred J.M . “ P recom m itm ent by Central Bank Independence,” O pe n E c o n o m ie s R e v ie w (1991), pp. 9 5 -1 1 2 . Nurske, Ragnar. In te rn a tio n a l C u rre n c y E xp e rie n ce (C olum bia U niversity Press, 1944). O bstfeld, M aurice. “ T he A d justm ent M ech a n ism ,” in M ichael D. Bordo and Barry E ichengreen, eds. A R e tro s p e c tiv e on th e B re tto n W ood s S y s te m (U niversity of Chicago Press, 1993), pp. 2 0 1-56. Kydland, Finn E., and Edward Prescott. “ Rules Rather than Discretion: The Inconsistency of O ptim al P lans,” J o u rn a l o f P o litic a l E c o n o m y (June 1977), pp. 4 7 3-91. O fficer, Lawrence. “ The E fficiency of the Dollar-S terling Gold Standard, 1 8 9 0 -1 9 0 8 ,” J o u rn a l o f P o litic a l E c o n o m y (O ctober 1986), pp. 1038-73. Lindert, Peter. K e y C u rre n c ie s a n d G old, 1 9 0 0 -1 9 1 3 , Prince ton Studies in International Finance, No. 24 (Princeton U niversity Press, 1969). Persson, Torsten and G uido T abellini. M a c ro e c o n o m ic P olicy, C re d ib ility a n d P o litic s (Harwood Press, 1990). Lucas, R obert E. Jr., and Nancy L. Stokey. “ O ptim al Fiscal and M onetary Policy in an Econom y W ithout C a p ita l,” J o u rn a l o f M o n e ta ry E c o n o m ic s (July 1983), pp. 5 5 -9 3 . M ankiw , G regory. “ The O ptim al C ollection of Seigniorage— T heory and E vidence,” J o u rn a l o f M o n e ta ry E c o n o m ic s (Septem ber 1987), pp. 3 2 7-41. M arston, R ichard. “ Real Exchange Rates and Productivity G rowth in the U nited States and Ja p a n,” in Sven Arndt and David R ichardson, eds. R e a l-F in a n c ia l L in k a g e s A m o n g O pe n E c o n o m ic s (M assachusetts Institute of T ech nology Press, 1987), pp. 7 1 -9 6 . M cKinnon, Ronald I. M o n e y in In te rn a tio n a l E x c h a n g e : The C o n v e rtib le C u rre n c y S y s te m (O xford U niversity Press, 1979). ________ “ An International G old Standard W ithout G o ld ,” C a to J o u rn a l (Fall 1988), pp. 3 5 1-73. ________ . “ A lternative International M onetary System s: The Rules of the G am e R econsidered,” J o u rn a l o f E c o n o m ic L ite ra tu re (forthcom ing 1993). M eltzer, Allan H. K e y n e s ’s M o n e ta ry Theory: A D iffe re n t In te rp re ta tio n (C am bridge University Press, 1988). ________ “ Som e Em pirical Findings on D ifferences Between EMS and Non-EM S Regim es: Im plications for C urrency B lo cs,” C a to J o u rn a l (Novem ber 1990), pp. 4 5 5 -8 3 . ________ . “ U.S. Policy in the Bretton W oods E ra,” this R e v ie w (M ay/June 1991), pp. 5 4 -8 3 . ________ , and Saranna Robinson. “ S tability Under the Gold S tandard in P ractice,” in M ichael D. Bordo, ed. M o n e ta ry, H is to ry a n d in te rn a tio n a l F in a n c e : E ssa ys in H o n o r o f A n n a J. S c h w a rtz (U niversity of Chicago Press, 1989), pp. 163-95. M ilward, Alan S. The R e c o n s tru c tio n o f W estern E uro pe 1 9 4 5 -1 9 5 1 (U niversity of C alifornia Press, 1984). FEDERAL RESERVE BANK OF ST. LOUIS Rockoff, Hugh. “ Som e Evidence on the Real Price of Gold, Its Cost of Production, and C om m odity P rice s,” in M ichael D. Bordo and A nna J. Schwartz, eds. A R e tro s p e c tiv e on th e C la s s ic a l G o ld S ta n d a rd , 1821 -1931 (U niversity of Chicago Press, 1984), pp. 6 1 3-50. Rogoff, Kenneth. “ The O ptim al Degree o f C o m m itm ent to an Interm ediate M onetary T a rg e t,” Q u a rte rly J o u rn a l o f E co n o m ic s (Novem ber 1985a), pp. 1169-90. ________ “ Can International M onetary Policy C ooperation Be C o u nterproductive?” J o u rn a l o f In te rn a tio n a l E c o n o m ic s (M ay 1985b), pp. 199-217. Roll, R ichard. “ Interest Rates and Price E xpectations During the Civil W a r,” J o u rn a l o f E c o n o m ic H is to ry (June 1972), pp. 4 7 6 -9 8 . Rueff, Jacques. “ Increase the Price of G o ld ,” in Lawrence H. O fficer and T hom as D. W illett, eds. The In te rn a tio n a l M o n e ta ry S y ste m : P ro b le m s a n d P ro p o s a ls (Prentice Hall, 1964), pp. 179-85. Salant, Stephen W. “ The V u ln e ra bility of Price Stabilization Schem es to Speculative A tta ck,” J o u rn a l o f P o litic a l E co n o m y (February 1983), pp. 1-38. Scam m ell, W illiam M. In te rn a tio n a l M o n e ta ry P o lic y : B re tto n W oods a n d A fte r (H alsted Press, 1975). Schwartz, Anna J. “ In tro d u ctio n ,” in M ichael D. Bordo and Anna J. Schwartz, eds. A R e tro s p e c tiv e o n th e C la s s ic a l G o ld S ta n d a rd , 1 8 2 1 -1 9 3 1 (U niversity of Chicago Press, 1984), pp. 1-22. ________ “ A lternative M onetary Regim es: The Gold S tand a rd ,” in C olin D. C am pbell and W illiam R. Dougan, eds. A lte rn a tiv e M o n e ta ry R e g im e s (The Johns Hopkins U niver sity Press, 1986), pp. 4 4 -7 2 . Solom on, Robert. The In te rn a tio n a l M o n e ta ry S ystem , 1 9 4 5 -1 9 7 6 : An In sid e r’s View (Harper and Row, 1976). 187 Stockm an, Alan C. “ Real Exchange Rates U nder Alternative N om inal Exchange Rate S ystem ,” J o u rn a l o f In te rn a tio n a l M o n e y a n d F in a n c e (August 1983), pp. 147-66. ________. “ Real Exchange-Rate V a riability Under Pegged and Floating Nom inal Exchange-Rate System s: An E quilibrium T he o ry,” C a rn e g ie R o c h e s te r C o n fe re n c e S eries o n P u b lic P o lic y (Autum n 1988) pp. 2 5 9-94. W illiam s, John H. Extract from A Paper On “ The Adequacy of Existing C urrency M echanism s U nder Varying C ircu m sta n ces,” in K. H orsefield, ed. In te rn a tio n a l M o n e ta ry F u n d : 1 9 4 5 -1 9 6 5 . Volum e III: D ocum ents (international M onetary Fund, 1969a), pp. 119-23. Tem in, Peter. L e s s o n s fro m th e G re a t D e p re s s io n (M as sachusetts Institute of T echnology Press, 1989). ________ . Extract from “ C urrency Stabilization: The Keynes and W hite P lans,” in K. H orsefield, ed. The In te rn a tio n a l M o n e ta ry F u n d : 1 9 4 5 -1 9 6 5 . Volum e III: D ocum ents (Inter national M onetary Fund, 1969b), pp. 124-27. Tew, Brian. The E v o lu tio n o f the In te rn a tio n a l M o n e ta ry S y s tem , 1 9 4 5 -7 7 , 4th ed. (Hutchinson, 1988). W illiam son, John. The E x c h a n g e R a te S yste m (Institute for International Econom ics, 1985a). Townsend, Robert M. “ The Eventual Failure of Price Fixing S chem es,” J o u rn a l o f E c o n o m ic Theory (February 1977), pp. 190-99. ________. “ On the System in Bretton W o o d s,” A m e ric a n E co n o m ic R e v ie w (M ay 1985b), pp. 7 4 -9 . T riffin, Robert. E u ro p e a n d the M o n e y M u d d le (Yale Univer sity Press, 1957). ________ G o ld a n d th e D o lla r C risis (Yale U niversity Press, 1960). Yeager, Leland B. In te rn a tio n a l M o n e ta ry R e la tio n s : Theory, H is to ry P o licy, 2nd ed. (H arper and Row, 1976). ________ “ The Im age of the G old S ta n d a rd ,” in M ichael D. Bordo and A nna J. Schwartz, eds. A R e tro s p e c tiv e o n the C la s s ic a l G o ld S ta n d a rd , 1 8 2 1 -1 9 3 1 (U niversity of Chicago Press, 1984), pp. 6 5 1-70. MARCH/APRIL 1993 FEDERAL Appendix Table 1 RESERVE Supply (Permanent) and Demand (Temporary) Shocks: 1880-1989 Annual Data: Standard Deviations of Shocks (percent); Dispersion of shocks across countries (percent) BANK Gold Standard O ST. LOUIS F Dem and United States United Kingdom C anada Italy G4 G 4* Dispersion Supply 2.36 3.09 1.01 4.79 2.23 2.10 2.69 4.77 3.09 2.80 5.00 3.53 3.30 4.30 Dem and U nited States U nited Kingdom C anada Italy G4 G 4* D ispersion Supply 5.06 2.96 3.65 10.98 3.66 3.79 3.93 7.76 3.18 3.55 7.95 5.87 5.89 4.60 4.27 4.34 6.47 10.60 4.53 4.30 3.87 Demand Supply Demand 9.36 4.32 6.45 8.50 6.82 6.64 7.03 6.73 6.18 6.28 11.56 6.73 6.35 4.15 6.31 5.23 8.01 6.46 4.67 5.12 5.50 2.22 5.59 2.31 18.29 2.49 1.99 5.62 Bretton W oods (Convertible) 1959-1970 Demand 1.15 1.71 0.97 2.71 0.95 0.90 1.61 1940-1945 Supply Supply 1.87 1.32 1.56 2.21 1.50 1.47 2.10 Floating Exchange 1.46 3.27 2.41 3.43 1.41 1.56 2.07 7.10 7.15 6.68 4.92 5.91 6.30 6.96 Demand Supply 3.69 2.42 2.67 8.15 2.67 2.76 2.82 5.72 2.47 2.73 5.86 4.33 4.38 3.40 Post W W II 1946-1989 1973-1989 Demand Supply Bretton W oods (Total) 1946-1970 188 Bretton Woods (Preconvertible) 1946-1958 Demand W orld W ar II 1919-1939 1914-1918 1883-1913 Interw ar W orld W ar 1 Supply 2.47 4.81 2.71 2.63 2.56 2.37 2.05 Demand 2.92 2.85 2.59 6.78 2.25 2.34 2.57 Supply 4.55 3.49 2.75 5.01 3.60 3.59 2.79 G4: G 4-aggregate data G 4 *: W eighted average of in d ivid u a l country shocks; the weights are calculated as the share of each co u n try’s N ational Incom e in the Total Incom e in the G4 countries, w here th e G N P/G D P data are converted to dollars using the actual exchange rate. D ispersion = 2 (w eightj(sho ck|- 2 w e ig h t,'sh o ck,)2)0-5 for i = United States, United Kingdom , Canada, Italy. 189 A p p e n d ix F ig u re 1 Supply and Demand Shocks: 1880-1989, Including the War Years, Annual Data, United States Percent 20- Supply ------ 1-------- 1-------- 1 --------- 1-------- 1-------- 1-------- 1-------- 1-------- 1-------- 1 — 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 Supply and Demand Shocks: 1880-1989, Including the War Years, Annual Data, United Kingdom Percent -------1-------- 1 --------- 1-------- 1-------- 1-------- 1 --------- 1 --------- 1 — 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 M ARCH/APRIL 1993 190 Appendix Figure 1 (continued) Supply and Demand Shocks: 1880-1989, Including the War Years, Annual Data, Canada Percent 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 Supply and Demand Shocks: 1880-1989, Including the War Years, Annual Data, Italy Percent -40 1880 1890 1900 1910 FEDERAL RESERVE BANK OF ST. LOUIS 1920 1930 1940 1950 1960 1970 1980 191 Appendix Figure 1 (continued) Supply and Demand Shocks: 1880-1989, Including the War Years, Annual Data, G4 Percent 12 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 M ARCH/APRIL 1993 Manfred J. M. Neumann M a n fre d J .M . N e u m a n n is the d ire c to r o f the In s titu te fo r In te r n a tio n a l E c o n o m ic s a n d a p ro fe s s o r o f e c o n o m ic s a t the U n iv e rs ity o f B onn. Commentary J V i l C H A E L BORDO PROVIDES US with a comprehensive, scholarly study o f the history of the three main international monetary regimes: the gold standard, the dollar standard, and the floating exchange rate. He focuses on two important questions. First, which regime provided the best performance with regard to the levels of inflation and real growth? Second, what makes an international monetary regime viable? Because I am not a historian, I will limit my comments to two areas. I will first discuss the comparative evidence on the performance of the three monetary regimes and use Bordo’s statistics to infer a little more information on the role of demand shocks under the different regimes. Thereafter I will concentrate on the important issue of determining a monetary system’s credibility. I find Bordo's thoughtful discussion of the issue useful. I should add, however, that sometimes he takes the literature too seriously—especially the affirmative litera ture on the European Monetary System (EMS). Nevertheless, Bordo forces us to consider which monetary system or standard, if any, can solve the credibility problem in terms of firmly anchoring market expectations about its viability. 1The G roup of Seven countries are Canada, France, Germany, Italy, Japan, the United Kingdom and the United States. FEDERAL RESERVE BANK OF ST. LOUIS W H IC H REGIME PERFORM ED BEST? It is natural to evaluate the welfare implica tions of monetary regimes by asking what different regimes achieve with respect to the level and stability o f inflation and real growth. Any monetary regime can be described as a mechanism or device that delivers an average rate of monetary expansion and a variance of money growth. With respect to economic per formance, the essential difference is whether a particular regime provides governments with more or less freedom to manipulate the aver age rate of and the variance of monetary expan sion. It follows that regime differences should be reflected in inflation levels and variances of inflation and per capita growth. Table 1 draws from Bordo's tables 1 and 4. I consider the Group of Seven countries as a whole, the United States, Germany and France and concentrate on the three major periods: the pre-World W ar I gold standard, the Bretton Woods system of the 1950s and 1960s, and the floating exchange rate in place since the mid-1970s.1 Note that, in contrast to Bordo, I do not separate out the favorable performance of the Bretton Woods convertible subperiod 193 Table 1 Inflation, Real Growth and Shocks in Different Monetary Regimes for the Group of Seven Countries, the United States, Germany and France1 ______________________________________ Gold Standard 1881-1913 Bretton W oods 1946-1970 Floating 1974-1989 Mean Var Mean Var Mean Var Inflation Mean G roup of Seven United States G erm any France 1.0 0.3 0.6 -0 .0 11.56 9.61 6.76 24.01 3.6 2.4 2.7 5.6 21.16 6.76 16.00 16.81 7.1 5.6 3.3 8.8 10.24 5.76 1.69 10.24 Per capita growth Mean G roup of Seven United States G erm any France 1.5 1.8 1.7 1.5 13.69 26.01 8.41 21.16 4.2 2.0 5.0 3.9 7.29 7.84 10.89 4.41 2.1 2.1 2.1 1.7 5.29 7.29 3.61 2.25 Dem and shocks Mean G roup of Seven United States G erm any France 8.64 4.12 5.62 20.98 7.91 5.43 8.29 12.25 6.26 2.96 2.76 3.72 Supply shocks Mean G roup of Seven United States G erm any France 9.27 14.52 5.38 14.06 4.47 2.37 7.02 3.06 5.01 3.76 1.93 2.31 'T h e se data com e from tables 1 and 4 in the Bordo article in this R e view . The 1.69 estim ated variance of inflation for G erm any in the floating exchange rate period is based on a standard deviation of 1.3. This differs from the 1.2 standard deviation in B ordo’s table 1 because of differences in rounding of the standard deviation. The G roup of Seven countries are C anada, France, Germ any, Italy, Japan, the United Kingdom and the United States. (1959-1970) in terms of inflation and output. The subperiod looked good on the surface; how ever, it was in fact the period when the break down of Bretton Woods was programmed. More generally speaking, for any regime we might find e,x post a good looking subperiod.2 As Bordo and others have pointed out, the data permit the following observations: • First, average inflation was negligible under the gold standard and highest under the floating exchange rate. • Second, the variability of inflation, as well as that of real growth, was higher under the gold standard than under the floating exchange rate. 2As Anna Schwartz pointed out in the discussion, an evalu ation of the EMS that bypasses the m ost recent period, • Third, the Bretton Woods regime exhibited the highest variability of inflation, whereas output variability was closer to its level under the float than under the gold standard. The first observation on average inflation per formance is well known and understood. It is widely accepted that the classical gold standard prevented the manipulation of monetary expan sion by enforcing a direct link between the base money stock, the national stock of gold and the balance of payments. Though devaluation was possible by raising the gold parity in national currency, it was rare. Thus the gold standard deliv ered the lowest average rate of inflation, given that the available gold stock did not grow much. when the EMS cam e close to collapse, would be seriously m isleading. MARCH/APRIL 1993 194 At the other extreme, fiat money cum floating does not put any external constraint on domes tic money production. Thus governments are free to use money production to collect inflation tax and to dampen the business cycle. The addi tional advantage to governments of the floating exchange rate is that the regime spares them the political cost of negotiating devaluation. In sum, the floating exchange rate is the monetary regime most conducive to inflationary policies. Finally, the Bretton Woods system was in between the gold standard and the floating exchange rate in that it started as a gold exchange standard but was permitted to degenerate into a pure fiat money standard (the dollar standard) during the early 1960s when the United States gold stock fell short of the value of outstanding dollar liabilities. More interesting than the average inflation performance is the observed difference in the volatility of inflation and output growth among regimes. But to what extent can this volatility be attributed to the operation of the different monetary systems? EX PLO R ING THE ROLE OF DEM AND SHOCKS Apart from determining the level of inflation, monetary regimes differ with respect to nomi nal demand shock variability. I propose the following conjectures. First, nominal demand shock variability is highest under the floating exchange rate and lowest under the gold standard. This reflects the differences in the limits to monetary discre tion. Because the degree of monetary discretion is close to zero under the gold standard and unlimited under the floating exchange rate, we should observe that the variance of inflation was caused predominantly by nominal demand shocks under the floating exchange rate but by supply shocks under the gold standard. Second, in a fixed-exchange rate system the system leader sets the floor for nominal demand shock variability. Consequently, for the Bretton Woods period we should observe that nominal demand shock variability was lowest in the United States. Similarly, during the floating rate period nominal demand shock variability should have been lower in Germany than in any other 3See B ordo’s table 4. FEDERAL RESERVE BANK OF ST. LOUIS member country of the European snake or EMS. Checking the empirical validity of these con jectures requires estimating the variance of nominal demand shocks. Bordo’s study provides us with some valuable information in this respect. Following Blanchard and Quah (1989) and Bayoumi and Eichengreen (1992) in particu lar, he has estimated for each country and each monetary regime a bivariate vector autoregres sion (VAR) for the rates of change of the price level and output. The lower panel of table 1 provides the variances of the estimated aggregate supply and demand shocks.3 Under the straight forward assumption that the distribution of real demand shocks was the same over the different monetary regimes, differences in demand shock variability can be attributed to the operational differences of the regimes. The empirical findings are mixed. The data reject our first conjecture. For the Group of Seven countries demand shock variability was highest under the gold standard and lowest under the potentially permissive floating exchange rate regime. The most puzzling aspect is the high demand variability during the gold stand ard period because not only was monetary pol icy discretion constrained by the rules of the regime, but also fiscal discretion was negligible, at least by today’s standards. Our second conjecture, in contrast, is con firmed. Demand shock variability was lowest in the United States during the Bretton Woods period and in Germany during the float. More over, it can be shown for Germany using an F-test that the demand shock variance o f the float differed significantly from its value under Bretton Woods at the 1 percent level of sig nificance. In the United States the level of significance was 10 percent. Though we have not seen any test statistics of Bordo’s VAR estimates, let us assume that the estimates are clean. On this assumption we may use them to investigate the contribution of the aggregate demand shocks to the variability of inflation and output growth under the different monetary regimes. To do so requires a model of aggregate supply and demand to determine the unknown price elasticities of aggregate demand and supply. Table 2 provides the bare bones o f such a model. The model is written in logs and has a 195 Table 2 A Minimal Structure (1) y = 6 + a (p (2) y = p (m - E _ ! p) p) O utput supply O utput dem and (3) 0 = 8 _ i + s Productivity (4) m = m _ , + d M oney stock Solutions (5) it = p - p_, =d _ ! - - y - s (6) Ay = y - y . , = s + [p (d - + ~ - + j IP < d -1 d _ ,) - (s - d - i) - (s - s _ ,)l s _ ,)] Variances (7 o* = fL ± Ji r02 + ?s!i ) w d (a + PY L , _ a2 + P2 (8) ohy - (a + p)2 °s P2 J , ^ 2 (a PY + (a + p)2 °d Table 3 The Contribution of Demand Shocks to the Variability of Inflation and Real Growth1 Bretton Woods Variance Gold Standard Variance Actual Adjusted Adjusted percentage of actual Actual Adjusted Adjusted percentage of actual Floating Exchange Rate Variance Actual Adjusted Adjusted percentage of actual Inflation Mean G roup of Four United States G erm any France 10.6 9.6 6.8 24.0 4.8 3.5 3.3 12.0 (45) (36) (49) (50) 12.1 6.8 16.0 16.8 5.2 4.2 6.0 7.4 (43) (62) (38) (44) 6.1 5.8 1.7 10.2 2.5 2.4 1.5 2.4 (41) (41) (86) (24) Real growth Mean G roup of Four United States G erm any France 15.9 26.0 8.4 21.2 8.8 13.8 5.2 13.1 (55) (53) (59) (62) 7.5 7.8 10.9 4.4 4.4 6.0 5.8 2.6 (59) (77) (53) (59) 4.7 7.3 3.6 2.3 2.8 4.2 2.6 0.8 (60) (58) (72) (35) ’ The Group of Four countries are Canada, France, Germ any and the United States. The adjusted variance excludes the contribution of supply shocks. Lucas-type supply equation and an aggregate demand equation. The evolution of prices and output is driven by productivity and the money supply (both modeled as random walks) with shocks d and s assumed to be independently distributed. The model's solutions [equations (5) and (6)] show that it meets the restrictions used in Bordo’s VAR estimates. Supply shocks have permanent effects on the price level and output, whereas demand shocks have no permanent effect on output. Given the variances of inflation, real growth, demand and supply shocks, equations (7) and (8) can be used to compute the slope coefficients in a p - y plane o f aggregate demand, - 1 1(), and aggregate supply, 1la. Solving by numerical iteration does not yield real solutions in all MARCH/APRIL 1993 196 cases—notably the Bretton Woods convertible subperiod.4 Given the estimates o f the slope coefficients, we can compute the contribution of the variance o f aggregate demand shocks to the observed variances of inflation and real growth in table 3. Table 3 presents for each monetary regime the measured variances of inflation and real growth, as well as adjusted variances, which exclude the contribution to volatility of the aggregate supply shocks. The numbers printed in parentheses indicate the percentage share in the measured variance of the contribution of the demand shock variance. Note that data from only four of the Group of Seven countries are included; data from Italy, Japan and the United Kingdom had to be removed because it was impossible to compute the slope coefficients for these countries (in at least one subperiod). Table 3 can be summarized as follows: • First, for the United States, the leader of the Bretton Woods regime, we find that the variances of inflation and real growth were dominated by the volatility of demand shocks during this period. About 62 per cent of the variance of inflation and 77 percent of the variance of output growth can be attributed to the variance of demand shocks. Similarly, we find that for Germany under the floating regime the inflation and the output variance were dominated by demand shocks, which accounted for 86 percent of the inflation variance and 72 p ercen t o f the output variance. • Second, for the four Group of Seven countries we find that demand shocks produced a higher inflation variance under Bretton Woods (5.2) than under the gold standard (4.8) or the floating exchange rate (2.5). The result probably reflects the differential performance of the two leading countries. • Third, for the four Group of Seven countries as a whole, the variance of demand shocks did not dominate the inflation variance under any of the three monetary regimes. Its contribution never exceeded 45 per cent. Thus we find over all regimes that the inflation variance was dominated by 4O f the up to four real solutions for each case, I chose the one w hich com bines a negative slope of aggregate dem and w ith a positive slope of aggregate supply. FEDERAL RESERVE BANK OF ST. LOUIS the volatility of aggregate supply shocks. This is a little surprising. Are we prepared to accept that systematic differences in the level of demand shock variability are not a characteristic feature o f international monetary systems? W e cannot, however, rule out that these findings are statistical artifacts enforced by an inability to separate demand from supply innovations accurately in the VAR estimation. Bordo himself has noted that in some cases the overidentifying restriction (according to which positive supply shocks should permanently raise output and drive down the price level) is not satisfied. Another indication of a possibly insufficient identification is the estimated change in the slopes o f aggregate supply and demand between regimes. Figure 1 presents the average slopes of aggregate supply and demand for the four Group o f Seven countries. What effect do we expect monetary regimes to have on these slopes? Consider the model printed in table 4 which provides more structure than the model in table 2. Because the model is linear in logs, the size of the alpha and beta coefficients depends on the agents’ price responsiveness, as well as on the share in output of the respective input in the production function or of the respective expenditure. Comparing the regimes of the gold standard and Bretton Woods periods, we find that both aggregate supply and demand schedules were steeper under Bretton Woods. I would have expected the opposite on the argument that the economies were generally more open to interna tional trade under Bretton Woods than before. Comparing the Bretton Woods regime with the float, we observe that the aggregate supply schedule became steeper under the floating exchange rate but the aggregate demand sched ule became more flat. The first observation is in line with the model in table 4 because the posi tive dependence o f the nominal exchange rate (its log is denoted by e) on the domestic price level implies a steeper aggregate supply schedule. For the same reason, the demand schedule should also be steeper under floating. However, the data 197 Figure 1 G-4 Output Figure 2 United States Output M ARCH/APRIL 1993 198 Table 4 Slopes and Exchange Rate Regimes Model ( 1 ) y d = 0 o g - P i [ i - ( E p c + i - P c)l + f t > ( P * + e - p ) ; (2) ys = e + o 1( p - E _ 1p ) - o 2(pR* + e - p ) (3) pc = yp + (1 - y ) ( p * + e ) (4) m = p + y - A i (5) i = i* + E e + 1 - e Slopes Fixed rates: e = 0 yd : —— — P +P -\Y 2 y • — + a2 a1 I— Flexible exchange rates (fcr*fe) 0 - If) * 0 If . ys : --------------------------------- "1+<,2 0 " I p ) do not comply. Also note that the United States data imply that both schedules are more flat under the float. See figure 2. In sum, I agree with Bordo that his VAR esti mates should be viewed with great caution. Moreover, because we are after the differential effect of monetary regimes, a serious drawback is that we cannot differentiate between nominal demand shocks, which we wish to study, and real demand shocks, which are irrelevant because they are not caused by the monetary regime. Also, I must emphasize that we are studying international regimes, which implies that we cannot treat countries as independent entities. Monetary regime shocks are transmitted inter nationally. For example, a nominal demand shock produced by the Fed will show up in Ger many as a demand shock that raises German output temporarily and German prices perma nently. At the same time, however, the shock will show up in Germany as a supply shock, changing the relative price of imported raw materials. This reduces German output perma FEDERAL RESERVE BANK OF ST. LOUIS nently and raises German prices permanently. Consequently, the identifying restrictions of Bordo’s VAR estimates will classify the nominal demand shock from the United States as a sup ply shock in Germany. In conclusion, I believe we have to make another, more sophisticated attempt at inves tigating the conjecture that international mone tary regimes systematically differ with respect to the variability they impose on world economies. THE CREDIBILITY PR O BLEM I now take up the fundamental question of which international monetary system, if any, can solve the credibility problem in the sense of firmly anchoring market expectations about the viability of the system? Bordo’s careful examination of the history of monetary regimes leads him to conclude that an international monetary system will be stable if its rules are credible. The rules will be credible if the member countries of the system are ready 199 to honor the rules. And member countries will honor the rules if there is a center country that enforces the rules. Accordingly, the classical gold standard did not break down because the United Kingdom, its center country, was com mitted to convertibility. In contrast, the United States, as the center country of Bretton Woods, was not committed to convertibility and main tenance of price stability. Bretton Woods consequently broke down. Finally, Germany’s commitment to price stability made the EMS a successful and viable system. Unfortunately, the latter prediction held only until last September. Though Bordo’s reasoning makes a lot of sense, it fails to address two essential questions. First, by which means or under what conditions will the center country be able to enforce the rules? Second, and more fundamentally, what conditions are required to make the center country keep its commitment? In my view, any international monetary sys tem that is based on commitment to rules will be fragile. Commitment should be replaced by precommitment. The game theory reformulation of the pathbreaking analysis by Kydland and Prescott proves that governments cannot com mit to price stability.5 In contrast to commit ment, precommitment does not depend on a government's good will or interest. Instead it is created by setting up an external mechanism that effectively ties the hands of current and future governments. An international monetary regime will be sta ble and therefore durable if it provides the institutional constraints for a subgame-perfect supergame. The fundamental constraint is effec tive precommitment by all member govern ments. There are two types of precommitment: precommitment to price stability at home and precommitment to a fixed exchange rate vis-avis another currency. Consequently, we can design two alternative regimes. A first regime resembles the EMS but commit ment is replaced by precommitment. The center country precommits on price stability at home by providing its central bank with the constitu tional status of independence from government. Elsewhere I have laid out a sufficient set of institutional elements that provides an incentivecompatible status of independence.6 The other countries precommit on a fixed exchange rate vis-a-vis the center currency by writing the fixed exchange rate into the country’s constitu tion as Sweden did during the gold standard. The alternative international regime is created by an agreement that all governments precom mit to price stability at home by providing their central banks with constitutional independence. Which of the two regimes is preferred? The first regime provides price stability for all coun tries in the medium to long run. The precom mitment to fixed exchange rates by n-1 members implies that idiosyncratic shocks will be dis tributed over member countries at full force, as was the case under the classical gold standard. Because fiscal policy is an important source of idiosyncratic shocks, the regime will hardly be attractive without a (enforceable) rule prohibit ing public deficits. The alternative regime of uniform precommit ment to price stability at home might be rejected by some as a nonsystem. But semantics apart, the setup is not to be equated with unconstrained floating. The regime will provide nominal ex change rate stability though not fixity. Depending on the judgment of central bankers, the regime might evolve into an adjustable peg system where up to n-1 central banks unilaterally peg their currencies to the currency of a center country in a flexible manner. This means that in the ad vent of a sizable country-specific shock at home or in the center country, they will permit exchange rate adjusting. In contrast to the non-precommitted governments in Europe, the independent central bankers will have no interest in defending mis aligned exchange rate parities. In conclusion, the Bretton Woods system and the EMS broke down because both systems were built on unenforceable commitment instead of precommitment. REFERENCES Bayoum i, Tam in, and Barry Eichengreen. “ Shocking Aspects of European M onetary U n ifica tio n ,” NBER W orking Paper No. 3949 (January 1992). Blanchard, O livier, and Danny Quah. “ The Dynam ic Effects of Aggregate Dem and and A ggregate S upply D isturbances,” A m e ric a n E c o n o m ic R e v ie w (Septem ber 1989), pp. 6 5 5-73. Kydland, F.E., and E.C. Prescott. “ Rules Rather than Discretion: The Inconsistency of O ptim al P lans,” J o u rn a l o f P o litic a l E cono m y, (June 1977), pp. 4 7 3-91. Neum ann, M .J.M . “ P recom m itm ent by C entral Bank Inde pen de n ce ,” O pe n E co n o m ie s R e v ie w (1991), pp. 9 5-112. 5See Kydland and Prescott (1977). 6See N eum ann (1991). MARCH/APRIL 1993 Federal Reserve Bank of St. Louis Post Office Box 442 St. Louis, Missouri 63166 The Review is published six times p e r y ea r by the Research and P u b lic In fo rm a tio n Departm ent o f the Federal Reserve Bank o f St. Louis. Single-copy subscriptions are available to the p u b lic f r e e o f charge. M ail requests f o r subscriptions, back issues, o r address changes to: Research and P u b lic In fo rm a tio n Departm ent, Federal R eserve Bank o f St. Louis, P.O. B ox 442, St. Louis, M issouri 63166. The vietvs expressed are those o f the individual authors and do n o t necessarily re fle c t o ffic ia l positions o f the Federal R eserve Bank o f St. Louis o r the Federal Reserve System. A rticles herein may be reprinted p rovid ed the sou rce is credited. Please p rovid e the B ank’s Research and P u b lic In fo rm a tio n D epartm ent with a co p y o f rep rin ted material.