The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Money Stock Measurement History, Theory and Implications Proceedings of the Eighteenth Annual Economic Policy Conference o f the Federal Reserve Bank o f St. Louis THE FEDERAL J RESERVE BANK of A r ST.IiH LS i Federal R eserve Bank o f St. Louis R e v ie w March/April 1994 In This Issue . . . iii v vii 1 Contributing Authors President’s Message Thomas C. Melzer Editor's Introduction Richard G. Anderson A H istorical P erspective on the Federal R eserve’s M onetary A ggregates: Definition, C onstruction and Targetin g Richard G. Anderson and Kenneth A. Kavajecz 32 The E volution o f the F ederal R eserve’s M onetary Aggregates: A Tim eline Kenneth A. Kavajecz 67 Commentary Charles W. Calomiris 73 E m pirical Evidence on the Recent B eh avior and U sefuln ess of Sim ple-Sum and W eighted M easures o f the M oney Stock K. Alec Chrystal and Ronald MacDonald 110 Commentary Charles R. Nelson 117 M oney D em and in a Flexible D ynam ic F o u rie r E xpenditure System Douglas Fisher and Adrian Fleissig 129 Commentary James L. Swofford 133 Financial F irm s’ P ro d u c tio n and Supply-Side M onetary A ggregation U n d e r Dynam ic Uncertainty William A. Barnett and Ge Zhou 166 Commentary William C. Brainard 169 Response to Commentary William A. Barnett and Ge Zhou MARCH/APRIL 1994 175 Can the Central Bank A chieve Price Stability? Jerome L. Stein 205 Commentary Frederic S. Mishkin 209 A Conference P an el D iscussion Michael J. Boskin, Philip H. Dybvig and Bennett T. McCallum http://fraser.stlouisfed.org/ FEDERAL RESERVE Federal Reserve Bank of St. Louis BANK OF ST. LOUIS iii Contributing Authors R ichard G. A n d e rso n Research & Public Information Department Federal Reserve Bank o f St. Louis St. Louis, MO 63102 W illiam A. Barnett Department of Economics Washington University St. Louis, MO 63130 M ichael J. B osk in American Enterprise Institute Washington, D.C. 20036 W illiam C. B ra in ard Department of Economics Yale University New Haven, CT 06520 Ch arles W. Calom iris Department of Finance University o f Illinois Urbana, IL 61820-6271 K. Alec Chrystal Centre for Banking The City University of London London EC2 England P h ilip H. D y b v ig Graduate School of Business Washington University St. Louis, MO 63130 D o u glas Fisher Department of Economics North Carolina State University Raleigh, NC 27695-8109 Kenneth A. Kavajecz Kellogg Graduate School of Management Northwestern University Evanston, IL 60201 R onald M acDonald Department of Economics University o f Strathclyde Glasgow G4 OLN Scotland Bennett T. M cCallum Graduate School of Industrial Administration Carnegie-Mellon University Pittsburgh, PA 15213 Federic S. M ishkin Graduate School o f Business Columbia University New York, NY 10027 Charles R. Nelson Department of Economics University of Washington Seattle, W A 98105 Jerom e L. Stein Department o f Economics Brown University Providence RI 02912 James L. S w o ffo rd Department of Economics and Finance University o f South Alabama Mobile, AL 36688 A d ria n Fleissig Department of Economics North Carolina State University Raleigh, NC 27695-8109 MARCH/APRIL 1994 V President’s Message The Federal Reserve Bank o f St. Louis has long emphasized the central role o f monetary aggregates in the conduct of monetary policy. Appropriate long-run growth of the supply of money is essential to attaining the principal goal of monetary policy—price stability—which, in turn, is necessary to maximize sustainable growth in the economy. And the monetary ag gregates have, in fact, played important roles as indicators of monetary policy in connection with the actions leading to the deceleration of U.S. inflation over the last several years. Translating the abstract scientific concept of "money” into a measurable empirical counter part has often been controversial. Obviously, a monetary aggregate must include the exchange medium of the economy, including currency and checkable deposits. The spending plans of households and firms are also affected by and reflected in the quantities of other liquid assets that they hold. How many of these should be included in a monetary aggregate? And, by what criteria should they be selected? Studies o f the measurement o f money tell us that the answers to these questions change through time. Financial innovation changes in stitutional arrangements and practices, forcing us to revise our measures o f money. In some cases, new financial instruments such as money market mutual funds are added to the monetary aggregates. In other cases, increasing similarity among financial institutions may require a major change in the institutional coverage of the aggregates. W e experienced both of these phenomena in the 1970s. Technological progress continues to cloud our measures of money by reducing the transaction costs of quickly ex changing one asset for another. in the Federal Reserve Bulletin. Twenty-five years ago, the St. Louis adjusted monetary base first appeared in this Bank’s Review. Through widely circulated publications such as Monetary Trends and U.S. Financial Data, Homer Jones and his successors have disseminated monetary ag gregates data to a worldwide audience of analysts and researchers. Research at St. Louis has also examined the policy implications of linkages between various monetary aggregates and nominal economic variables. This year, for example, marks the 25th anniversary of publication in the St. Louis Fed’s Review o f Leonall Andersen and Jerry Jor dan’s seminal research linking the growth of nominal income to the growth o f the M l mone tary aggregate. Today, some suggest that the monetary ag gregates may no longer be useful guides for monetary policy. The weaker-than-anticipated growth of M2 during the recent economic recovery and expansion has brought to the fore front once again issues regarding the measure ment, modeling, and continued policy usefulness of monetary aggregates. I am hopeful that the presentations and discussions presented at this conference will improve our understanding of the issues involved in measuring money and the implications of alternative measures for the con duct of monetary policy. Thomas C. Melzer President and Chief Executive Officer Federal Reserve Bank o f St. Louis Our staff in St. Louis has contributed in a number o f ways to the measurement o f mone tary aggregates. Thirty-three years ago, William Abbott's revised M l monetary aggregate appeared MARCH/APRIL 1994 vii Editor’s Introduction Monetary aggregates have played a prominent role in policy research at the Federal Reserve Bank of St. Louis for more than 25 years. The Bank’s 18th annual Economic Policy Conference in October 1993 brought together a variety of evidence on the interaction between the use of monetary aggregates in policymaking and mea surement o f the money stock. The first session of the conference addressed issues in the construction o f monetary aggregates. Milton Friedman and Anna Schwartz have noted that measurement of the stock of money in the United States is an activity almost as old as the republic itself. Their well-known histories of these data, however, largely precede both the first modern monetary aggregates published by the Federal Reserve in 1960 and the aggregates used today in macroeconomic research. In the first paper presented at the conference, Richard Anderson and Kenneth Kavajecz review the his tory and construction of the Federal Reserve’s monetary aggregates. Following a broad introductory discussion of definitional and statistical issues, Anderson and Kavajecz trace the history of the Federal Reserve’s m o n eta ry aggregates since 1943. T h e y describe in detail the sources of data used in bu ildin g the current aggregates, cautioning the reader that a wide variety o f data are received and in corporated into the aggregates throughout the year. Because various Federal Reserve publica tions are released at different times, observations on a monetary aggregate in one publication may differ significantly from observations in another. Moreover, data in different issues of the same publication more than a year apart may not be comparable since the monetary aggregates are benchmarked each year to incorporate addition al incoming data and new seasonal adjustment factors. The authors find that these annual benchmarks often significantly change published growth rates for the monetary aggregates, although the size of the revisions is small except for the most recent years. The authors conclude with a summary of the Federal Reserve’s use of monetary aggregates as monetary targets. The article is followed by a unique timeline compiled by Kenneth Kavajecz that traces the history of the Federal Reserve’s monetary ag gregates from 1960-93. The date of each change in definition and benchmark revision is included, as well as descriptions o f many special events that affected the monetary aggregates. Of general interest to all readers, the chronology will be in valuable to researchers working with highfrequency data on the monetary aggregates. In his commentary, Charles Calomiris proposes a number o f reasons why empirical economists should be concerned about the construction of the monetary aggregates data that they use in their research. Since tests of many hypotheses in modern macroeconomics require long time series of data, researchers may be at risk by ignoring issues such as changes in sampling and seasonal adjustment procedures used by the data constructors. Further, the construction of long time series is complicated by the Fed’s fre quent retrospective revisions and redefinitions of the monetary aggregates. Calomiris also notes that the redefinitions discussed by Anderson and Kavajecz call into question the usefulness of the monetary aggregates for testing many pro positions in macroeconomics. If the redefinitions are motivated by a desire to make the new aggregate better track economic activity, then the redefined aggregates may not be suitable for tests o f the structural stability of macroeco nomic relationships, including money demand. A number of economists have argued over the last 15 years that simple-sum monetary aggregates of the type published by the Federal Reserve Board are not defensible in terms of either economic aggregation or statistical index number theory. These researchers have suggested a number o f alternative measures of the money stock including the Divisia monetary aggregate MARCH/APRIL 1994 viii proposed by Barnett and the currency-equivalent aggregate suggested by Rotemberg. In the con ference’s second paper on the policy implica tions of differing measures of the money stock, K. Alec Chrystal and Ronald MacDonald com pare the indicator properties o f simple-sum aggregates to those o f alternative measures of money in seven industrialized countries. The authors' first set o f tests is based on a variant of the classic St. Louis reduced-form equation for nominal output. Perhaps as might be expected, the results show little difference between the indicator properties of narrow simple-sum and Divisia aggregates. For broader aggregates, however, the Divisia aggregates are generally found to be preferable to simple-sum aggregates. Next, the authors conduct a series of sophisticated multivariate causality tests based on estimated error-correction models. These tests also suggest that Divisia aggregates are preferred to simple-sum aggregates, although the results are not so strong as to find that a Divisia ag gregate has significant indicator value when a simple-sum aggregate does not. In a test on U.S. data since 1980—a period of extensive financial innovation—the authors find particularly strong support for the superior indicator properties of a Divisia M2 index relative to the simple-sum M2 aggregate. In his commentary, Charles Nelson notes that the authors' specification of the St. Louis equa tion for the U.S. is not comparable to that for their other countries, with the form er better seen as a structural demand equation and the latter as reduced-form equations. For the U.S., although differences between results based on alternative various M l and M2 aggregates may be reasonable, he finds puzzling the sharp differ ences among results for M l and Divisia M l and M IA when one might have expected the three aggregates to closely resemble each other. Nelson also questions the authors’ causality inferences drawn from their estimated errorcorrection models. Emphasizing that monetary aggregates enter the error-correction models through both the first differences of their growth rates and the error-correction terms (which are specified in growth rates rather than levels), he suggests that Chrystal and MacDonald’s emphasis solely on the significance o f the coefficients on the first differences o f growth rates may be misplaced. Strong significance of the errorcorrection terms in some equations suggests http://fraser.stlouisfed.org/ FEDERAL St. Louis Federal Reserve Bank of RESERVE BANK OF ST. LOUIS more o f a role for monetary aggregates than the authors perhaps recognize. The papers presented in the second session ad dressed a pair of econometric issues in measure ment o f monetary aggregates. Financial assets, like other goods, are demanded by households because they yield a flow o f services. This simple insight suggests the potential value o f analyzing the demand for money in the context of a multi variate expenditure system, rather than as a sin gle isolated demand equation. Despite its intuitive appeal, the expenditure system approach has had limited acceptance due to a number o f short comings. Most prominent perhaps has been un certainty regarding the correct functional form. This uncertainty has led to widespread use of flexible functional forms able to furnish (at least) a second-order approximation to the true un known function at (at least) one point. The Fourier flexible functional form proposed by Gallant solves the approximation problem by providing an arbitrarily accurate global approxi mation to any unknown function and its partial derivatives. Expenditure systems based on this functional form typically have been static, how ever, limiting their usefulness with economic time series data. Douglas Fisher and Adrian Fleissig propose and compare two dynamic ex tensions of the Fourier functional form. Their estimates o f dynamic expenditure systems that include monetary assets suggest that the dynamic models are more consistent with the data than the Fourier static model. In particular, the dynamic models seem to provide much sharper estimates o f the elasticities o f substitution be tween the various types of monetary assets held by households. No econometric model can be all things, but James Swofford concludes in his commentary that Fisher and Fleissig have done a commenda ble job of achieving the goals they set forth for their model. Their dynamic extension of the Fourier functional form is an important contri bution, likely of value to many future research ers. He notes, however, that although their elasticity estimates are plausible, many readers may find them difficult to interpret. The reader who is primarily interested in understanding household money demand may miss entirely the importance o f estimating expenditure systems if authors, including Fisher and Fleissig, fail to provide a thorough discussion o f their findings. Swofford also concludes that Fisher and Fleissig’s ix model fares laudably well against the very demanding criteria proposed by Carl Christ at last year’s St. Louis economic policy conference. The next paper addresses the relatively new topic of supply-side monetary aggregation. Meas ured money stocks in most economies are pri marily composed o f inside money or, in other words, of the liabilities of profit-maximizing firms. The supply-side aggregation conditions applicable to the monetary services produced by these liabilities differ from those more commonly studied in the demand-side monetary aggrega tion literature. Recognition o f the risk and un certainty facing these intermediaries further complicates aggregation, since existing economic aggregation conditions and index number theory (such as that for Divisia monetary aggregates) have usually considered only cases of perfect certainty. William Barnett and Ge Zhou introduce to the literature a stochastic model of monetary services production by banks under uncertainty. In the model, banks are treated as neoclassical competitive firms that maximize the present value o f expected utility. The banks contract for deposits and real factor inputs (labor, for exam ple) at the beginning o f each period. During the period, three variables—the economy’s average price level, reserve requirement ratios for each deposit type, and the ex post realized rate of return on loans—are determined by random processes not controllable by the firm. The em pirical results support the hypothesis that the banks' deposit liabilities are weakly separable from purchased real factor inputs such as labor. A comparison o f the Divisia, simple-sum, and currency-equivalent monetary aggregates to the model’s estimated exact monetary aggregate sug gests that the ability o f the Divisia index to track the exact aggregate is little diminished un der uncertainty. This conclusion is invariant to whether the exact aggregate is constructed from model estimates based on alternative as sumptions of risk neutrality and risk aversion. In his commentary, William Brainard notes the increasing importance o f studies of the sup ply of monetary assets. Unlike simpler times, when the money stock could be well measured by summing currency and demand deposits, to day’s relatively low costs of substituting among a wide variety of financial assets makes less cer tain both the measurement and control of monetary aggregates. Brainard notes, however, that the dynamic structure of the model may not be as rich as the authors suggest. In partic ular, the period-by-period balance sheet con straint imposed by Barnett and Zhou as equation 2 prevents the model firm from carrying re tained earnings (or losses) forward. Each period, the firm’s available resources include only the deposits and real inputs contracted for at the beginning o f that period plus a fixed amount of capital; in turn, all earnings must be paid out to the owners o f the firm at the end of the period since the balance sheet constraint prevents any from being carried forward into the next. He suggests that the apparent dynamic structure of the profit function in their equation 3 arises be cause Hancock’s profit function, equation 1 in Barnett and Zhou, differs from the cash flow that the firm will in fact receive in each period, conditional on its decisions and the stochastic nature o f the economy. This reservation aside, the richness o f Barnett and Zhou’s paper is reflected in the numerous extensions proposed by Brainard for future researchers. In a response to Brainard, Barnett and Zhou present additional results clarifying the dynam ics of their model. The model requires some type of temporal separability restriction on either the discounted profit stream or the inter temporal utility function to avoid the intractable problem of estimating a system of simultaneous Euler equations. The formulation employed by Barnett in previous work, and preferred by Brainard, appears as but one o f a number of al ternative separability hypotheses. The relative plausibility o f the hypotheses remains a subject for further empirical research. Papers at the conference’s final session once again turned to the implications of alternative measures o f the money stock for the conduct of monetary policy. Monetary policymakers often rank price stability first among their goals. Dur ing the 1970s, central banks worldwide adopted growth targets for monetary aggregates that they hoped would guide them toward price sta bility. In many countries, however, initial opti mism became disappointment as Goodhart’s law— that the behavior of a monetary aggregate will change when the central bank targets its growth —seemed to prevail. Jerome Stein studies whether Goodhart’s law has applied with equal force in the United States to all measures of the money stock. Working with the dynamic model he developed with Infante in the 1980s, Stein demonstrates that the short-run stability o f the linkage between inflation and money growth is MARCH/APRIL 1994 X apparent only when the model includes a varia ble representing the state of the economy, measured in his model by the difference be tween the current and long-run equilibrium un employment rates. In that case, the growth of M2 arises as a good indicator of movements in both inflation and unemployment. Further, M2’s indicator properties appear superior to those of statistical index number monetary aggregates, including Divisia M2, the currency-equivalent aggregate CE, and a Divisia CE aggregate. Regardless of its indicator value, a monetary ag gregate must be controllable before it can be chosen as a policy target. Stein concludes that none of the broad monetary aggregates are sufficiently controllable to be used as targets. He finds, however, that adjusted bank reserves appear to be an acceptable target for control of the inflation rate. Although monetary aggregates may be valua ble indicators of the stance of monetary policy, they are not necessary for central banks to achieve price stability. Agreeing with Stein that the long-run inflation rate is largely determined by growth o f the money stock, Frederic Mishkin notes that Federal Reserve policy has supported a relatively low, steady inflation rate during the last decade without strict adherence to any monetary target. He suggests that the highly dy namic nature of Stein’s model might help ex plain the relatively poor showing of M2 per se as an indicator for individual variables such as inflation and real output while being a valuable indicator for nominal GDP. Since real output growth accelerates more quickly following a monetary shock than inflation and later tends to slow while inflation accelerates, cyclical move ments in M2 may be more closely correlated with both short- and long-run movements in nominal GDP than with either inflation or real output separately. At the same time, Mishkin finds troubling the poor fit of the model to quarterly data which may indicate that Stein’s empirical surrogate model is not capturing well the dynamic interactions prominent in the SM theoretical model. Also puzzling are the very different conclusions reached by Stein and by Chrystal and MacDonald regarding the relative indicator properties of simple-sum and Divisia M2. Finally, Mishkin emphasizes that the omis sion o f rational expectations from Stein’s model prevents him from analyzing the importance of credibility in policymaking. Announced targets for monetary aggregates might help prevent sharp jumps in inflationary expectations by sig http://fraser.stlouisfed.org/ FEDERAL RESERVE Federal Reserve Bank of St. Louis BANK OF ST. LOUIS nalling the public that the central bank is serious about achieving its inflation targets. In this event, monetary aggregate targets might help the central bank stabilize the inflation rate even when measurement of the monetary aggregate is uncertain or monetary aggregates are not highly controllable. The conference concluded with a panel dis cussion o f the role of monetary aggregates in feedback rules for the conduct of monetary policy. Monetary aggregates have historically been constructed to guide monetary policy. The introduction of rational expectations into macroeconomic models emphasized that the feedback rules by which policymakers adjust growth of monetary aggregates are an important part of the structure o f the economy. In the panel discussion, Michael Boskin sug gests that Federal Reserve actions under Alan Greenspan, and to some lesser extent under Paul Volcker, should be viewed as a rules-based policy. He sees the Fed as setting out a strategy whereby its actions in most periods are governed by pursuit of its goal of long-run price stability, rather than by a feedback rule based on a monetary aggregate. Temporary deviations from pursuit o f the goal are permitted for ex igencies that are well understood by the public. Further, in his view, the Federal Reserve will never find satisfactory any policy rule that in cludes only a small set of monetary aggregates or similar indicacor variables. Behavioral rules arise naturally as solutions in decision-theoretic models. Could a monetary policy rule based on monetary aggregates arise as the solution to a decision problem? The se cond panelist, Philip Dybvig, proposes a com plete prototypical decision framework for the Fed, including an objective function, control variables, constraints and a well-defined infor mation set. Although too much of the structure remains unknown to obtain explicit solutions, he concludes that future research on the value of monetary policy rules and the role of mone tary aggregates might usefully be guided by such a framework. Some researchers have argued that monetary aggregates have little value as either policy tar gets or indicators. If so, discussion of their measurement seems vacuous. The third panelist, Bennett McCallum, concludes the conference by suggesting that monetary aggregates are indeed irrelevant to the conduct of monetary policy. In xi his framework, the central bank’s main job is to keep nominal GDP growing smoothly at a noninflationary rate. Even when the penultimate goal is price stability rather than stable growth of nominal output, he argues that we know much better what growth rate for nominal GDP is likely to be consistent with long-run price stability than we do the appropriate long-run growth rates for M l or M2. McCallum’s research sug gests that directly targeting the growth of nomi nal GDP through control of the monetary base is preferable to targeting any monetary ag gregate, no matter how measured. Richard G. Anderson St. Louis, Missouri April 8, 1994 MARCH/APRIL 1994 1 Richard G. Anderson and Kenneth A. Kavajecz, Richard G. Anderson is a research officer at the Federal Reserve Bank of St. Louis. Kenneth A. Kavajecz is a Ph.D. candidate in finance at Northwestern University. An earlier version of this paper was completed while the authors were in the Division of Monetary Affairs at the Federal Reserve Board. We wish to thank numerous former colleagues at the Board for their generous assistance and access to their unpublished writings, without which this study would not have been possible, including Sean Collins, Dennis Farley, David Lindsey, Leigh Ribble and Jack Walton. We thank Richard Kopcke for stressing the importance of regarding changes in Regulation Q as equivalent to redefinitions of the monetary aggregates. We also thank Heather Deaton and Christoph Hinkelmann for research assistance. A Historical Perspective on the Federal Reserve’s Monetary Aggregates: Definition, Construction and Targeting "...the Federal Reserve should use as an intermediate target that monetary total (aggregate), or those to tals, through which it can most reliably affect the behavior o f its ultimate objectives — the price level, employment, output, and the like. Which total or totals best satisfy that requirement depends in turn on (1) how accurately the total can be measured; and (2) how precisely, and at what costs including unwant ed side effects, the Fed can control the total; and (3) how closely and reliably changes in the total are related to the ultimate policy objectives. “In general, though by no means uniformly, the broader the concept, the greater the problems o f measurement and control." Improving the Monetary Aggregates (Report of the Advisory Committee on Monetary Statistics), 1976, p. 7. D TA ON THE MONETARY AGGREGATES A are the fundamental raw material of research in many facets of economics and finance. Money demand modelling, measurement o f money stock announcement effects, tests of the ration ality of preliminary money stock forecasts and financial market efficiency, and comparison of alternative seasonal adjustment procedures are just a few such areas. Monetary aggregates also are used by Federal Reserve System staff in for mulating policy alternatives for the Federal Open Market Committee (FOMC). Perhaps no government data are more important or more widely used in economic and financial research MARCH/APRIL 1994 2 than the monetary aggregates. Often unap preciated by researchers, however, is the extent to which the appropriate use of monetary ag gregates data is intimately connected with changes through time in the data's definitions, construction, revision and publication. A failure to appreciate the interdependence of time, data, definitions and procedures may adversely affect or vitiate research and policy conclusions. This paper discusses the construction, publica tion and evolution of monetary aggregates data since the inception o f the Federal Reserve Sys tem in 1914. In opening their seminal volume on U.S. monetary data, Friedman and Schwartz (1970) set a similar objective: This book attempts to provide a comprehensive survey o f the construction o f estimates o f the quantity o f money in the United States — an ac tivity that dates back almost to the beginnings of the Republic. The survey covers sources, methods o f construction, and the end product, (p. 1) Friedman and Schwartz present a consistent time series of monetary aggregates based on their own data for 1867-1946 and Federal Reserve data through the mid-1960s. This paper and the companion timeline (Kavajecz, 1994) extend Friedman and Schwartz by reviewing the construction and publication of the Federal Reserve’s monetary aggregates from 1960 through 1993. We focus on the years since 1960, the period for which the Federal Reserve Board staff currently publishes official monetary ag gregates. The interested reader will find few (if any) available descriptions o f the Federal Reserve’s monetary aggregates comparable to Friedman and Schwartz’s narrative. The evolution of the monetary aggregates as economic statistics has been influenced by both economic thought and statistical practice.1 Struc tural change in financial markets and the in troduction of new financial instruments require periodic redefinition of the monetary aggregates to accurately reflect the portfolio choices availa ble to households and firms. Never defined nor constructed in the abstract, however, monetary aggregates exist largely as indicators and/or tar gets of monetary policy. Thus, to an unknown but perhaps considerable extent, selection of the 1We do not discuss in this paper the work on aggregation theory and related monetary aggregates such as the Divisia and MQ aggregates. These were consistently labelled by Board staff as experimental and not adopted for policy analysis. The interested reader is referred to Barnett (1980) and Spindt (1985). http://fraser.stlouisfed.org/RESERVE BANK OF ST. LOUIS FEDERAL Federal Reserve Bank of St. Louis definitions of the monetary aggregates has been based on the relative ability o f alternate ag gregates to predict economic activity. Prior to 1980, commercial banks furnished most transac tion deposits and their nontransaction deposits seemed to be the closest substitutes for money. In turn, the Federal Reserve’s monetary ag gregates emphasized both the distinctions between types o f deposits and between commer cial banks and thrift institutions. The narrower M l and M2 aggregates first published in 1971, for example, included only deposits at banks, while thrifts were included in M3. These distinc tions were preserved in 1975 when M3 was re vised and M4 and M5 were introduced. Perceived breakdowns in the historical rela tionship between a monetary aggregate and eco nomic activity, reflected, say, in a putative permanent shift in its velocity, may lead to calls for redefinition o f the aggregate. Such pressures on M l and M2 (as initially defined in 1971) were apparent throughout the 1970s. Reinforced by accelerations in inflation and a shift by some macroeconomists toward increased emphasis on the monetary aggregates, these pressures led in early 1974 to the appointment of the Advisory Committee on Monetary Statistics, chaired by professor George Bach o f Stanford. By 1980, the Depository Institution Deregulation and Monetary Control Act (DIDMCA) permitted a redefined set of monetary aggregates to be con structed from a greatly expanded, much richer and much more costly flow of data than had ever previously been available. The new ag gregates also seemed to have more stable re lationships to economic activity. Published analyses at the time o f the 1980 redefinition cit ed with approval the lack of trend in the veloci ty of the new M2 relative to the old measure, although they stopped short of proposing a less variable long-run velocity as a choice criterion.2 Although such pragmatic redefinition seems clearly to be in the spirit of Friedman and Schwartz3 it may account for at least some part , of the ex post stationarity of the GNP velocity of M2 (as currently defined) identified by Hallman, Porter and Small (1991). The ideal monetary aggregate would be com posed o f assets that are capital-certain (or 2Simpson (1979, 1980). Other descriptions of the construc tion of the Federal Reserve’s monetary aggregates include Broaddus (1975), Duprey (1982), Lawler (1977) and Walter (1989). 3See especially chapter 4. 3 nearly so), highly liquid and closely related to economic activity. Narrow monetary aggregates composed primarily o f medium of exchange seem to satisfy at least the first two criteria acceptably well, while broader aggregates do so somewhat less well. Broader aggregates often include assets that are capital-uncertain or, in other words, as sets whose market values vary with market in terest rates, the pace o f economic activity, or expectations of such variables. Broad monetary aggregates are uniformly defined to include the nominal (face) value of capital-uncertain assets rather than the market value, however. Small time deposits included in the non-Ml compo nent of M2, for example, may be taken to be capital-uncertain when there are penalties for withdrawal before maturity.4 Money market mutual fund (MMMF) shares, also included in the non-Ml component o f M2, appear capitalcertain to their holders even though the market value o f the funds’ assets varies inversely with market interest rates. So long as the MMMFs satisfy a variety of Securities and Exchange Commission rules (including restrictions on the maturity of the funds’ assets) and short-term market interest rates don’t move too rapidly the funds need not pass through changes in the market value of their assets to shareholders. The market values of money market instruments included in very broad aggregates such as M3 and (the seldom used) L vary considerably more, however. Such instruments include negotiable large time deposits included in the non-M2 com ponent o f M3, and most items included in the non-M3 component of L. Monetary aggregates defined to include the nominal rather than mar ket value of these assets necessarily omit some actual portfolio constraints faced by firms and households, who must necessarily substitute among financial assets at market rather than nominal values. Including these assets in mone tary aggregates at market values, however, would cause the measured size of the aggregate to vary with market rates. This might reduce the useful ness of the aggregate as an indicator of the im pact of policy actions. A policy action that reduced reserve availability could reduce not only the quantity of money demanded as mar 4Under Regulation Q, depositories were required to impose early withdrawal penalties. Many institutions have chosen to continue such penalties even in the absence of Regula tion Q. On the demise of Regulation Q, see Gilbert (1986). The liquidity of time deposits has varied through time. Pri or to Reg Q, some time deposits were indistinguishable from modern savings and transaction deposits; see Fried man and Schwartz (1970), p. 76-7. ket interest rates increased, but also the appar ent quantity "supplied” as prices of the included money market instruments fell. The indicator properties o f movements in such capital-uncertain monetary aggregates for economic activity have not been established.5 The statistical issues in building monetary ag gregates also are formidable. If cost were no ob ject, an ideal monetary aggregate would be built from daily observations on all its components at all financial intermediaries. In fact, cost/benefit tradeoffs figure prominently in both data collec tion and the definition of the aggregates. The Congress has mandated that a cost/benefit analy sis be part of each application for renewal of major deposit reports, typically required every three years. Reporting burden is generally to be kept as low as possible while obtaining adequate data for the conduct o f monetary policy. This position has led to deposit reporting strategies based on survey sampling wherein deposit coverage and reporting frequency vary by size o f institution. Most of these issues have largely been omitted from the literature on money demand. As fine a work as Laidler’s (1993) classic text on money demand fails to discuss the definition, construc tion or revision of monetary aggregates, except to acknowledge Friedman and Schwartz’s re search. Nowhere is the reader warned of the potential pitfalls in monetary aggregates data awaiting the unwary. This problem arises largely from the difficulty and high cost to researchers of locating relevant institutional details. This paper attempts to reduce that cost. SOURCES OF MONETARY AGGREGATES DATA Throughout U.S. history, every definition of money has been composed primarily of the lia bilities of private financial institutions, both notes and deposits. During most periods, these financial institutions have been subject to government regulation. In turn, the primary sources of current and historical monetary ag gregates data are government reports filed by these financial institutions. 5The difficulties of interpreting monetary aggregates that in clude capital-uncertain instruments are prominent in proposals to include bond and equity mutual funds in a redefined M2. See, for example, Collins and Edwards (1994) and Orphanides, Reid and Small (1993). MARCH/APRIL 1994 4 The Federal Reserve's first published monetary aggregate appeared in 1943 in Table 9 of Bank ing and Monetary Statistics. The table showed currency demand deposits and time deposits for June call dates from 1892 to 1922 and for June and December call dates from 1923-41. The sum of currency and demand deposits was defined as "the supply o f money” or "means of pay ment" although it was noted that time deposits often were used for current payments "...during the 1920s.” Subsequent data were published in the Federal Reserve Bulletin.6 Later, Copeland and Brill (1948) presented a series based on the lastday-of-the-month consolidated condition state ment of the banking system. In 1949, the Board began monthly publication of this series. The first modern monetary aggregate based on averages of daily data, labelled M l, was con structed by William Abbott and Marie Wahlig of the Federal Reserve Bank o f St. Louis and ap peared in the Federal Reserve Bulletin in 1960 (Abbot, 1960); a revision was published in 1962 (Abbot, 1962). Building monetary aggregates from daily data is important because seasonal patterns within a month may cause data for in dividual days to be unrepresentative of both the month’s average level and the aggregate’s trend growth rate. Abbott and Wahlig’s data, which began in 1947, reflected available deposit reports and were shown at half-monthly and monthly frequencies. Member banks had begun report ing in 1944 averages of daily data at the middle and end of each month. Data for nonmember banks and mutual savings banks (MSBs) were es timated from Federal Deposit Insurance Corpo ration (FDIC) call reports, although the precise interpolation method is not stated. Monetary aggregates data subsequently were published on the Board’s statistical release, known as the J.3 and entitled Demand Deposits, Currency, and Related Items, twice a month from November 1960 through July 1965. The release included averages of daily data at half-monthly and monthly frequencies, seasonally adjusted, and at weekly, half-monthly and monthly fre quencies, not seasonally adjusted.7 The most re cent data included on the release predated the publication date by two weeks. 6For details, see the introductory notes to section 1 in Bank ing and Monetary Statistics and the notes to chapters 1-4 in Banking and Monetary Statistics 1941-1970. 7Member banks began reporting daily data each week in December 1959. For years after 1959, the weekly data were prorated to obtain monthly and half-monthly frequencies. http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis The J.3 was succeeded by the current release, known as the H.6 and entitled Money Stock, Liquid Assets, and Debt Measures, on July 30, 1965. It shows averages of daily figures at week ly and monthly frequencies. A revised monetary aggregates series based on weekly averages of daily data beginning in 1959 was later presented by Fry, Beck and Weaver (1970).8 The current definitions o f the monetary aggregates were largely established in 1980; see Kavajecz (1994) and Simpson (1979, 1980). At the time of the redefinition, monetary aggregates based on the new definitions were constructed back to 1959. Details o f their construction are discussed in the appendix. For researchers, monetary data extracted from individual issues of the J.3 and H.6 releases pro vide contemporaneous estimates of the mone tary aggregates based on a well-defined infor mation set: the data available to Board staff as of the publication date. These statistical releases allow a researcher interested in announcement effects or the policy formation process o f the FOMC to observe Federal Reserve Board staff es timates o f the level o f the money stock at each point in time, or permit a researcher interested in market efficiency or the “rationality” o f initial money stock estimates to study the timing and extent o f revisions to initially published data. The statistical releases are not very useful for longer-run studies, however, because the infor mation set underlying the release changes each week as Board staff receives both new data and revisions to previously reported data. Further, the definitions of the monetary aggregates have changed through time. While the Federal Reserve Board has published a number of historical volumes, each with unique features making it a valuable source of data, use of these data also is complicated by varying defi nitions and observational frequencies. Ideal his torical data would be computed at similar frequen cies under consistent definitions. The two most comprehensive volumes, Banking and Monetary Statistics and Banking and Monetary Statistics 1941-1970, were published by the Federal Reserve in November 1943 and September 1976, 8Some independent researchers have attempted to build monetary aggregates data for earlier periods using current definitions. For a careful discussion of the issues, see Rasche (1987, 1990). 5 respectively.9 Observational frequency differs across data series, with various data at monthly, weekly or daily frequencies. There are also im portant conceptual distinctions through time in the data, such as the difference between mem ber and nonmember banks and the difference between thrifts and commercial banks. When using data from other sources in conjunction with the Banking and Monetary Statistics volumes, researchers should appreciate that data published subsequently are not strictly compara ble, since more recent publications incorporate further revisions to the data. A closely related publication, and the yearly counterpart to the Banking and Monetary Statistics volumes, is the Annual Statistical Digest. The Digest is released at the end o f each year and contains data for the previous year. The Board’s Annual Report also contains information about the monetary aggregates, but the information tends to be more descriptive than numerical. These publications provide a long-run, consis tent perspective o f the monetary aggregates over their respective published date ranges, since within each issue of each publication the observations are based on a single, consistent in formation set. They perhaps are less appropri ate, however, for lines of research where the hypotheses depend on the information set used in constructing the money stock estimate, since the date the estimate was formulated is not explicitly given. Similar concerns suggest that data sets con structed from various issues of the Federal Reserve Bulletin may not be suitable for a variety of research. Board staff have published compo nents of the monetary aggregates, such as de mand deposits and currency, in the Bulletin since its inception in May 1915. In February 1944, the staff first showed demand deposits and currency in the same table, foreshadowing the later M l monetary aggregate. While the Bulletin’s current T&ble 1.10 (first published in its present form in January 1977) descends from the 1944 table, the data published in this table through the years are not a consistent time ser ies due to definition changes, reporting changes, 9The 1943 edition of Banking and Monetary Statistics was reprinted in August 1976. See also the Board’s corrected 1959 reprint of All-Bank Statistics. ,0The title of this publication has changed somewhat through time. It currently is produced by the Money and Reserves Projections Section of the Division of Monetary Affairs. Pri or to 1988, it was produced by the Banking Section of the Division of Research and Statistics. Prior to 1993, the print ed publication was offered to the public as a supplement annual benchmark revisions, and reestimation of seasonal adjustment factors. At the same time, the Bulletin is an excellent resource for tracking the various changes that have occurred in the definitions and construction of the monetary ag gregates through time. Due to its somewhat longer time span, data extracted from various is sues of the Bulletin illustrate how the monetary aggregates have evolved; occasional articles have presented detailed information on changes in the monetary aggregates. Unfortunately, like many other Federal Reserve historical publica tions, the Bulletin does not specify the date at which the estimates were made, that is, the time-indexed information set on which they were based. In general, data in the Bulletin pre cede by two months the Bulletin’s publication date, but at times it has been longer. Since monetary aggregates data appear with differing lags in various System publications (for example, 10 days on the H.6), data from different sources may be based on quite different information sets even when the dates that they first appear in print are close together. This suggests that, in general, a database built from one Federal Reserve source or publication should not be up dated from another. Finally, a publication that presents compre hensive, consistent time series is Money Stock Revisions. 0 This publication is offered to the 1 public early in each year as a supplement to the issue of the H.6 release that incorporates the Board staff’s annual benchmark revisions, in cluding reestimated seasonal adjustment factors. The publication presents a comprehensive set of monetary aggregates data, beginning in 1959 for monthly data and in about 1975 for weekly data.1 Unlike other Board staff publications, the 1 information set and definitions used in con structing the data are well-defined, making the data ideal for longer-run studies. Note, however, that since each year’s publication uses that year's current definitions — and the definitions of the monetary aggregates and their compo nents have changed through time — the data may differ significantly from previously pub lished data. to the issue of the H.6 release that contained the newly benchmarked monetary aggregates data; data in machine readable form were sold by the National Technical Informa tion Service of Springfield, Virginia. In 1993, the publica tion and associated data were first offered for sale by Publications Services at the Board of Governors. "Subject to the availability of the particular series. See Table 2 for the availability of specific series. MARCH/APRIL 1994 6 DATA COLLECTION The data collection process is the foundation of the construction of monetary aggregates data. The collection o f data useful for the monetary aggregates has changed (and improved) dramati cally during the last eight decades. We present here a brief outline of the principal data inflows to the Federal Reserve during a small number of distinct periods over which data collection and publication practices differed significantly. 1915-43 The data collected during this period have been extensively documented by Friedman and Schwartz (1970), chapters 12-15. Beginning in 1923, data for all member banks are available. From April 1923-December 1928, the Federal Reserve collected and published deposits as o f a single day each month; from January 1929March 1944, monthly averages of daily data; af ter March 1944, averages o f daily data were col lected twice a month. Data also continued to be reported each week on Wednesday by a sample of several hundred weekly reporting banks that held a majority of bank deposits. Data for non member banks and for MSBs were available on call reports. 1944-80 Averages of daily member bank deposit data were collected twice a month through Decem ber 1, 1959, when weekly averages began to be collected. Regular publication beginning in November 1960 of monthly money stock figures on the J.3 release necessitated estimates o f the monetary liabilities o f nonmember banks. Nonmember bank data continued to be collected on call reports, typically two per year until 1960, when thereafter four per year were required. 1980-Present Perhaps the least appreciated aspect of the Monetary Control Act of 1980 was a significant 12ln particular, thrift institutions and nonmember banks be gan reporting deposits weekly to the Federal Reserve. 13A zero reserve requirement ratio applies to the reserve ex emption amount of deposits. The reserve exemption amount is not to be confused with the low reserve tranche. The tranche allows a lower 3 percent reserve requirement ratio to be applied to some portion of deposits, while a higher ratio (currently 10 percent) applies to the balance. Both the reserve exemption amount and the low reserve tranche are indexed. For 1993, the reserve exemption and http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis improvement in the quantity and quality of data flowing to the Federal Reserve. A water shed in data collection, the act empowered the Federal Reserve System to impose reporting requirements on all depository institutions with reservable liabilities above a prescribed minimal amount. The act significantly eased estimation of the money stock, as deposit re porting by financial institutions became nearly universal and was no longer a function of mem bership status or charter type.1 'IWo years later, 2 in the Garn-St. Germain Act, Congress mandated that the Federal Reserve establish guidelines to ease reporting burden borne by financial in stitutions while maintaining adequate coverage o f the outstanding monetary liabilities o f the banking system. In response, a system o f report ing categories was established wherein the reporting burden — measured by frequency of reporting and number of items reported — depends upon both total deposits and reservable liabilities. Under this system, the Federal Reserve Board staff each year establishes a cutoff level o f total deposits and an exemption level o f reservable liabilities. Increases in both levels are indexed to the year-over-year increase in aggregate deposits at all depository institutions as calculat ed from second quarter (June 30th) call reports.1 Tible 1 summarizes the System’s 3 reporting categories and the type/frequency of report submitted by financial institutions in each category for 1992, 1993 and 1994.1 The 4 deposit cutoff and reserve exemption levels were established at $25.0 and $2.4 million, respective ly, beginning January 1985. These have subse quently been indexed each year, based on 80 percent of the growth in aggregate deposits, except in 1988. In that year, Board staff research suggested that little accuracy would be sacrificed, and a significant reporting burden reduced for smaller institutions, by increasing the deposit cutoff more rapidly. The deposit cutoff, which had automatically increased in January to $30 million from the previous year’s $28.6 million, low reserve tranche amounts are $3.8 and $46.8 million, respectively. For 1994, the amounts are $4.0 and $51.9 mil lion, respectively. ' “Values for each year are typically published in the respec tive January issues of Federal Reserve Bulletin. Values for 1992, 1993 and 1994, for example, appear on pp. 36-7, 18 and 23-4 of the January 1992, 1993 and 1994 issues, respectively. 7 Table 1 Depository Institution Reporting Categories 1992-94 by Deposit Cutoff and Reserve Exemption Amount Reserve Exemption Amount reservable liabilities Deposit Cutoff total deposits if more than $3.6 ($3.8) [$4.0] if less than $3.6 ($3.8) [$4.0] if more than $44.8 ($44.8) [$44.8] the institution must file the FR2900 report weekly the institution must file the FR2910Q report quarterly if less than $44.8 ($44.8) [$44.8] effective as of January 1992 (1993) [1994] the institution must file the FR2900 report quarterly the institution might be exempt from reporting Note: All figures are in millions of dollars. was raised in September to $40.0 million. Sever al thousand smaller banks were exempted from weekly reporting by this change. ed), or, including nondeposit liabilities, about 80 percent o f the aggregate liabilities o f financial institutions included in the monetary aggregates. Institutions that file the FR2900 at a weekly frequency (Table 1, the upper left-hand box) report daily levels for about a dozen deposit and nondeposit liabilities. Institutions falling in the other boxes have a sharply reduced reporting burden. Institutions that file the FR2900 at a quarterly frequency (the lower left-hand box) re port the same items but only for a single week each quarter (the week that contains the third Thursday in the last month of the quarter). Insti tutions that file the FR2910Q (upper right-hand box) report weekly average data on fewer items for one week each quarter. Institutions in the lower right-hand box of 'Fable 1 are exempt from filing reports with the Federal Reserve if and only if Federal Reserve staff are able to accurately obtain required data from other sources, such as call reports.1 For institutions other than weekly 5 reporters (all categories except those in the up per left-hand box), Federal Reserve Board staff must estimate their deposits during the periods between reports. In 1992, daily data were re ceived each week from approximately 9,100 financial institutions, about 30 percent of all depositories. These data comprised about 90 per cent o f the aggregate deposits included in the monetary aggregates (the balance being estimat Construction of weekly values of broad mone tary aggregates such as M2 and M3 also relies on a variety o f weekly reports o f data for non deposit liabilities such as repurchase agreements (RPs), Eurodollar deposits, and reports from non bank financial institutions such as MMMFs. The numerous sources and reports used by Board staff in the construction of the monetary ag gregates are shown in Table 2. In general, broader aggregates such as M2 and M3 are less precisely measured than M l because a larger proportion o f the data included in the aggregate is either not reported directly to the Federal Reserve, and/or is reported less frequently than the data included in M l. In addition, a larger number of various nonmoney stock items are netted out of the broader aggregates. In the non-Ml components of M2 and M3, MMMF shares have been among the more com plex items. A dynamic industry characterized by rapid growth, new funds have frequently ap peared and old ones vanished. In addition, funds may merge, change names or change in vestment objective by, say, lengthening the maturity of their assets to become a short-term bond fund. All these events complicate accurate 15lf not, the institution is required to file an annual report. MARCH/APRIL 1994 FEDERAL RESERVE Table 2 Information about the Definition, Availability and Source Data for the Monetary Aggregates________________ BANK O ST. LOUIS F This table provides information on the construction of the monetary aggregates M1, M2, M3 and L as of October 1993. Readers are cautioned that some definitions and data sources may differ in earlier periods. Each aggregate reflects the amounts of the designated assets held by the nonbank public, which includes households, business es and government entities other than the U.S. Treasury. Assets issued in the U.S. are included whether they are held by foreign or domestic residents. Certain dollardenominated assets issued abroad and held by U.S. residents also are included. The aggregates are constructed by consolidation rather than aggregation, such that the liabilities of one money stock issuer that are held by another issuer within the same aggregate cancel each other. For example, the amount of large time deposits held by money market mutual funds is subtracted from gross large time deposits in building M3, because these deposits are both a liability of one money stock issuer (banks) and an asset of another (money market mutual funds). Monetary aggregates published by the staff of the Board of Governors as of October 1993 were: M1 M2 M3 L = = = = currency + checkable deposits; M1 + certain nontransaction deposits and other liquid assets; M2 + certain assets that are either less liquid and/or issued in large denominations; and M3 + certain money market instruments. Federal Reserve System reports are referred to below by the prefix FR and reports of the interagency Federal Financial Institutions Examination Council by the prefix FFIEC. Call reports are administered by the FFIEC, a joint agency including the Federal Reserve, the Federal Deposit Insurance Corporation (FDIC), the Treasury Department and the National Credit Union Administration (NCUA). Complete report titles and reporting frequency are shown only the first time a report is cited; references thereafter are abbreviated. NSA published data begin Money Stock Component M1 = ( + ) Money stock currency = Currency held by the nonbank public (in other words, held out side the U.S. Treasury, Federal Reserve Banks and the vaults of depository institutions). monthly weekly 1/59 Definition 1/6/75 1/59 Source of Information 1/6/75 Federal Reserve Board staff have judged that adequate data are not available before these dates to construct monetary aggregates based on current definitions. ( + ) Currency in circulation Currency held outside the U.S. Treasury and Federal Reserve Banks. Federal Reserve Statement of Condition (internal Fed balance sheet) (FR34), daily; Treasury and Mint Reports on currency and coin in circulation. ( - ) Vault cash Cash held by depository institu tions (including cash in automatic teller machines). Report of Transaction Accounts, Other Deposits and Vault Cash (FR2900), from weekly and quarterly reporters; Quarterly Report of Selected Deposits, Vault Cash and Reservable Liabilities (FR2910Q); Annual Report of Total Deposits and Reservable Liabilities (FR2910A); Consolidated Reports of Condition and Income (call reports) (FFIEC 031, 032, 033, 034), quarterly, last business day of the quarter. The FR2900 is the core report for the monetary ag gregates. More than 9,000 financial institutions file the FR2900 report weekly following their Monday close of business, each report containing daily deposit data for the preceeding week. Some smaller institutions file the FR2900 report only for one week each quarter. See the text for discussion. co ■■■■ NSA published data begin Money Stock Component monthly weekly ( + ) Travelers checks Outstanding amount of U.S. dollar-denominated travelers checks issued by nonbanks (checks issued by banks are in cluded in demand deposits). 1/59 1/6/75 ( + ) Demand deposits adj usted = Demand deposits at all depository institutions in the U.S. other than those due to other depositories (including money market mutual funds [MMMFs]), the U.S. government, and foreign banks and official institutions, less cash items in the process of collection (CIPC) and Federal Reserve float. 1/59 1/6/75 ( + ) Gross demand deposits Definition Source of Information Monthly Report of Travelers Checks Outstanding (FR2054), last business day of the month; weekly data are interpolated from seasonally adjusted monthly data. FR2900; FR2910Q/A; call reports Deposit liabilities of banks payable on demand; time deposits with original maturity of less than seven days; travelers checks and money orders that are the primary obligation of the issuing deposi tory institution. ( - ) Demand deposits due to depository institutions, for eign banks and official insti tutions, and the U.S. Treasury Weekly Report of Assets and Liabilities for Large Banks (FR2416), includes about 160 large banks, weekly, close of business Wed nesday; call reports for other depositories, quarterly, last business day of quarter. ( + ) Other money orders Money orders and official checks issued by nonbank subsidiaries or bank holding companies. Weekly Report of Money Orders and Similar Payments Instruments issued by Nonbank Subsidiaries of Bank Holding Companies (FR2053), close of business Monday. ( - ) Cash items in process of collection Third-party payment instruments (checks) redeemable in immedi ately available funds if presented today. Same as gross demand deposits; all checks being collected are deducted from demand deposits regardless of the type of account wherein the deposit was made. FR34 ( - ) Float on the Federal Reserve ( + ) Other checkable deposits MARCH/APRIL 1994 NOW and automatic transfer service (ATS) accounts at com mercial banks, U.S. branches and agencies of foreign banks, and Edge Act corporations; NOW and ATS accounts at thrifts; credit union share draft balances; and demand deposits at thrifts. 1/63 1/6/75 FR2900; FR2910Q/A; call reports, quarterly FEDERAL NSA published data begin Money Stock Component RESERVE ( + ) Savings deposits, net = BANK O ST. LOUIS F ( + ) savings and MMDA deposits at banks and thrifts Passbook and statement savings deposits plus money market de posit accounts (MMDA) other than those due to general purpose and broker/dealer money market funds, foreign banks and official institutions and the U.S. government. MMDAs are a special type of savings ac count that permits a small num ber of third-party payments per month. weekly 1/5/81 Adequate weekly thrift data are not available before 1981; see Appen dix 1 for discussion of monthly thrift data for 1959-80. 1/59 12/82* (*MMDAs) 11/3/80 12/20/82* MMDAs were first authorized in December 1982; separate savings and MMDA data were collected until September 1991. Thereafter, only a single combined series has been collected. FR2900; FR2910Q/A; call reports Deposit or account in which the depositor is not currently, but may be at any time, required by the financial institution to give written notice of intent not less than seven days prior to with drawal. FR2416; call reports ( - ) savings and MMDA deposits due to foreign banks, foreign official institutions and the U.S. Treasury ( + ) Adjusted small time deposits = Deposits, including retail repur chase agreements (RPs), issued in amounts of less than $100,000 with original maturities of seven days or more, less all IRA/Keogh retirement account balances at banks and thrifts. 1/59 11/3/80 FR2900; FR2910Q/A; call reports ( + ) gross small time deposits ( + ) retail RPs at commercial banks and mutual savings banks (MSBs) Source of Information monthly 1/59 Definition Non-M1 component of M2 = Retail RPs are issued in small denominations most often to households and small businesses. Monthly Survey of Selected Deposits (FR2042), last Wednesday of the month. ( + ) retail RPs at savings and loan associations Office of Thrift Supervision, quarterly thrift balance sheet ( - ) IRA/Keogh balances at commercial banks and MSBs FR2042 ( - ) IRA/Keogh balances at savings and loan associations Office of Thrift Supervision, quarterly thrift balance sheet NSA published data begin Money Stock Component_________ Definition__________________________ monthly weekly______ Source of Information Non-M1 component of M2 = (continued) ( + ) Share balances in general purpose and broker/dealer MMMFs MMMFs are certain types of investment companies that agree to abide by the SEC’s Rule 2a-7 and a variety of other regulations regarding the types and maturi ties of allowable assets. Shares in these funds may be held by households, businesses and vari ous institutions. 1/74 2/4/80 ( + ) Overnight RPs, net = One-day and continuing-contract RPs issued by all depository in stitutions to other than depository institutions, MMMFs and foreign official institutions. 11/69 1/6/75 ( + ) gross overnight RPs RPs as of close business, one day each week Report of Selected Borrowings (FR2415), for commercial banks, weekly, close-of-business Monday; Weekly Report of Repurchase Agreements on U.S. Government and Federal Agency Securities with Specified Holders (FR2415t), for thrifts, close of business Monday ( - ) overnight RPs held by MMMFs ( + ) Overnight Eurodollars, net = The Investment Company Institute (ICI) voluntarily collects informa tion for the Federal Reserve. Weekly and monthly reports cover both the funds’ liabilities (shares) and assets. The amounts of individual assets held by MMMFs are important because most assets— including RPs, Eurodollars, large time deposits and Treasury bills —are netted from the monetary aggregates during the consolidation of M2, M3 or L. Data are labeled by Federal Reserve staff as the Weekly (Monthly) Report of Assets of Money Market Mutual Funds [FR2051a (FR2051 b)]; Weekly Report of Assets for Selected Money Market Mutual Funds (FR2051 c); or the Weekly Report of Overnight Eurodollars for Selected Money Market Mutual Funds (FR2051d). The ICI data are as of close of business on Wednesday. The Wed nesday level is included in the aggregate for the week ending the following Monday. For example, M2 and M3 for the week of January 10, 1994, contained data on MMMF shares as of Wednesday, January 5. FR2051a, c MARCH/APRIL 1994 Eurodollar deposits with original maturity of one day issued by foreign branches of U.S. banks worldwide to U.S. nonbanks (U.S. addresses other than depository institutions and MMMFs) 2/77 12/31/79 ( + ) gross overnight Eurodollars Report of Selected Deposits in Foreign Branches held by U.S. Ad dresses (FR2050), weekly reporting of daily data, close of business Monday; Monthly (Quarterly) Report on Foreign Branch Assets and Liabilities [FR2502, (FR2502s)], last business day of the period ( - ) overnight Eurodollars held by MMMFs FR2051a, c FEDERAL NSA published data begin Money Stock Component RESERVE ( + ) Large time deposits, net = BANK O ST. LOUIS F Deposits issued by banks and thrifts in amounts of $100,000 or more with initial maturities of seven days or more, other than those held by MMMFs, other depository institutions, and for eign banks and official insti tutions monthly weekly 1/59 Definition Non-M2 component of M3 = 1/5/81 1/59 Source of Information 11/3/80 ( + ) gross large time deposits FR2900; FR2910Q/A; call reports ( - ) large time deposits due to foreign banks and official insti tutions, and the U.S. Treasury FR2416; call reports, quarterly ( - ) large time deposits held by MMMFs FR2051a, c ( - ) mortgage-backed bonds at savings and loan associations ro 10/69 ( + ) Term RPS, net = ( + ) gross term RPs Office of Thrift Supervision, Statement of Condition (call report), quarterly Mortgage-backed bonds are reported as a reservable liability on the FR2900. They are not deposits, however, and, hence, are subtracted from the monetary aggregates. 1/6/75 FR2415 RPs issued by all depositories with original maturities greater than one day, other than continu ing contract and retail RPs and RPs issued to other depositories and foreign banks and official in stitutions. FR2051a, c ( - ) term RPs held by MMMFs 1/59 ( + ) Term Eurodollars, net = ( + ) gross term Eurodollars { - ) term Eurodollars held by MMMFs Eurodollar deposits due to U.S. nonbank addresses with maturity longer than one day at all foreign branches of U.S. banks and at offices of non-U.S. banks in the U.K. and Canada 12/31/79 FR2050; FR2502; data furnished by the Bank of England and Bank of Canada. FR2051a, c NSA published data begin Money Stock Component Definition monthly weekly 4/74 2/4/80 Source of Information Non-M2 component of M3 = (continued) ( + ) Shares in institution-only (l-O) MMMFs, net = ( + ) shares in l-O MMMFs, gross MMMFs that under SEC guide lines require large minimum in vestments (typically $50,000 + ) and sell shares only to sophisti cated investors and institutions, thereby gaining exemption from certain SEC accounting rules. These shares may be held by households, businesses or insti tutions. FR2051a, c ( - ) overnight RPs and Euro dollars held by l-O MMMFs Note that term RPs and Eurodol lars held by MMMFs were netted above. FR2415 for banks; FR2415t for thrifts Non-M3 component of L = ( + ) Bankers acceptances, net = 1/59 Bankers acceptances held by the nonbank public other than ac cepting banks, Federal Reserve Banks, foreign official institutions, Federal Home Loan Banks and MMMFs. NA 1/59 NA CO (+ ) gross bankers acceptances Monthly Survey of Eligible Bankers Acceptances (FR2006), month ly, last day of the month; call reports, quarterly ( - ) acceptances held by Federal Reserve Banks FR34 ( - ) acceptances held by MMMFs ( + ) Commercial paper, net = FR2051a, c Commercial paper held by the nonbank public other than MMMFs. 1/59 NA MARCH/APRIL ( + ) gross commercial paper Report of Commercial Paper Outstanding Placed by Brokers and Dealers (FR2957a), weekly, Wednesday; Report of Commercial Paper Outstanding Placed Directly by Issuers (FR2957b), weekly, Wednesday and last day of the month ( - ) commercial paper held by MMMFs FR2051a, c 1994 FEDERAL RESERVE NSA published data begin Money Stock Component ( + ) Short-term U.S. Treasury securities, net = Definition BANK O ST. LOUIS F Treasury bills and coupons with remaining maturities of less than 12 months held by the nonbank public other than depositories, Federal Reserve Banks, MMMFs, and foreign banks and official institutions. monthly weekly 1/59 NA Source of Information ( + ) gross short-term Treasuries Monthly Statement of Public Debt, U.S. Treasury Department ( - ) Federal Reserve Bank holdings of short-term Treasuries FR34 FR2051a, c ( - ) MMMF holdings of short term Treasuries ( + ) U.S. savings bonds U.S. government savings bonds held by the nonbank public. 1/59 N.A. SOURCE: Compiled by the authors from published and unpublished Federal Reserve documents. Monthly Statement of Public Debt, U.S. Treasury Department 15 measurement of the aggregate amount of MMMF shares held by the nonbank public. Retirement accounts (IRA/Keogh) at banks, thrifts and MMMFs also have sometimes been nettlesome. These deposits, netted from the monetary aggregates, are not collected in the same manner as other deposit data included in the aggregates. As shown in Tkble 2, retirement balances at banks are collected in the FR2042 report. This report surveys fewer banks less fre quently than the FR2900 report that provides most deposit data. Retirement balances at MMMFs are collected by the Investment Compa ny Institute from member mutual funds and, like data for commercial banks and thrifts, lags somewhat behind the reporting of deposits and other liabilities included in the aggregates. Measurement problems also arise regarding Eurodollars and RPs. High-quality timely data are available on the overnight Eurodollar com ponent of M2 because these deposits are largely held at Caribbean branches of U.S. banks.1 6 Tferm Eurodollars held in foreign branches of U.S. banks are reported on approximately the same basis. Term Eurodollars, however, also are held extensively at non-U.S. banks in England and Canada, not subject to Federal Reserve reporting. The Bank of England and the Bank of Canada collect quarterly data for U.S.-dollar denominated deposits due to U.S. nonbank ad dresses. Although aggregate totals are given to Federal Reserve staff, data for individual banks are confidential and, hence, can neither be checked nor edited by Federal Reserve staff.1 7 For RPs, the problem is more a conceptual is sue than a matter of data reporting. Overnight RPs are included in the non-Ml component of M2 because, at least in part, they are an attrac tive alternative to holding transaction balances. RPs with maturity o f more than one day also, of course, may serve the same purpose. RPs with a maturity longer than one day however, are reported as term RPs and included in the nonM2 component of M3. An investor who accepts a two-day RP contract rather than a sequence of two, one-day contracts may reduce the size of 16ln fact, these deposits are recorded in New York while be ing legally booked through “ nameplate” branches in the Caribbean (so-called because the office largely consists of a brass nameplate). M2 without any economic significance. It seems likely that much of the predictable part of such switches, say, due to holiday weekends, is cap tured in the seasonal adjustment factors. The balance remains as statistical noise. Overall, weekly first-published values of M2 and M3 shown on the current H.6 release are based about 80 percent on data that are report ed weekly, with the balance estimated from lesser frequency reports.1 8 MAJOR OPERATIONS RY ROARD STAFF THAT AFFECT THE MONETARY AGGREGATES In addition to the principal sources o f data, well-informed researchers should be aware of the more important revision practices and schedules used by Federal Reserve Board staff that affect the continuity of the data. Bench marks, seasonal factor reestimation and defini tion changes may have significant impacts on the monetary aggregates and, correspondingly, on research employing that data. B en ch m a rk R evision s All monetary aggregates data are subject to a “benchmark” revision annually. In its most general form, a benchmark of the monetary ag gregates by Board staff would be (ideally) a measurement of the universe o f money stock is suers and their holdings of monetary liabilities. A benchmark serves three main purposes. First, it allows Board staff to incorporate deposit data on institutions that are exempt from reporting directly to the Federal Reserve. These data are obtained either from bank and thrift call reports or from other annual reports filed by the institutions. Second, it allows the incorpora tion o f corrected/revised data submitted by depository institutions throughout the year. Third, it allows staff to update estimates of some nondeposit components of the aggregates. Depository institutions generally submit re vised deposit data throughout the year. Such 18Detailed estimates of such coverage ratios are prepared about every three years and furnished to the Office of Management and Budget as part of the reauthorization process for the report. See Walton and others (1991). 17ln addition, few statistics are available for coverage ratios, error rates, and so on. MARCH/APRIL 1994 16 data from weekly reporting institutions are in corporated into the monetary aggregates pub lished on the H.6 release only during the first three weeks following the week in which the report was due, that is, the four most recent weeks shown on the H.6 release. Deposit data submitted after that time are held in abeyance and incorporated at the annual benchmark, along with data received from institutions that report only once per year. (Deposit data received from quarterly reporting institutions are incor porated when received during the year, as are nondeposit data received from many sources. See Table 2.) This three-step process begins with aggregation of all deposit data reported by financial institutions during the past six or seven years. Next, data are matched to call reports for all depository financial institutions to identify missing institutions (if any) and ob tain deposit levels at the call dates for those in stitutions exempt from filing deposit reports with the Federal Reserve. Finally, miscellaneous data collected during the year regarding items not covered by deposit reports are incorporated. Benchmarks constitute a clear break-in-series for monetary aggregates data, changing signifi cantly not only past data but altering the base upon which new estimates will be published during the coming year. Since 1964, a bench mark of the monetary aggregates has been done at least annually In recent years, Board staff have published the benchmark data prior to the February Humphrey-Hawkins testimony of the Federal Reserve Chairman before Congress. From 1974 through 1980, however, benchmark revisions of the monetary aggregates were con ducted approximately every quarter. The in creased frequency of benchmarks addressed a concern, raised by the Bach Commission, that the methods used at the time to estimate nonmember bank deposits could introduce a bias into the monetary aggregates. It was felt that more timely benchmarks would serve to keep the Federal Reserve’s estimates more closely aligned with the true, unobserved figures. This was not a new concern, however, and in fact all benchmarks prior to the Monetary Control Act had focused heavily on nonmember bank deposits, since these institutions were not re quired to report to the Federal Reserve.1 The 9 power to enforce near-universal reporting that 19The quarterly deposit data reported on the call reports by nonmember banks also were not without problems. The definitions of “deposits” differ somewhat between the Fed’s Regulation D and the call report instructions, making the FEDERAL http://fraser.stlouisfed.org/RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis was endowed on the Federal Reserve by the Monetary Control Act obviated the need for fre quent benchmarks after 1980. Tbday, bench marks focus on special items not covered on deposit reports. The effects o f these revisions on quarterly growth rates o f the monetary aggregates are shown in the first page of Table 3. The columns of the table correspond to the annual bench marks published in early 1986-93. Each entry in the table is the change in the annualized growth rate o f the corresponding monetary aggregate during that quarter due to revisions of the un derlying source data. The largest revisions due to any benchmark occur in the most recently completed year, shown as the shaded areas in the table. Revisions for prior years, not shaded, are smaller. While not following a consistent pattern, the data suggest that any particular quarter may be revised significantly especially for the broader aggregates. In part, the latter are related to the higher percentage of non deposit components in those aggregates. Seasonal Adjustm ent Seasonal adjustment o f the monetary ag gregates has long been an important area of research. The FOMC formulates its monetary policy in terms of seasonally adjusted data, and both the public and policymakers often take re cent movements in adjusted data as indicating the underlying trend growth rate of the mone tary aggregates. Seasonal adjustment methods attempt to separate recurring calendar-related patterns in data (due to, say, calendar dating, payroll sched ules, tax filing deadlines, and so on) from ran dom shocks and the underlying trend. In general terms, the data generating process for the monetary aggregates is assumed to be well represented as the product of three compo nents: a time-varying trend, a time-varying seasonal and an irregular. Each year, Board staff publish revised seasonal factors for most historical periods and projected seasonal factors for the upcoming year. With few exceptions, these seasonal factors are based on, and published simultaneously with, the andata not fully comparable. For earlier analyses of the effect of benchmark revisions, see Lang (1978) and Simpson and Williams (1981). Table 3 Page 1: Revisions to Previously Published Quarterly Growth Rates of the Monetary Aggregates (s.a.) Due to Benchmark Data Revisions Year of annual benchmark (usually published in February; see Kavajecz, 1994) 1986 1987 1988 1989 1990 Periods M1 M2 M3 M1 M2 1984 Q4 0.4 0.6 -0 .2 0.4 0.5 -0 .6 -1 .2 1985 Q1 Q2 Q3 Q4 0.1 - 0 . 4 -1 .0 0.1 0.3 -0 .3 0.1 0.0 0.0 0.6 -0 .1 -0 .6 -0 .2 -0 .3 0.2 0.0 0.1 0.2 0.5 0.8 -0 .4 -0 .1 -0 .1 0.2 0.0 0.2 0.2 -0 .1 0.0 0.0 0.0 0.0 0.1 0.1 -0 .1 -0 .1 0.1 0.7 0.5 0.6 0.4 0.5 -0 .1 -0 .1 - 0 .5 0.0 - 0 .5 0.1 0.1 -0 .1 0.0 0.0 -0 .1 0.0 0.0 0.0 0.0 -0 .1 - 0 .2 - 0 .4 0.4 0.1 0.3 0.0 0.0 0.0 0.2 0.0 0.1 0.3 - 0 .2 0.1 0.1 0.0 0.0 0.0 0.0 1991 0.0 0.0 0.0 0.1 1986 Q1 Q2 Q3 Q4 1987 Q1 Q2 Q3 Q4 1988 Q1 Q2 Q3 Q4 1989 Q1 Q2 Q3 Q4 1990 Q1 Q2 Q3 Q4 M3 0.5 0.1 0.0 0.0 -0 .1 - 0 .7 - 0 .4 -0 .1 - 0 .2 -0 .1 -0 .1 -0 .1 0.1 M1 M2 M3 M1 M2 0.1 0.0 0.1 0.0 0.3 0.3 M1 M2 -0 .1 M3 0.2 M3 0.0 0.1 0.1 0.2 - 0 .2 0.0 0.2 - 0 .3 - 0 .3 0.0 -0 .1 -0 .1 MARCH/APRIL 1994 0.0 0.1 0.1 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.1 0.1 0.0 0.0 -0 .1 0.0 -0 .1 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -0 .1 0.0 0.0 0.0 0.0 0.0 0.0 0.1 0.0 0.0 0.1 0.0 0.1 0.3 0.1 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.1 0.1 -0 .1 -0 .1 0.0 0.0 - 0 .3 0.0 0.3 0.4 0.0 0.0 0.0 0.0 0.0 0.0 0.1 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.1 0.0 0.0 0.0 0.0 0.0 0.3 -0 .1 0.1 0.1 0.0 0.1 0.1 0.0 0.0 -0 .1 0.1 -0 .1 0.2 0.1 0.1 0.1 0.4 0.0 0.0 0.0 0.0 0.0 0.1 0.1 0.1 0.0 -0 .1 0.0 0.0 0.0 -0 .1 0.0 -0 .1 0.3 0.3 0.0 0.0 0.0 0.4 0.2 0.1 0.1 0.4 0.2 0.0 -0 .1 -0 .1 0.1 0.2 0.0 0.1 0.3 0.1 0.0 -0 .1 0.0 -0 .1 -0 .1 0.1 0.0 0.0 -0 .1 0.2 0.1 -0 .1 0.1 -0 .1 0.0 0.2 0.5 0.2 0.7 - 0 .2 0.1 0.0 0.3 0.3 M3 0.0 -0 .1 0.0 -0 .1 0.0 0.0 0.0 0.0 -0 .1 -0 .2 -0 .2 -0 .1 1992 Q1 Q2 Q3 Q4 Note: These revisions do not include effects due to revisions in seasonal adjustment factors and/or changes in definitions. M2 0.0 0.0 -0 .1 M1 1993 M2 1991 Q1 Q2 Q3 Q4 1992 M1 0.0 0.0 0.1 -0 .1 -0 .1 -0 .1 0.0 0.0 0.2 M2 M1 M3 M3 0.0 0.0 0.0 0.0 -0 .1 -0 .1 0.1 0.1 0.0 0.1 0.0 0.1 0.0 0.2 0.1 -0 .1 -0 .2 -0 .1 -0 .1 -0 .2 -0 .1 -0 .1 0.2 - 0 .4 0.0 0.2 0.1 0.7 -0 .1 -0 .2 -0 .3 0.0 -0 .1 - 0 .2 FEDERAL RESERVE Table 3 (cont.) Page 2: Revisions to Previously Published Quarterly Growth Rates of the Monetary Aggregates (s.a.) Due to Revisions to Seasonal Adjustment Factors Year of annual seasonal review (usually published in February, along with benchmark data revisions) 1987 1986 0.9 0.0 - 0 .3 -0 .6 0.0 0.4 0.2 0.7 0.6 - 0 . 6 - 0 . 7 - 0 .6 1.1 0.1 - 0 . 4 1986 Q1 Q2 Q3 Q4 1987 Q1 Q2 Q3 Q4 1988 Q1 Q2 Q3 Q4 1989 Q1 Q2 Q3 Q4 M1 M2 0.8 0.6 - 0 . 3 -0 .3 - 0 . 4 -0 .2 - 0 . 3 -0 .3 1.0 1.0 0.1 - 0 . 2 - 0 .4 0.1 - 0 . 7 - 0 .3 -0 .4 -0 .1 - 0 .2 0.1 -0 .1 -0 .1 -0 .1 0.0 - 0 .2 0.0 0.0 0.1 - 0 .2 -0 .1 0.1 -0 .1 0.0 0.1 0.1 0.1 0.0 -0 .1 -0 .1 -0 .1 - 0 .3 0.1 0.2 0.0 0.0 0.0 0.0 0.0 0.3 -0 .2 -0 .2 0.1 0.0 0.0 0.0 0.0 -0 .3 -0 .3 0.0 - 0 .4 -0 .1 0.3 0.6 0.6 - 0 .2 0.0 0.4 0.3 0.0 0.1 - 0 .5 -0 .3 0.0 0.0 0.1 0.2 0.0 -0 .1 - 0 .2 0.0 -0 .3 -0 .1 0.0 0.0 0.2 0.0 0.1 0.0 0.2 -0 .1 -0 .1 0.1 0.0 0.0 0.0 0.0 -0 .5 - 0 .4 0.0 - 0 .5 - 0 .3 0.3 1.0 0.7 -0 .1 0.2 0.7 0.4 0.1 - 0 .2 - 0 .9 -0 .5 0.0 -0 .1 0.3 0.2 - 0 .2 -0 .2 -0 .1 0.0 -0 .2 -0 .1 0.2 0.1 0.0 0.0 0.1 0.0 0.1 -0 .1 -0 .1 0.1 0.0 0.0 0.0 0.0 -0 .1 0.3 1.1 0.5 0.1 - 0 .3 -1 .3 -0 .6 -0 .6 0.0 -0 .1 0.4 0.3 - 0 .3 -0 .2 -0 .1 0.0 -0 .2 -0 .1 - 0 .2 -0 .1 0.4 0.2 0.1 0.0 0.0 0.0 -0 .1 0.1 0.0 0.0 0.1 0.0 0.0 0.4 0.2 1.3 0.5 0.6 0.2 - 0 .3 -0 .1 - 1 . 6 - 0 . 8 - 0 .6 0.0 -0 .1 0.4 0.4 - 0 .4 -0 .3 0.0 0.0 - 0 .5 - 0 .3 -0 .1 - 0 .3 - 0 .2 0.5 0.3 0.0 0.0 -0 .2 -0 .1 0.0 0.0 0.1 0.2 0.0 0.0 0.1 - 0 .2 -0 .1 0.4 0.5 -0 .1 - 0 .4 - 0 .3 -0 .3 -0 .1 0.0 0.6 - 0 .4 0.0 - 0 .3 - 0 .4 0.6 0.4 0.2 0.0 -0 .4 -0 .3 0.1 0.1 0.7 0.4 -0 .3 -0 .1 0.7 0.1 0.0 -0 .1 0.0 - 0 .3 0.0 0.2 0.3 0.4 0.1 0.2 -0 .4 0.2 0.4 -0 .3 -0 .1 -0 .2 - 0 .2 0.1 0.3 0.4 -0 .6 0.2 0.1 0.6 - 0 .2 -0 .3 -0 .2 - 0 .2 -0 .2 0.1 0.0 M2 M2 M2 M3 0.0 0.0 0.1 -0 .1 0.2 0.1 -0 .3 -0 .1 -0.1 M1 M2 0.0 M3 0.2 0.7 - 0 .6 - 0 .7 -0 .3 -0 .1 - 0 . 6 -0 .5 - 0 .2 0.5 0.1 1.1 0.9 0.6 M3 1990 Q1 Q2 Q3 Q4 1991 Q1 Q2 Q3 Q4 1992 Q1 Q2 Q3 Q4 Note: These revisions shown do not include effects of benchmark data revisions to and/or changes in definition. M1 M2 M3 1993 M1 M1 M1 1992 1991 1990 1989 1988 M3 CO 1984 Q4 1985 Q1 Q2 Q3 Q4 M3 o I M2 CD M1 o I BANK O ST. LOUIS F Periods M3 M1 M2 M3 0.1 -0 .1 0.2 - 0 .6 0.0 -0 .6 - 0 .2 - 0 .4 -0 .2 0.6 0.6 0.6 0.2 0.0 0.2 -0 .8 -0 .4 0.3 0.1 1.1 0.6 - 0 .4 - 0 .2 - 0 .3 - 1 .2 - 0 .5 -0 .2 0.6 0.1 0.3 1.4 0.7 0.4 - 0 .7 - 0 .5 -0 .3 oo Table 3 (cont.) Page 3: Revisions to Previously Published Quarterly Growth Rates of the Monetary Aggregates (s.a.) Due to Changes in Definition_______________________________________________________________________________ Year of redefinition (published at time of benchmark and seasonal review) 1986 Quarters M1 M2 1987 M3 M1 M2 1988 M3 1989 M1 M2 M3 1986 Q1 Q2 Q3 Q4 0.4 0.0 0.1 0.4 0.3 0.2 0.0 0.0 0.0 0.0 M2 1990 M3 M1 M2 1991 M3 M1 M2 1992 M3 M1 M2 1993 0.0 1987 Q1 Q2 Q3 Q4 M1 M3 M1 0.0 0.0 0.0 0.0 M2 M3 1984 Q4 1985 Q1 Q2 Q3 Q4 CO 1988 Q1 Q2 Q3 Q4 0.0 1989 Q1 Q2 Q3 Q4 0.0 0.0 0.0 0.0 - - 0.1 0.0 0.1 0.1 0.2 0.1 0.0 0.0 0.0 0.0 1990 Q1 Q2 Q3 Q4 MARCH/APRIL 1994 1991 Q1 Q2 Q3 Q4 Note: These revisions shown do not include effects due to benchmark data revisions and changes in seasonal adjustment factors. Source: Data shown in shaded areas are taken from the issues of the Federal Reserve Board’s H.6 statistical release, published after the annual benchmark. See Kavajecz (1994) for exact dates. Other data shown are the authors’ calculations from annual issues of Money Stock Revisions. 20 nual benchmark data.2 Monthly seasonal factors 0 are estimated by a variant of the Statistics Cana da Xll-ARIMA method.2 In the first step of this 1 method, the observed data are extended by the addition of one or two years of forecasts. The forecasts are obtained via an ARIMA model that includes exogenous intervention variables for each month and, in some cases, a small number of special events.2 In recent years, intervention 2 variables have been included for events such as the impact of the 1986 Tax Reform Act on the levels of liquid deposits in early 1987 and the dramatic surge in M l that occurred during Hurricane Gloria’s sweep up the east coast of the United States in September 1985. Seasonal factors are then obtained by applying standard X ll algorithms to the lengthened series. Weekly seasonal factors are estimated via a two-step process. In the first, initial estimates of weekly seasonal factors are obtained from an unobserved-components time series model.2 In 3 the second, these initial estimates are modified via a quadratic programming model such that averages of a particular path o f seasonally ad justed weekly data equal the previously estimat ed monthly seasonal pattern.2 Projected weekly 4 seasonal factors are obtained in a similar man ner, subject to judgmental adjustment by Board staff for events such as unusual calendar dating and holiday effects that are not captured by the statistical models. Like other aspects of the monetary aggregates, the methods used for seasonal adjustment have evolved over time. From 1955 — when the first seasonally adjusted numbers were published — through 1981, seasonal adjustment was done us ing the classic Census X ll procedure.2 In 1982, 5 the Xll-ARIMA procedure proposed by Dagum was adopted to reduce well-known potential problems due to the use of truncated moving20The very few exceptions in which the seasonal review was completed and published after the benchmark are noted in Kavajecz (1994). 21See Farley and O’Brien (1987). 22See Box and Tiao (1975). 23The statistical model has been developed over a number of years; see Cleveland and Grupe (1983), Pierce, Grupe and Cleveland (1984), and Cleveland (1986). The model allows for a noninteger number of weeks during the year and other effects. Statistically, it seeks to estimate trend, seasonal and irregular components of a time series that is sampled at a frequency which differs from the fundamental frequencies of the data generating processes for its com ponents. 24See the appendix to Farley and O'Brien (1987) for details of the algorithm. 25See Pierce and Cleveland (1981). http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS FEDERAL Federal Reserve Bank of St. Louis average filters near the ends of the sample.2 6 Other features that have been added to improve the estimation include trading day effects, pay ment schedules and holiday dating. Following recommendations of the Advisory Committee on Monetary Statistics, the Federal Reserve publishes both seasonally adjusted and unadjusted data. The weekly H.6 release, for ex ample, currently includes adjusted data for four monetary aggregates and 25 components, and unadjusted data for the four aggregates, 26 com ponents and 11 related miscellaneous series. Most of the adjusted components are furnished for ease of analysis, however, and are not used in construction o f the monetary aggregates. Seasonally adjusted M l is constructed as the sum of four separately adjusted components: currency, travelers checks, demand deposits and other checkable deposits (OCDs). The non-Ml component of M2 and the non-M2 component of M3 are adjusted as a whole, with adjusted M2 equal to the sum of adjusted M l and the non-Ml component of M2; M3 similarly is formed by summing M2 and the adjusted nonM2 component of M3. Early each year, Board staff forecast seasonal adjustment factors for the monetary aggregates during the coming year. These projected factors are published on the H.6 release at the same time as the benchmark data, and are not re vised during the year on the basis o f incoming data.2 Hence, published monetary growth rates 7 throughout the year are based on ex ante fixed seasonal factors that incorporate no information received during the current year. Thus, it perhaps is not surprising that revised seasonal factors for the most recently completed year may differ significantly from those that were forecast a year earlier. Revisions to the mone tary aggregates due to revisions to seasonal fac26While X11 uses two-sided moving-average filters for most observations, the filters must be truncated near the ends of the time series. This effect tends to increase the size of the revisions to the most recent year’s seasonal factors when they are reestimated the following year. Further, it also tends to underestimate the degree of seasonality near the end of the sample. Extending the sample via ARIMA model forecasts seems to attenuate both problems. See Dagum (1983). 27Experimental estimates of concurrent seasonal factors, up dated using incoming data, were published as an appendix to the H.6 for several years but never incorporated into any official monetary aggregate. The Board’s committee of ex perts on seasonal adjustment had recommended explora tion of concurrent factors; see Pierce and Cleveland (1981). A similar recent review at the Bank of England (1992) sug gested that concurrent adjustment might reduce the size of subsequent revisions. 21 tors, shown on the second page of Table 3, often have exceeded those due to either revi sions to underlying source data (shown on the first page of the table) or to changes in defini tions (the third page of Table 3). mation about the stance of monetary policy with respect to economic activity. The Federal Reserve responded by creating the monetary ag gregates M2 and M3 in 1971, and M4 and M5 in 1975. Although the concept of seasonal movements in data may be fairly straightforward, there is no generally accepted statistical definition of seasonality. ‘"Ii'ue” seasonal factors are never ob served nor measured, even with error. Thus, seasonally adjusted monetary aggregates neces sarily retain a significant subjective component, even in the long run. Lindsey and others (1981) notes that the adjusted monetary aggregates have tended to become somewhat smoother through time as their seasonal adjustment fac tors have been subjected to successive annual revisions. Although he attributes this to in creases in our knowledge about, and precision in, estimation of the seasonal adjustment factors, an alternative hypothesis is that the seasonal component is absorbing more of the irregular component, leaving an adjusted time series that more closely resembles its trend component. Despite the increasing attention focused on near-moneys, the multiple definitions of the monetary aggregates during the 1970s continued to reflect legislative distinctions between the as set and liability powers of banks and thrifts. These distinctions faded after passage o f the Monetary Control and Garn-St. Germain Acts, permitting a new set of nested definitions such that M l became a subset of M2, and M2 a sub set of M3.2 By internalizing within M2 9 opportunity-cost-induced shifts of funds be tween medium-of-exchange and liquid near moneys for all intermediaries, this design en hanced the usefulness of M2 as an intermediate policy target through better estimates of a (nominally) stable demand curve for M2.3 0 Changes In D efinitions Although financial innovation has been an im portant factor, the evolution of the Federal Reserve Board staff’s definitions o f monetary ag gregates primarily has been governed by econo mists' changing empirical perceptions of the appropriate concept of money.2 In the 1960s, 8 economists’ focus on the medium o f exchange function of money made M l the principal ag gregate. As empirical relationships for M l ap peared to break down in the 1970s and attention turned once again to the role of liquid near moneys, some suggested that multiple monetary aggregates might collectively reveal more infor 28Our view is that many of the theoretical arguments for the inclusion and/or exclusion of specific assets are ex post ra tionalizations of workable empirical definitions. The same argument is, of course, made by Friedman and Schwartz (1970). 29There are a few qualifications to this characterization. From 1980-87, a portion of the vault cash and demand deposits held by thrifts had been included in M1 (but not in M2 and M3), while the balance was excluded (none of the vault cash and interbank deposits held by commercial banks were included in the aggregates). In 1988, the treatment of these items for thrifts was changed to be comparable to that for banks. Similarly, in constructing M3, a variety of netting items are deducted, such as large time deposits at commercial banks held by M2-type money market funds. In general, in moving from narrower to broader aggregates, any asset held by a money stock issuer (say, a money market fund) that was issued by another money stock Since monetary aggregates data first appeared on the J.3 statistical release in 1960, the broad monetary aggregates (roughly corresponding to M l, M2, M3) have been redefined about a dozen times. Changes have ranged in magnitude from the massive redefinition in February 1980 to small additions and subtractions such as the in clusion of nonbank travelers checks in June 1981. Whenever a definition change is put in place, Board staff recompute all historical data for the monetary aggregates and components under the most recent definitions.3 Available 1 Federal Reserve publications, including Money Stock Revisions, show monetary aggregates data solely in terms of current definitions. For re searchers studying Federal Reserve behavior, "knowing what money was" at a particular time is complicated by changes in definitions as well issuer (say, a commercial bank) is netted out of the broad er consolidated monetary aggregate. 30For discussion, see Simpson and Porter (1980). 3,The 1980 redefinition, for example, required Board staff to “ rebuild” M2 for years prior to 1980 with an expanded set of thrift deposit data. Some details are discussed in the appendix. MARCH/APRIL 1994 22 as by the annual benchmark and seasonal review process. Definitional changes perhaps are usefully sum marized in three categories. First, there is the inclusion (or, less often, exclusion) of an existing money market instrument or depository liabili ty.3 A prominent example is the addition in 2 1980 of general purpose and broker/dealer MMMFs to the M2 aggregate.3 While M2 was 3 recomputed on a consistent basis for all prior periods following the redefinition, conceptually this is a nontrivial change. During the 1970s, when the first surge in money market fund growth occurred, the contemporaneous M2 ag gregate excluded money market funds; shifts by households into the funds were (in principal) embedded in the elasticity o f M2 with respect to its opportunity cost and reflected in shifts in the income velocity of M2. Researchers using the redefined M2, however, see an aggregate that in ternalizes these shifts, has a smaller interest elasticity, and different velocity behavior. Of course, the importance of this change in defini tion for analysis o f Fed behavior is mitigated by the FOMC’s emphasis on M l during the period. Other examples are the inclusion in M2 o f retail RPs (which were basically uninsured small time deposits exempt from Reg Q) in 1982, the exclu sion of retirement accounts from the monetary aggregates in 1983, and the addition o f term Eu rodollar deposits to M3 in 1984. While the last had been discussed earlier, inclusion of the deposits had to await a reliable source of data. The second type of definition change is the in clusion of a new money market instrument or depository institution liability. In some cases, the new instrument or deposit may simply reflect the removal o f a prohibition against that type of deposit or of a ceiling on a deposit offering rate (Regulation Q ceilings). To the ex 32The precise definition of M1 has changed several times due to changes in the treatment of demand deposits due to foreign commercial banks and official institutions. Includ ed in M1 prior to 1980 (see Kavajecz, 1994), these deposits were excluded thereafter following recommendations of the Advisory Committee on Monetary Statistics. See Advisory Committee on Monetary Statistics (1976), p. 4, or Farr and others (1978). These changes also complicate building M1 based on current definitions for years prior to 1959; see Rasche (1987). 33Tax-exempt general purpose and broker/dealer MMMFs, ex cluded in 1980, were added in February 1983. 34See Kavajecz (1994) for details. More obscure examples in clude certain assets sold by depositories with recourse, bank investment contracts (BICs), and bank deposit notes (the latter classified as a deposit under Federal Reserve Regulation D but not by the FDIC). Brokered deposits pro http://fraser.stlouisfed.org/RESERVE BANK OF ST. LOUIS FEDERAL Federal Reserve Bank of St. Louis tent that deregulation or the authorization of new instruments permanently changes the be havior of depositories, its affect on the monetary aggregates is similar to a change in definition. Examples include the authorization of NOW ac counts nationwide in 1980, the introduction of money market deposit accounts (MMDAs) in 1983, and the major discrete steps in the phaseout of Regulation Q that occurred in 1982, 1983 and 1986.3 In many cases, this type of 4 deposit account was already included in the ag gregates (both OCDs and MMDAs are types of savings deposits). The authorization o f these new instruments, largely born of deposit in terest rate controls, likely induced unusual tran sitory volatility in published data during the period when money may be shifting between components and may also have permanently changed the income and interest elasticities of the monetary aggregate.3 5 The third type o f definition change is reclas sification o f the liabilities o f different types of financial institutions. Prior to the 1980 redefini tion, deposits at banks and thrift were included in separate monetary aggregates. Deposits at thrifts were included in M3 and M5 while com parable deposits at banks were included in M2 and M4. The 1980 redefinition restructured the monetary aggregates to combine similar types of deposits at commercial banks and thrifts. Al though strongly motivated by the increasing similarity of the deposits offered by banks and thrifts during the 1970s, some economists coun selled against the pooling of bank and thrift lia bilities in the new aggregates. Their arguments were based largely on the joint product nature of depositories. To the extent that firms and households tend to purchase a bundle o f serv ices from a single institution rather than separate products from a number o f institutions, there may be value to aggregation by institutionvide another example. Although a bank or thrift might receive a deposit of a million dollars (or more) from a broker, the amount of the deposit is included in M2 as a small time deposit if the deposit is placed entirely for the benefit of individuals. In this manner, the development of the brokered retail CD market could potentially have affect ed the apparent interest elasticity of M2 by altering the be havior of its small time deposit component. 35There is no doubt this was the case in 1983, when the FOMC decided to rebase its target growth rate ranges for the year following the introduction of MMDAs. The implica tions of deregulation during the 1980s, including the demise of Reg Q, for money demand models are dis cussed by Moore, Porter and Small (1990). 23 al type rather than by product. In response, the Board adopted the recommendation that, to ev ery extent feasible, data for banks and thrifts should be published separately so as to permit such analysis. This argument is similar to Fried man and Schwartz’s position that financial assets may appropriately be aggregated if they are sufficiently close substitutes in either demand or supply. Overall, annual revisions to the monetary aggregates due to revisions to source data, sea sonal factors and definitions render treacherous any attempt by a researcher to update or extend previous studies by mixing differing vintages of monetary aggregates data. One recent empirical study (Dewald, Thursby and Anderson, 1986) found in an extensive computer simulation ex periment that empirical results may be highly sensitive to the mixing o f different vintages of data, including data on the monetary aggregates. A complete chronology of revisions and redefini tions of the monetary aggregates is shown in Kavajecz (1994). CONCLUSION: THE MONETARY AGGREGATES AS MONETARY TARGETS We conclude our historical examination of the Federal Reserve’s monetary aggregates with a summary of their use as monetary policy tar gets. The FOMC’s target and monitoring ranges for the aggregates are shown in T&ble 4.3 G Targeting of monetary aggregates began with House Concurrent Resolution 133 in 1975, later formalized in the Humphrey-Hawkins Act of 1978 as an amendment to the Federal Reserve Act. From 1975 through 1978, the committee rebased each quarter its annual four-quarter tar get range for the monetary aggregates. The resulting base drift in the committee’s targets has been controversial.3 Since 1978, the com 7 mittee has set one, fourth-quarter-to-fourthquarter range each year except 1983. Authoriza tion of MMDAs in late 1982 led to a surge in M2 growth as aggressive bidding by depositories against money market funds apparently drew nonmonetary balances into M2. (Recall that taxa ble general purpose and broker/dealer MMMFs had been included in M2 in 1980 and that MMDAs, a type of savings deposit, were always included in M2. M2 was redefined slightly in February 1983 to include tax-exempt general purpose and broker/dealer money market funds.) The committee subsequently reset its 1983 tar get ranges using a February-March base. While relatively narrow through the early 1980s, target ranges widened during the decade as an accelerating pace of innovation in financial markets apparently complicated money demand forecasting and money stock control. The range for M l was widened to 4 percentage points in 1983 and to 5 points in 1985. Citing uncertainty regarding the demand for M l and its relation ship to economic activity, the committee did not set a target range for M l in 1987 or beyond.3 8 The target range for M2 similarly was widened over this interval, although it has remained at its current width of 4 percentage points since 1988. In part, the widening of the range in 1988 reflects the increased difficulty of forecasting the demand for M2 during an era o f turmoil in financial markets, including the restructuring of the thrift industry, capital and earnings difficul ties at commercial banks, and a restructuring (deleveraging) o f household and firm balance sheets. The monetary aggregates during most years have grown within their target ranges, as shown in Figures 1 and 2. Growth often has run well toward the upper or lower bounds of the cones, however, suggesting that the midpoint of the committee’s target range may not always be the best forecast o f an aggregate’s growth. 36Target and monitoring ranges differ in terms of the strength of the implied policy reaction function. In general, deviation of an aggregate from a target range suggests a somewhat stronger policy response than deviation from a monitoring range, ceteris paribus. 37For contrasting views, see for example Axilrod (1982), Broaddus and Goodfriend (1984) and Walsh (1986). “ “ Monetary Policy Report to the Congress,” Federal Reserve Bulletin, April 1987. MARCH/APRIL 1994 24 Table 4 Growth Cones for the Monetary and Credit Aggregates (percent annual rate)_______________________________ Target and monitoring ranges Date established Apr.75 Jun.75 Jul.75 Oct.75 Jan.76 Apr. 76 Jul.76 Nov.76 Jan.77 Apr. 77 Base period Mar. 75 Jun.75 7502 75Q3 7504 7601 7602 76Q3 7604 7701 Span Mar.75-Mar.76 Jun.75-Jun.76 75Q2 - 7602 7503 - 7603 7504 - 7604 7601 - 7701 7602 - 7702 7603 - 7703 7604 - 7704 7701 - 7801 M1 5.0 5.0 5.0 5.0 4.5 4.5 4.5 4.5 4.5 4.5 - M2 7.5 7.5 7.5 7.5 7.5 7.0 7.0 6.5 6.5 6.5 8.5 8.5 8.5 7.5 7.5 7.5 7.5 7.5 7.0 7.0 - Bank credit proxy M3 10.5 10.5 10.5 10.5 10.5 10.0 9.5 10.0 10.0 9.5 10.0 10.0 10.0 9.0 9.0 9.0 9.0 9.0 8.5 8.5 - 12.0 12.0 12.0 12.0 12.0 12.0 11.0 11.5 11.5 11.0 6 . 5 - 9.5 6.5 - 9.5 6 . 5 - 9.5 6 . 0 - 9.0 6.0 - 9.0 6 . 0 - 9.0 5.0 - 8.0 5 . 0 - 8.0 7.0 - 10.0 7.0 - 10.0 Bank credit Jul.77 Oct.77 Feb.78 Apr.78 Jul.78 Oct.78 Feb. 79 Feb.80 Feb.81 Feb.82 7702 77Q3 7704 7801 7802 7803 7804 7904 8004 8104 7702 7703 7704 7801 78Q2 7803 7804 7904 8004 8104 - 7802 7803 7804 7901 79Q2 7903 7904 8004 8104 8204 4.0 4.0 4.0 4.0 4.0 2.0 1.5 4.0 3.5 2.5 - 6.5 6.5 6.5 6.5 6.5 6.0 4.5 6.5(M1B) 6.0{M1B) 5.5 7.0 6.5 6.5 6.5 6.5 6.5 5.0 6.0 6.0 6.0 - 9.5 9.0 9.0 9.0 9.0 9.0 8.0 9.0 9.0 9.0 8.5 8.0 7.5 7.5 7.5 7.5 6.0 6.5 6.5 6 .5 - 11.0 10.5 10.0 10.0 10.0 10.0 9.0 9.5 9.5 9.5 7.0 7.0 7.0 7.5 8.5 8.5 7.5 6 .0 6.0 6 .0 - 10.0 10.0 10.0 10.5 11.5 11.5 10.5 9.0 9.0 9.0 Debt Feb.83 Feb.83 Jul.83 Jan.84 Feb. 85 Jul.85 Feb.86 Feb.87 Feb.88 Feb.89 Feb. 90 Jul.90 Feb.91 Feb.92 Feb.93 Jul.93 83Feb/Mar 8204 8302 8304 8404 85Q2 85Q4 86Q4 87Q4 8804 8904 89Q4 90Q4 9104 9204 9204 83Feb/Mar-83Q4 8204 - 8304 8302 - 8304 83Q4 - 84Q4 84Q4 - 85Q4 85Q2 - 85Q4 8504 - 86Q4 86Q4 - 87Q4 87Q4 - 88Q4 8804 - 8904 8904 - 9004 8904 - 90Q4 9004 - 9104 9104 - 92Q4 9204 - 9304 9204 - 9304 _ 4.0 5.0 4.0 4.0 3.0 3.0 - 8.0 9.0 8.0 7.0 8.0 8.0 NS NS NS NS NS NS NS NS NS 7.0 - 10.0 _ _ NC 6.0 - 9.0 6.0 - 9.0 NC 6.0 - 9.0 5.5 - 8.5 4.0 - 8.0 3.0 - 7.0 3.0 - 7.0 NC 2.5 - 6.5 2.5 - 6.5 2.0 - 6.0 1.0 - 5.0 6.5 - 9.5 NC 6 . 0 - 9.0 6 . 0 - 9.5 NC 6 . 0 - 9.0 5 . 5 - 8.5 4.0 - 8.0 3 . 5 - 7.5 2.5 - 6.5 1 . 0 - 5.0 1 . 0 - 5.0 1 . 0 - 5.0 0 . 5 - 4.5 0 . 0 - 4.0 8.5 - 11.5 NC 8.0 - 11.0 9.0 - 12.0 NC 8.0 - 11.0 8.0 - 11.0 7.0 - 11.0 6.5 - 10.5 5.0 - 9.0 NC 4.5 - 8.5 4.5 - 8.5 4.5 - 8.5 4.0 - 8.0 The FOMC first set desired longer-run growth targets for M1, M2, M3 and the bank credit proxy at its meeting on April 14-15, 1975. On February 15, 1977, ranges for the monetary aggregates were added to the Domestic Policy Directive sent to the Open Market Desk at the Federal Reserve Bank of New York. On April 18, 1978, the range for bank credit was added to the Domestic Policy Directive. NC: Not Changed NS: None Specified http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis 25 Figure 1 M2 Historical Target Ranges Billions of dollars Quarterly data Figure 2 M3 Historical Target Ranges Billions of dollars Quarterly data MARCH/APRIL 1994 26 REFERENCES Abbott, William J. “ Revision of Money Supply Series,” Feder al Reserve Bulletin (August 1962), pp. 941-51. ________“A New Measure of the Money Supply,” Federal Reserve Bulletin (October 1960), pp. 1102-23. Advisory Committee on Monetary Statistics. Improving the Monetary Aggregates: Report of the Advisory Committee on Monetary Statistics. Board of Governors of the Federal Reserve System, 1976. Axilrod, Stephen. “ Comments in ‘Is the Federal Reserve’s Monetary Control Policy Misdirected? Resolved: That the Federal Reserve’s Current Operating Procedures for Con trolling Money Should be Replaced,’ ” Journal of Money, Credit and Banking (February 1982), pp. 119-47. Bank of England. Report of the Seasonal Adjustment Working Party, Occasional Paper no. 2 (October 1992). Barnett, William A. “ Economic Monetary Aggregates: An Application of Index Number and Aggregation Theory,” Journal of Econometrics, Annals of Applied Econometrics 1980-3, a supplement (1980), pp. 11-48. Beck, Darwin L. “Sources of Data and Methods of Construc tion of the Monetary Aggregates,” in Improving the Mone tary Aggregates: Staff Papers. Board of Governors of the Federal Reserve System, 1978, pp. 117-33. Board of Governors of the Federal Reserve System. All-Bank Statistics 1896-1955. Board of Governors of the Federal Reserve System, 1959. ________Annual Report. _______ . Annual Statistical Digest. _______ . Banking and Monetary Statistics 1941-1970. Board of Governors of the Federal Reserve System, 1976. ________Banking and Monetary Statistics. Board of Gover nors of the Federal Reserve System, 1943. _______ . Demand Deposits, Currency, and Related Items (J.3). _______ . Federal Reserve Bulletin. ________Money Stock, Liquid Assets, and Debt Measures (H.6). ________Money Stock Revisions, annual supplement to the H.6. Box, G.E.P., and G.C. Tiao. “ Intervention Analysis With Applications to Economic and Environmental Problems,” Journal of American Statistical Association (March 1975), pp. 70-9. Broaddus, Alfred. “Aggregating the Monetary Aggregates: Concepts and Issues,” Federal Reserve Bank of Richmond Economic Review (November/December 1975), pp. 3-12. _______ , and Marvin Goodfriend. “ Base Drift and the Longer Run Growth of M1: Experience from a Decade of Monetary Targeting,” Federal Reserve Bank of Richmond Economic Review (November/December 1984), pp. 3-14. Cleveland, William P. “ Calendar Adjustment and Time Ser ies,” Board of Governors of the Federal Reserve System Special Studies Paper, Division of Research and Statistics, no. 198 (October 1986). Copeland, Morris A., and Daniel H. Brill. “ Banking Assets and the Money Supply Since 1929,” Federal Reserve Bulle tin (January 1948), pp. 24-32. Dagum, Estella Bell. The X-11-ARIMA Seasonal Adjustment Method. Statistics Canada, 1983. Dewald, William G., Jerry G. Thursby, and Richard G. Anderson. “ Replication in Empirical Economics: The Jour nal of Money, Credit and Banking Project,” The American Economic Review (September 1986), pp. 587-603. Duprey, James N. “ How the Fed Defines and Measures Money,” Federal Reserve Bank of Minneapolis Quarterly Review (Spring-Summer 1982), pp. 10-9. Farley, Dennis E., and Yueh-Yun C. O’Brien. “ Seasonal Ad justment of the Money Stock in the United States,” Journal of Official Statistics (1987, vol. 3, no. 3), pp. 223-33. Farr, Helen T., Lance Girton, Henry S. Terrell, and Thomas H. Turner. “ Foreign Demand Deposits at Commercial Banks in the United States,” in Improving Monetary Aggregates: Staff Papers. Board of Governors of the Federal Reserve Sys tem, 1978, pp. 35-54. Friedman, Milton, and Anna J. Schwartz. Monetary Statistics of the United States. Columbia University Press, 1970. Fry, Edward R., Darwin Beck, and Mary F. Weaver. “ Revision of the Money Stock,” Federal Reserve Bulletin (December 1970), pp. 887-909. Gilbert, R. Alton. “ Requiem for Regulation Q: What it Did and Why It Passed Away,” this Review (February 1986), pp. 22-37. Hallman, Jeffrey J., Richard D. Porter, and David H. Small. “ Is the Price Level Tied to the M2 Monetary Aggregate in the Long Run?” The American Economic Review (Septem ber 1991), pp. 841-58. Kavajecz, Kenneth A. “The Evolution of the Federal Reserve’s Monetary Aggregates: A Timeline,” this Review (March/April 1994). Laidler, David W. The Demand for Money: Theories, Evidence, and Problems, 4th edition. Harper and Row, 1993. Lang, Richard W. “ Benchmark Revisions of the Money Stock and Ranges of Money Stock Growth,” this Review (June 1 9 7 8 ), pp. 11 -9 . Lawler, Thomas A. “ Seasonal Adjustment of the Money Stock: Problems and Policy Implications,” Federal Reserve Bank of Richmond Economic Review (November/December 1977), pp. 19-27. Lindsey, David, and others. “ Monetary Control Experience Under the New Operating Procedures,” in New Monetary Control Procedures. Board of Governors of the Federal Reserve System, 1981. Moore, George R., Richard D. Porter, and David H. Small. “ Modeling the Disaggregated Demands for M2 and M1: The U.S. Experience in the 1980s,” in Peter Hooper and others, eds., Financial Sectors in Open Economies: Empiri cal Analysis and Policy Issues. Board of Governors of the Federal Reserve System, 1990, pp. 21-105. National Credit Union Association. Annual Report. _______ , and Michael R. Grupe. “ Modeling Time Series When Calendar Effects Are Present,” in Arnold Zellner, ed., Conference on Applied Time Series Analysis of Economics Data. U.S. Department of Commerce, 1983, pp. 57-67. Orphanides, Athanasios, Brian Reid, and David Small. “The Empirical Properties of a Monetary Aggregate That Adds Bond and Equity Mutual Funds to M2.” Board of Gover nors of the Federal Reserve System Financial and Eco nomics Discussion Paper no. 93-42, Division of Monetary Affairs (December 1993). Collins, Sean, and Cheryl L. Edwards. “ Redefining M2 to Include Bond and Equity Mutual Funds,” mimeo. Board of Governors of the Federal Reserve System, 1994. Pierce, David A., and William P. Cleveland. “ Seasonal Adjust ment Methods for the Monetary Aggregates,” Federal Reserve Bulletin (December 1981), pp. 875-87. http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis 27 _______ , Michael R. Grupe, and William P Cleveland. . “ Seasonal Adjustment of the Weekly Monetary Aggregates: A Model-Based Approach,” Journal of Business and Eco nomic Statistics (July 1984), pp. 260-70. _______ , and John R. Williams. “ Recent Revisions in the Money Stock: Benchmark, Seasonal Adjustment, and Cal culation of Shift-Adjusted M1-B,” Federal Reserve Bulletin (July 1981), pp. 539-42. Rasche, Robert H. “ Demand Functions for Measures of U.S. Money and Debt,” in Peter Hooper and others, eds., Finan cial Sectors in Open Economies: Empirical Analysis and Policy Issues. Board of Governors of the Federal Reserve System, 1990, pp. 113-61. Spindt, Paul A. “ Money Is What Money Does: Monetary Aggregation and the Equation of Exchange,” Journal of Po litical Economy (1985), pp. 175-204. _______ . “M1-Velocity and Money-Demand Functions: Do Stable Relationships Exist?” Carnegie-Rochester Confer ence Series on Public Policy (autumn 1987), pp. 9-88. Simpson, Thomas D. “The Redefined Monetary Aggregates,” Federal Reserve Bulletin (February 1980), pp. 97-114. ________“A Proposal for Redefining the Monetary Aggre gates,” Federal Reserve Bulletin (January 1979), pp. 13-33. _______ , and Richard Porter. “ Some Issues Involving the Definition and Interpretation of the Monetary Aggregates,” Controlling Monetary Aggregates III, Federal Reserve Bank of Boston Conference Series no. 23 (October 1980), pp. 161-234. Van Peski, Neva. “Appendix: Data Sources and Construction of the Proposed Monetary Aggregates,” Federal Reserve Bulletin (January 1979), pp. 33-42. Walsh, Carl. “ In Defense of Base Drift,” The American Eco nomic Review (September 1986), pp. 692-700. Walter, John R. “ Monetary Aggregates: A User’s Guide,” Fed eral Reserve Bank of Richmond Economic Review (Janu ary/February 1989), pp. 20-8. Walton, Jack, and others. “ Performance Evaluation of Money Stock, Liquid Assets, and Debt Measures (H.6) Statistical Release,” unpublished memorandum. Board of Governors of the Federal Reserve System, Division of Monetary Affairs, 1991. MARCH/APRIL 1994 28 Appendix Building Historical Monetary Aggregates 1959-80 The 1980 redefinition of the monetary ag gregates confronted Board staff with the daunt ing task of building comparable historical data. In some cases, large amounts o f additional data needed to be collected. In others, various esti mates and approximations had to be made since required historical data had not been collected in the needed detail, at the desired frequency, or on the basis of consistent definitions. Although the data sources available as o f 1977 have been described elsewhere, little has been written about the earlier data.1 This appendix, based on published and unpublished material, summarizes available information about the data sources and methods used to construct monetary aggregates for years prior to 1980. Monetary aggregates are built by consolidation of data, not addition. Consolidation requires not only data on the types and amounts of outstand ing liabilities of financial intermediaries but also data on the ownership of such liabilities by other money-stock-issuing institutions, the latter being netted from the aggregate during consoli dation. So far as possible, the discussion below reviews available data on both items. DEPOSITS INCLUDED IN M l Most commercial bank deposit items were available at least twice a year from call reports. Demand deposits had been reported by member banks since well before 1959. Call report data were available quarterly from all banks begin ning in 1961, when quarterly call reports be came required by law. Daily data on OCD accounts were available for member banks. End-of-month data begin ning in September 1972 for other New England financial institutions were obtained from the Federal Reserve Bank of Boston. MSBs issued two types of demand deposits. One was used for regular third-party payments, 'Beck (1978) describes data available in 1977 and refers to unpublished memoranda for earlier sources and methods. Our discussion here draws from unpublished Federal Reserve Board memoranda by Neva Van Peski and Darwin Beck, and from Van Peski (1979). We thank them for help ful comments while absolving them of responsibility for remaining errors or omissions. 2The report form filed by weekly reporting banks had been revised in 1961 and 1966 to improve coverage of these http://fraser.stlouisfed.org/ FEDERAL Federal Reserve Bank RESERVE BANK OF ST. LOUIS of St. Louis that is, it was checkable. The other consisted mainly of escrow balances, not used for regular payments. Only the first is included in the monetary aggregates. Separation of the two types of deposits prior to 1980 was based on month-end data collected by the FDIC during an 18-month survey conducted from July 1975 to December 1976. The survey data themselves were included in M l for the 18 months they were available. Before and after this period, data on total demand deposits reported on semi annual or quarterly call reports were multiplied by the average ratio of checkable to total de mand deposits during the survey period. Month ly data were obtained by interpolation. Share draft balances at federal credit unions were obtained from the National Credit Union Administration (NCUA) as o f month-end for MaySeptember 1976. Thereafter, only end-of-quarter data were available. No data were available on share drafts at state credit unions. For total credit union savings deposits, as o f July 1977, federal credit unions held 55 percent o f savings deposits; their share o f share draft accounts is unknown. Under the 1980 definition o f the monetary ag gregates, demand deposits at commercial banks due to thrifts, foreign banks and foreign official institutions are subtracted from total demand deposits in building M l (see Table 2). Demand deposits at U.S. commercial banks due to foreign commercial banks and official institutions were available weekly (on Wednesday) for weekly reporting banks since May 1961, and quarterly or twice a year from call reports for all banks since (at least) 1959.2 Ml-type deposits at foreign-related institutions were available as of the last Wednesday of the month since Novem ber 1972 (beginning in 1977, Edge Act corpora tions reported only quarterly, but other institutions continue to report monthly). For earlier years, estimates were based on data items; see the introduction to chapter 4 in Banking and Monetary Statistics 1941-1970. Ironically, these data were originally collected from weekly reporting banks so that they could be added back into the monetary aggregates after being removed during earlier adjustments. Following the 1980 redefinitions, these reported data were used to remove the same items from the new aggregates. 29 taken from the Annual Report of the Superin tendent of Banks in New York and for Edge Act corporations from call reports submitted twice a year to the Federal Reserve Bank of New York. Deposits due to thrifts were handled in vari ous ways. For MSBs, demand deposits at weekly reporting (commercial) banks (FR2416 reporters) due to MSBs were available for each Wednesday since May 1961. Quarterly or semiannual data for all commercial banks also were available on call reports since before 1959. These deposits were netted out of Ml. For credit unions, de mand deposits at all commercial banks due to credit unions were estimated to equal 0.03 per cent of total year-end credit union assets for each year through 1974. After 1974, they were taken to equal the "cash” item in the annual reports o f the NCUA. (No adjustment was made for credit union vault cash, also included in this item.) For savings and loan associations (S&Ls), demand deposits at commercial banks before 1973 were assumed to be a constant fraction of the item “cash on hand and in banks” reported annually in condition statements issued by the Federal Home Loan Bank Board (FHLBB); we do not know the value of the fraction used. Begin ning September 1973, semiannual call reports are available in March and September from the FHLBB. DEPOSITS INCLUDED IN THE NON-MI COMPONENT OF M2 Savings D eposits The savings deposit component o f M2 includes deposits at commercial banks, MSBs, S&Ls and credit unions. As usual, construction o f mone tary aggregates requires both gross deposit amounts and, as a netting item, the amounts of deposits held by other money stock issuers. Monthly savings deposit data generally were available beginning in 1968. For prior years, sav ings deposits often were estimated as a constant share of total deposits, the share itself being es timated from data available circa 1968. The fol lowing paragraphs discuss estimates for each type of depositary. For commercial banks from June 1961 through June 1966, total savings deposits were taken from semiannual and quarterly call reports; monthly values were obtained by interpolation. For July 1966 through January 1968, savings deposits at member banks were estimated from monthly summary reports submitted by the Fed eral Reserve Banks (FR422). Beginning January 1968, member banks reported daily savings deposits each week. Monthly nonmember bank data were obtained by interpolation of quarterly call reports.3 The number of data items required as netting items in consolidation is small since commercial banks were not permitted to offer savings accounts to profit-making businesses (including other depositories) prior to November 1975. Thereafter, data regarding savings deposits due to domestic and foreign banks and foreign official institutions were available on Wednes days for weekly reporting banks and for all banks on quarterly call reports since March 1976. (Note that this corresponds to current practices shown in T&ble 2.) We have been unable to clarify precisely which data were used from 1959-67 for MSBs. From 1959-67, total deposits were available on a month-end basis from the National Association of Mutual Savings Banks (NAMSB), but no separate savings deposit series was available. For 1968-71, savings deposits were estimated using total deposit data and a deposit breakdown col lected in a quarterly survey by the FDIC.4 Begin ning in December 1971, month-end savings deposits were published by the NAMSB. Monthaverage data (to correspond to averages of daily data, so far as possible) were constructed by averaging month-end data. 3The discussion in this appendix is somewhat more precise than what we have been able to document. From July 1966 through January 1968, for example, Board staff wrote that “ nonmember bank data were estimated using ratios gener ated from call report data...,” but they do not say precisely how this was done or which ratios were used. The staff memos do note that nonmember bank data continued to be taken from call reports after January 1968, and that monthly values were obtained by interpolation of quarterly call report data. 4Unfortunately, we have been unable to locate a description of the estimation procedure. MARCH/APRIL 1994 30 T\vo netting items were needed for MSBs: sav ings deposits at MSBs due to the U.S. Treasury, and savings deposits held by MSBs at commer cial banks. Both series were available on call reports beginning in March 1976. Different ap proximations were used to generate data for pri or dates. U.S. Treasury deposits were in fact zero for all months prior to November 1974, the first month MSBs were permitted to offer interest-bearing savings deposits to governments. Government deposits were assumed to be $1 million in November 1974 and all intermediate months were obtained by linear interpolation. Similarly, savings accounts held by MSBs at com mercial banks were assumed to be $1 million in November 1975 and intermediate months through March 1976 were obtained by interpo lation. For S&Ls, total deposits for all operating S&Ls from 1959 to June 1968 were obtained from the Federal Savings and Loan Insurance Corporation (FSLIC).5 Beginning in July 1968, month-end sav ings deposits at all federally-insured S&Ls be came available from the FSLIC. For the earlier period (1959 to June 1968), savings deposits were assumed to equal total deposits multiplied by the July 1968 ratio of savings to total deposits. Month-average data were obtained by averaging month-end data. Savings deposits held by S&Ls at other deposi tories, netted out in consolidating M2, were available semiannually beginning in September 1973 from the March-September reporting sys tem release published by the FHLBB (essentially a semiannual call report). Values for prior months were obtained by linear interpolation between an assumed zero in December 1967 and the September 1973 value of $19 million. Credit union shares were obtained on a month-end basis from NCUA.6 Month-average data are constructed by averaging month-end data. Deposits of credit unions at other credit 5Conversations with former FHLBB staff during the course of this research suggest that these data never, in fact, co vered all operating S&Ls. Some data for non-FSLIC institu tions were apparently estimated rather than obtained directly. Other sources report that federally insured S&Ls likely held as much as 95 percent or more of total S&L deposits. Recall that state-insured thrifts in Massachusetts and New York were chartered as MSBs. 6lt isn’t clear whether these data covered all credit unions or only federally insured institutions. Our guess is the lat ter. If so, other credit union deposits would be excluded from the aggregates, perhaps one-half of total credit union deposits. http://fraser.stlouisfed.org/ FEDERAL RESERVE Federal Reserve Bank of St. Louis BANK OF ST. LOUIS unions, netted out in consolidation, are available annually for federal credit unions from the year-end report o f the NCUA beginning in De cember 1968; values for prior years are as sumed to be zero.7 Similar data for state credit unions were estimated by multiplying total as sets at state-chartered credit unions by the ratio of such inter-credit-union shares to total assets at federal credit unions. Small Tim e D ep osits The small time deposit component of M2 in cludes bank and thrift deposits under $100,000 with an original maturity of seven days or more. U.S. Treasury deposits and deposits of thrifts with commercial banks and other thrifts are netted out in consolidation. For commercial banks, small time deposits were computed as a residual by subtracting two series, savings deposits and time deposits of more than $100,000, from reported data on total time and savings deposits. Total time and savings deposits at member banks had been reported weekly since 1959. Small time deposits at nonmember banks were estimated by multiplying small time deposits at small member banks by the ratio o f small time deposits at nonmember banks to small time deposits at small member banks on call report dates.8 Time deposits due to the U.S. Treasury and due to MSBs were netted from the non-Ml component of M2 in consolidation. For weekly reporting member banks, these data were availa ble on Wednesday since 1959 and 1961, re spectively (however, see Banking and Monetary Statistics 1941-1970, chapter 4, for a discussion of changes in items reported). For other banks, semiannual and quarterly call report data were available since before 1959. For MSBs, month-end time deposits beginning in December 1971 were obtained from NAMSB. For prior periods, time deposits were estimated 7Smaller credit unions often hold, as a significant part of their assets, shares in large “corporate central” credit un ions. Although the latter have some retail business, they primarily act as an investor of excess funds deposited with them by other credit unions. 8As in some other cases, this is a somewhat more specific statement of what we believe was done than we have, in fact, been able to locate. 31 by Board staff from data on total deposits at MSBs (available at least from 1959) and from time deposit data collected on quarterly FDIC surveys (available at least since 1966). We have no description of what was done for 1959-65, but it is likely that the 1966 ratio of time deposits to total deposits was simply maintained over this period. (Precisely what was done may be of little importance, since time deposits at MSBs were only 1 percent of total deposits in 1966.) Time deposits of S&Ls at banks are netted from M2 in consolidation. Beginning in Septem ber 1973, time deposits of S&Ls at commercial banks were taken from a semiannual FHLBB publication, referred to in unpublished Board memoranda as the FHLBB March-September reporting system. For all dates prior to Septem ber 1973, it was assumed that S&Ls kept the same proportion of their cash assets in bank time deposits as they had in September 1973. In other words, S&L time deposits at banks from 1959-72 were assumed to be a constant fraction of the amount of "cash on hand and in banks” reported by S&Ls in annual condition state ments to the FHLBB. The value o f that fraction was the ratio o f bank time deposits to cash as sets shown in the September 1973 report in the FHLBB March-September reporting system. Time deposits of credit unions at banks and S&Ls also are netted from M2. Deposits of credit unions at S&Ls (assumed to be time deposits) were reported at year-end by federal credit unions, and were available from the NCUA Annual Report since before 1959. The ra tio of these assets to total assets was used to es timate these items for state-chartered credit unions. Annual reports issued by the NCUA and its predecessor were available since before 1959. Time deposits o f credit unions at commercial banks were estimated at year-end; until 1974, they were treated as a residual, the difference between “cash” reported in the annual reports and estimated demand deposits. After 1974, the cash item excluded time deposits, which were then estimated by applying the ratio o f time deposits to total assets in 1974 to total assets in later years. Year-end cash figures were available since before 1959 for federal credit unions, and since December 1964 for state-chartered credit unions from the annual reports. Large Tim e D ep osits in M 3 The large time deposit component of the monetary aggregate M3 consists of time deposits over $100,000 at all depositories less domestic interbank time deposits and time deposits due to other depositories, foreign commercial banks and foreign governments. The distinction be tween large and small time deposits essentially begins in 1961. Construction of large time deposit data beginning in 1961 is discussed by both Friedman and Schwartz (1970) and Beck (1978). MARCH/APRIL 1994 32 Kenneth Kavajecz The Evolution of the Federal Reserve’s Monetary Aggregates: A Timeline This timeline follows the history of the monetary aggregates published by the staff o f the Fed eral Reserve’s Board of Governors. The chronology is based on the Board’s J.3 and H.6 statistical releases as well as material from the Federal Reserve Bulletin, Money Stock Revisions, and other publications. The timeline includes descriptions of all definition changes and benchmark revisions, the basis on which data were published (monthly, bimonthly, weekly), and the day of the week and time of day that the money stock data were released to the public. The last are o f particular impor tance for financial researchers using high frequency data. Additional miscellaneous items related to the monetary aggregates are included, selected by the author on the basis of their likely im portance to the evolution of the monetary aggregates and/or the role o f monetary aggregates in monetary policy. Note the following in the timeline: • Each page gives information on events that occurred during a single year. • The lines at the top of the pages trace the life of every official monetary aggregate published by the Board staff between 1959 and 1993 (experimental aggregates are excluded). The names of monetary aggregates that were defined and being published during a year are shown in bold face on that page, and the period over which they were being published is shown as a solid line. • Each event of interest is shown as a vertical line with a parallelogram attached. Each event is also dated in the upper left corner of the parallelogram. • Definitional changes are distingished from other events by having a solid vertical line with a three-dimensional rectangle attached. Key: M1A < - ........................• ---------------------- > Undefined Defined i i i i r / Definition Change http://fraser.stlouisfed.org/ FEDERAL RESERVE Federal Reserve Bank of St. Louis BANK OF ST. LOUIS / Other Event 33 1960 , J , F , M , A M, J, J , A , S , O N, D, MI A M1/M1B M1+ M1 - shift adjusted M2 M3 M4 M5 L z N ovem ber 14,1960 The first Federal Reserve statistical release on the money supply was published. The J.3 release entitled Demand Deposits, Currency, and Related Items was thereafter published twice a month. The reported figures were averages o f daily figures rather than the on eday figures reported in the Federal Reserve Bulletin. The money stock was called ’’the money supply.” It measured a concept that would later be called M I A , namely currency plus demand deposits adjusted. The currency component included currency held outside the Treasury, the Federal Reserve, and the vaults o f all commercial banks. The demand deposit component consisted o f demand deposits other than those due to commercial banks and the U.S. Government, less cash items in process o f collection (C IP C ) and Fed eral Reserve float. C IPC items at member and nonmember banks w ere deducted sepa rately from demand deposits at member and nonmember banks, respectively. Since Fed eral Reserve float was not divisible on the basis o f a member-nonmember attribution, it was deducted in whole from the member bank demand deposit component. (See footnote on J.3 release). / 1/ Thursday Release Day of the week released and release time. Immediate Release Bi-monthly basis MARCH/APRIL 1994 34 1961 , J , F , M , A , M , J, J, A, S, O, N, D, MIA < --------------------------------------------------------------------------------------------------------------------------------► M1/M1B M1+ M1 - shift adjusted M2 M3 M4 M5 L http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis Thursday Release Immediate Release Bi-monthly basis 35 1962 , J , F , M . A , M . J , J , A . S , Q , N . P , MI A -------------------------------------------------------------------------- 4 f - ..... ---- ------------- M1/M1B M1 + M1 - shift adjusted M2 M3 M4 M5 L - Septem ber 11,1962 Annual benchmark and seasonal review. Benchmarked to call reports available for 1961. The definition o f M I A was expanded to include demand deposits held by banks located in U.S. territories and possessions at U.S. commercial banks plus foreign demand bal ances at Federal Reserve banks. Foreign demand balances included demand deposits due to foreign governments, central banks and international institutions. (See Federal Reserve Bulletin (F R B ), August 1962). ' .......................... .................... V Thursday R elea se Im m ed ia te R elea se B i-m o n th ly basis MARCH/APRIL 1994 36 1963 J , F , M , A , M , J, J, A, S, O, N, D, MIA -------------------------------------------------------------------------------------------------------------- M1/M1B M1+ M1 - shift adjusted M2 M3 M4 M5 L http://fraser.stlouisfed.org/RESERVE BANK OF ST. LOUIS FEDERAL Federal Reserve Bank of St. Louis Thursday Release Immediate Release Bi-monthly basis 37 1964 . J , F . M , A , M . J, J , A. S, O, N, P MI A M1/M1B M1 + M l - shift adjusted M2 M3 M4 M5 L June 29,1964 Annual benchmark and seasonal review. Benchmarked to call reports available for 1962 and 1963. (See FRB, June 1964). Thursday Release Immediate Release Bi-monthly basis MARCH/APRIL 1994 38 1965 , J F , M , A M, J, J A S O MIA l--------------------------------------------------------------------- M1/M1B M1+ M l - shift adjusted M2 M3 M4 M5 L J u ly 3 0 , 1965 Annual benchmark and seasonal review. Benchmarked to the June and December 1964 call reports. The J.3 release was replaced by the H.6 release, published weekly on Thursday. The H.6 release showed week averages o f daily data on a week ending Wednesday basis. (See FRB, July 1965). Thursday Release Immediate Release Bi-monthly basis http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis Thursday Release ! Immediate Release -L Week ending Wednesday basis N, D 39 1966 , J , F , M , A . M , J , J, A . S, O, N, D, MIA M1/M1B M1+ M l - shift adjusted M2 M3 M4 M5 L June 23, 1966 Effective June 9, 1966, balances accumulated for payment o f personal loans were reclassified for reserve purposes and were excluded from time deposits reported by member banks. Although this did not affect the reported money supply at the time, it did affect the time deposit series reported separately on the H.6. The estimated amount o f such deposits at all commercial banks ($ 1,140 million) was excluded from time deposits adjusted thereafter. (See H.6 release). Setem ber 29,1966 Annual benchmark and seasonal review. Benchmarked to the June and December 1965 call reports. (See FRB, September 1966). Thursday Release Immediate Release Week ending Wednesday basis MARCH/APRIL 1994 40 1967 J , F , M , A , M , J, J, A, S, O, N, D, MI A http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis Thursday Release Immediate Release Week ending Wednesday basis 41 1968 J, F , M , A , M , J, J, A, S, Q, N, D, MI A Thursday Release Immediate Release Week ending Wednesday basis MARCH/APRIL 1994 42 1969 ■ J. F , M , a . m , j , j , a . s , o . n , p , MIA M1/M1B M1+ M l - shift adjusted M2 M3 M4 M5 L August 14,1969 Effective August 6, 1969, the demand deposit component o f the money supply was increased substantially due to a change in accounting procedures associated with bank clearing o f Eurodollar transactions. Previously, an increasing volume o f such transactions had increased C IPC without increasing demand deposits. Since C IPC was deducted from gross demand deposits in computing the money supply, the net demand deposit concept measured in the money supply had been understated by an increasing amount in recent years. A tentative revision was made to correct the downward bias from June 1967 to July 1969. http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis Septem ber 25,1969 Annual benchmark and seasonal review. Benchmarked to the June and December 1968 and June 1969 call reports. (See FRB, October 1969). Thursday Release Immediate Release Week ending Wednesday basis 43 1970 J , F , M , A , M , J, J, A, S, O, N, D MIA M1/M1B M1+ M1 - shift adjusted M2 M3 F ebru ary 1,1970 Mr. Arthur F. Bums replaced Mr. W illiam McChesney Martin, Jr. as Chairman o f the Federal Reserve Board. Chairman Martin had served since April 2, 1951. N ovem b er 27,1970 A nn ual benchm ark and seasonal review . Benchmarked to the December 1969 and June 1970 call reports. The revision this year encompassed for the first time certain new data, mainly from agencies and branches in the U.S. o f foreign banks and from subsidiaries o f U.S. banks organized under the Edge act to engage in international banking business. These new data served to correct a downward bias in the money supply series caused by the generation o f C IPC on the books o f U.S. domestic banks as a result o f clearing a large daily volume o f international transactions. (See FRB, December 1970). Thursday Release Immediate Release Week ending Wednesday basis MARCH/APRIL 1994 44 1971 , J , F , M , A , M , J, J , A, S, O, N, D, MIA* M1*/M1B M1+ M l - shift adjusted M2 M3 M4 M5 L A p ril 22,1971 The Federal Reserve started to publish 3 monetary aggregates. M l , M 2, M3. M l and M2 were reported on a w eekly and monthly basis while M 3 was reported only on a monthly basis due to a lack o f data sources at the time. *M 1 was the same as the previously published money stock, listed above as M I A , only the name had changed. M 2 was a broader aggregate that included M 1 plus commercial banks’ savings deposits, time deposits open account, and time certificates o f deposit other than negotiable CDs issued in denominations o f $100,000 or more by large weekly reporting commercial banks. M 3 was M 2 plus deposits at mutual savings banks and savings and loan associations. FEDERAL RESERVE BANK OF ST. LOUIS N ovem ber 18,1971 Annual benchmark and seasonal review. Benchmarked to the December 1970 and June 1971 call reports. (See FRB, Novem ber 1971). j D ecem ber 9,1971 M oney stock measures have been revised, beginning in September 1971 to reflect the formation o f new banking institutions doing primarily international business. The vague description listed above was taken from a footnote on the H.6 release. To what this refers is subject to some debate. Thursday Release Immediate Release Week ending Wednesday basis 45 1972 , J , F , M , A M , J , J , A , S , O N, D, MI A ^ M1/M1B I M1 + I. . . . M l - shift adjusted ....................................................................... ^ M ................................................................................................................................................................................................................................................................................................................................................................ M2 I M3 -------------- 1------------------------------------------------------------------------ --------M4 M5 I L ► | ........................... I ........................... I Febru ary 24,1972 Benchmark and seasonal review o f M 3 data. Benchmarked to reflect new data fo r deposits at mutual savings banks and savings and loan shares. N ovem b er 24,1972 A change in Regulation J, governing check collection procedures, was implemented on Novem ber 9,1972. Because o f its effects on clearing accounts on bank balance sheets, it had the effect o f raising demand deposits as calculated for inclusion in the money supply. However, to avoid any discontinuities in the series, the resulting increase had been elim i nated from the current series until the annual benchmark and seasonal review. Thursday Release Immediate Release Week ending Wednesday basis MARCH/APRIL 1994 46 1973 J , f , m , a , m , j , j , a , s , o , n , d , MIA M1/M1B http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis Thursday Release Immediate Release Week ending Wednesday basis 47 1974 , J , F . M . A , M . J , J , A . S , Q , N . D MIA M1/M1B M1+ M1 - shift adjusted M2 M3 M4 M5 L J a n u a r y 3 1, 1974 Annual benchmark and seasonal review. Benchmarked to the December 1972 as w ell as the March, June and October 1973 call reports. 1973 was the first year since the early 1960s when call report data appropriate for money supply benchmarks had been available for the spring and fall. (See FRB, February 1974). M a y 23,1974 Benchmark. Benchm arked to the Decem ber 1973 c a ll report. (See H .6 release). ' August 22,1974 Benchmark. Benchmarked to the April 1974 call report. (See H.6 release). N ovem ber 21,1974 Annual benchmark and seasonal review. Benchmarked to the June 1974 call report. (See FRB, December 1974). Thursday Release Immediate Release Week ending Wednesday basis MARCH/APRIL 1994 48 1975 , J , F , M , A , M , J , J , A , S , Q , N , D , MIA ■ ' i ................. i ....................................................... i .......................... ........................ i i i 1 1 1 I I 1 - M1/M1B - i ................. i ....................................................... i ................................................... i i i j M1+ M1 - M2 M3 s h ift a d ju s te d - ----------------- 1.............. ........................ ................ 1 ................................................... I 1 ! 1 1 M4 " M5 W | I ^ 1 i\ 1 1 1 I | 1 ™ ' ^ | - - - - - 1i l 1 L ......................... I ................. I I / February 20, 1975 - _ - - - - - -j - ................. i ....................................................... i ................................................... 1 i ■ / | i i / Benchmark and seasonal review. / / Benchmarked to the October 1974 call report. / (See H .6 release). / i i A p ril 3, 1975 On April 3, 1975, the Federal Reserve published two additional monetary aggregates, M4 and M5. M l and M 2 remained unchanged from their inception in 1971. The definition o f M3 was revised to include credit union shares. M 4 was defined as M2 plus large negotiable time certificates o f deposits issued by large weekly reporting commercial banks. M5 was defined as M3 plus the same large time deposits added to M4. ✓ ............................................................................................................................. is _______I_________________________ _________________________ I_____ F O M C M eeting, A p r il 14-15,1975 Septem ber 18,1975 First target growth cones announced for the Benchmark. monetary aggregates. Benchmarked to the April 1975 call report. (See Anderson and Kavajecz, 1994, Table 4)./ (See H.6 release). http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis M a y 22,1975 Benchmark. Benchmarked to the December 1974 call report. (See H.6 release). Thursday Release Immediate Release Week ending Wednesday basis 49 1976 , J . F . M , A M, J, J , A , S , Q , N , D MI A / O ctob er 21, 1976 / Benchmark. / / Benchmarked to the March 1976 call report. (See H.6 release). Thursday Release Immediate Release Week ending Wednesday basis MARCH/APRIL 1994 50 1977 J . f . m . a . m . j . j . a . s . o . n . p MI A http://fraser.stlouisfed.org/ FEDERAL RESERVE Federal Reserve Bank of St. Louis BANK OF ST. LOUIS June 23,1977 Benchmark. Benchmarked to the December 1976 call report. (See H.6 release). Thursday Release Immediate Release Week ending Wednesday basis 51 1978 J, F , M , A , M , J, J, A, S O D N MI A T' 1 r ■ ■r M1 + M l - shift adjusted T* 1 M2 M3 M4 M5 June 1,1978 February 10,1978 Money Market Time Deposits Data from the Boston District estimatec were authorized by Congress. i Money stock measures for the week o f I I February 1, 1978 subject to larger than normal revisions. I June 22,1978 Benchmark. September 21,1978 Annual benchmark and seasonal review. Benchmarked to the Benchmarked to the March 1978 call report. December 1977 call report. Corrected a recently discovered downward cash (See H.6 release). items bias over the period mid-1975 through September 1978. The bias was created by foreign j March 8,1978 related institutions transferring funds fo r their Mr. G. William M iller replaced parent or subsidiaries. Mr. Arthur F. Burns as Chairman (See H.6 release). o f the Federal Reserve Board. Chairman Bums resigned on January 31, 1978. November 16,1978 On Novem ber 16, 1978, the Federal Reserve published yet another money stock mea sure, M1+. M arch 23,1978 M l, M2, M3, M 4 and M5 remained unchanged from the definitions outlined in 1975. Annual benchmark and seasonal review. M 1 + was defined as the narrow money stock measure, M 1, plus savings deposits at com Benchmarked to the December 1976 mercial banks, N O W accounts at banks and thrift institutions, credit union share drafts, as well as March, June, and and demand deposits at mutual savings banks. September 1977 call reports. nr* (See H.6 release). Thursday Release Immediate Release Week ending Wednesday basis MARCH/APRIL 1994 52 1979 J F . M A M . J . J . A S . O N D, MI A M 1/M1B ^ 1 M1 + M l - shift adjusted M2 \ M3 1 ■|...................r -------------- 1------- 1 ' i i i i i i i i 1 1 1 1 1 1 1 1 i i i i i i i i - i - ................ i-.................. i------- 1i i i i i i i i I I | i 1 ! i I ! I | i ■ \ M4 ^ I I I ! i ^ ^ ^ ! f- ^ M5 L I i /F ebru ary 8,1979 / / Annual benchmark and seasonal review. / / Benchmarked to the June 1978 call report. / / (See H.6 release). / i ■ i i i i i -|...................r .................. i------- 1 i i i i i i / August 6,1979 / j / Mr. Paul A . Volcker replaced / / Mr. G. W illiam M iller as Chairman / / o f the Federal Reserve Board. / 1 1 f M a y 24,1979 Benchmark. 1 O ctob er 6,1979 On Saturday October 6, 1979, Chairman Volcker Benchmarked to the September 1978 call report. called a special meeting o f the F O M C where he (See H.6 release). announced the Federal Reserve would switch to http://fraser.stlouisfed.org/ FEDERAL St. Louis Federal Reserve Bank of RESERVE BANK OF ST. LOUIS a nonborrowed reserve operation procedure. The m ove placed a greater emphasis on the M l aggregate due to its close relation to the outstanding supply o f reserves. N ovem ber 8,1979 The money supply figures published on Novem ber 8, 1979 for the weeks ending October 3, 10, 17, and 24th incorporated minor corrections made to the data due to an understatement o f the deposits provided by Manufacturers Hanover Trust Company in the last four weeks. The Federal Reserve had begun an inquiry, with the help o f outside counsel, to provide assurance that recent errors in the money supply data were inadvertent and that no individual or institution obtained improper advantage from the preparation, revision and release o f these figures. Thursday Release Immediate Release Week ending Wednesday basis 53 1980 M A ■ M , J , J O N D MIA M1/M1B «*- M1+ «*- M1 - shift adjusted M2 M3 M4 •<- M5 ■<- January 10,1980 June 20, 1980 N ovem ber 7,1980 Benchmark. Benchmark. The money supply figures that would normally be Benchmarked to the December 1978 Benchmarked to the June and published on Novem ber 14, 1980 may be delayed fo r / and March 1979 call reports. September 1979 call reports. a time in view o f changes in the flows o f data (See H.6 release). (See H.6 release). required by the Monetary Control A ct o f 1980. The next H.6 release went out on Novem ber 18th. Febru ary 8,1980 On February 8, 1980, the Federal Reserve radically reorganized how the monetary aggregates were defined. M l was renamed M I A without changing its definition. M 1B was defined to be M I A plus N O W and automatic transfer service (A T S ) accounts at banks and thrift institutions, credit union share draft accounts and demand deposits at mutual savings banks. M 2 was redefined to be M 1B plus overnight (and continuing contract) repurchase agreements (R P ) that are issued by commercial banks to the non-bank public, overnight Eurodollars issued by Caribbean branches o f member banks to U.S. non-bank customers, money market mutual fund shares, savings deposits and small time accounts (those issued in denominations less than $100,000) at commercial banks and thrift institutions. N ote that M 2 w ill differ from the sum o f its components by a consolidation adjustment made to avoid double-counting the public’ s monetary assets, namely, the amount o f demand deposits held by thrift institutions at commercial banks. M 3 was defined to be M 2 plus large time deposits (those issued in denominations o f $100,000 or more, net o f the holdings o f domestic banks, thrift institutions, the U.S. government, money market mutual funds, and foreign banks and official institutions), and term RPs at commercial banks and thrift institutions, net o f term RPs held by money market mutual funds. A new aggregate, L , was created and defined to be M 3 plus the non-bank public’s holdings o f U.S. savings bonds, short-term Treasury securities, commercial paper and bankers acceptances (which excludes money market mutual fund holdings o f these assets). In addition, two addenda were included on the H.6 release, overnight RPs at commercial banks plus overnight Eurodollars and money market mutual fund shares. Feb ru ary 15,1980 Seasonal factors for the newly defined aggregates were released on the H.6. (See FRB, February 1980). Friday Release Immediate Release Week ending Wednesday basis MARCH/APRIL 1994 54 1981 , J , M IA F , M . A , M . J, J. A, S, O, N, D. M1/M1B M1+ ! ..................................... I..................... M1B - shift adjusted -► M2 -► M3 -► M4 M5 L __ L. January 16 and 23,1981 M a y 22,1981 The H.6 emphasized caution when Another monetary aggregate, called M lB -sh ift adjusted, was introduced. It was defined interpreting the monetary aggregates to be M 1B less shifts to O C D from non-demand deposit sources. because o f the introduction o f N O W A ll the definitions o f the other monetary aggregates remained unchanged. accounts on a nationwide basis with heavy promotional efforts. January 23,1981 Benchmark. June 26,1981 Benchmarked to the December 1979 Benchmark. and March 1980 call reports. Benchmarked to the September and December 1980 call reports. This incorporated all the changes due to The definition o f the narrowest measure o f the money stock, M 1, was revised to include the implementation o f the Monetary non-bank travelers checks. Control Act. A ll the definitions o f the other monetary aggregates remained unchanged. (See H.6 release). M a y 1,1981 Annual seasonal review. Adjustment o f the monetary aggregates to include the effects o f N O W accounts. I (See H.6 release).___________________ M a rch 13,1981 Septem ber 18,1981 The H.6 cautioned the interprepation o f the aggregate measures The term R P component o f M 3 was revised due to the shifting o f demand deposits and savings deposits into and benchmarked to a survey o f ’’retail RPs” other checkable deposits (O C D ) accounts. Estimates o f the shifts conducted on August 31, 1981. The current obtained from various depository institution samples suggested methods o f estimation did not pick up the increase that in January and February, 75 to 80% o f the increase in which was attributed to recent active promotion. excess o f ’’trend” came from demand deposits and the other 20 (See H.6 release).________________________________ to 25% came from savings deposits and other sources. (See H.6 release)._________________________________________________ http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis Friday Release Immediate Release Week ending Wednesday basis 55 1982 J . F M A M J . J A S . O N . D , MI A Friday Release Friday Release @ 4:10 PM EDT_____________________________________________________________________________________*4:15 PM EDT Week ending Wednesday basis (W.E.W.) * (W.E.W.) MARCH/APRIL 1994 56 1983 , J , F , M , A M, J , J , A , S , O N, D, MI A M1/M1B M1+ L M1B - shift adjusted M2 -► M3 -► M4 M5 L ' January 28,1983 The Gam-St. Germain A ct o f 1982 had recently authorized money market deposit ac counts. Beginning on January 28, 1983, M M D A s were reported separately as a component o f the broader monetary aggregates. Due to the lack o f historical data, they were reported on a not seasonally adjusted basis. Note that this did N O T revise the monetary aggregates because the deposits had previously been included in the savings / component o f M2. I F eb ru a ry 14, 1983 A nn ual benchm ark and seasonal re view . Benchmarked to the December 1981 and March, June, and September 1982 call reports. T w o definitional changes have been implemented. M 2 was revised to include general purpose/broker dealer (GP/BD) tax-exempt money market mutual funds and exclude all IRA/Keogh balances at depository institutions and money market mutual funds. O ctob er 1,1983 The D ID C m oved to amend Regulation Q by eliminating interest rate ceilings on time deposits with maturities greater than 31 days and principal greater than $2,500. (See FRB, Novem ber 1983, Table 1.16). M 3 was revised to include institution-only (I/O) tax-exempt money market mutual funds. (See H.6 release). M a y 20th through June 10th 1983 / W eekly data on savings deposits and small time deposits were not reported due to report ing difficulties associated with M M D A s. In addition, historical data were revised to re flect corrections o f reporting errors beginning in December 1982. j (See H.6 release dated June 10, 1983).______________________________________________j http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis Friday Release @4:15 PM EDT Week ending Wednesday basis 57 1984 , J , F , M , A , M , J , J , A , S , Q , N , D , M IA M 1/M1B M1+ . . . . . . . . . . . . .I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. .. .. .. .. .. .. .. .. .. .. .. . . . . . . . . . . . . . . . . . . . . . . . . . . . -. -. -. -. .- .- .-. - r - - i i i i _ _ _ _ _ L . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ............................................ i - - - - - - - - - - - - - _ ......... M1B - shift adjusted ............. ................................... r .................. i 1 M2 M3 ^ • M4 ■ W W 1 £ ........................... ..................................... h .................. i i ..................................................................... r .................. i ---------------------------------------------------------- j----------------- ► M5 L -----------February 16,1984 Annual benchmark and seasonal review. Benchmarked to recent call reports. The H.6 published on February 10, 1984 was the last one that presents deposits data on a week-ending Wednesday basis. A ll data shown on the H.6 dated February 16, 1984, was shown on a week-ending Monday basis to correspond with the new reporting cycle under contemporaneous reserve requirements (C R R ). In addition, M 3 was redefined to include term Eurodollars in Canada and the United Kingdom that are held by U.S. residents. The rest o f the aggregates remained unchanged definitionally. (See H.6 release). M arch 22, 1984 The H.6 began being released at 4:30 PM E D T on Thursdays. / November 1,1984 Benchmark. Benchmark due to revised data received in conjunction with annual shifts among weekly, quarterly and annual reporting panels. Similar benchmarks were not needed in later years because o f improvements in the procedure used to handle the panel shifts at the Federal Reserve. In addition, institution-only money market mutual fund shares y were revised back to Novem ber 1980 to reflect new data. Friday Release Thursday Thursday Release @ 4:15 PM EDT| 4:15 PM j @ 4:30 PM EDT W.E.W. I Week ending Monday (W .E.M .) I W .E.M. MARCH/APRIL 1994 58 1985 , J . F . M , A M, J, J , A , S , O N, D MI A http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis / N ovem ber 21,1985 / The Bank o f N ew York experienced a computer j / failure that resulted in a substantial transitory increase in reported demand deposits. Thursday Release @ 4:30 PM EDT Week ending Monday 59 1986 , J , F , M , A M, J, J A S O N, D, MI A M1/M1B i M1 + M1B - shift adjusted M2 -► M3 ► M4 M5 L February 13,1986 Annual benchmark and seasonal review. Benchmarked to call reports through June 1985. (See H.6 release). August 21,1986 January 1,1986 Estimates o f M 2 and M3 were revised upward, Regualtion Q was further revised by reflecting new data for RPs obtained from regular the D ID C by abolishing interest rate quarterly and annual surveys for the end o f June. ceilings on both N O W accounts and (See H.6 release). time deposits with maturities less than 31 days. (See FRB, March 1986, Table 1.16). O ctob er 22,1986 Demand deposits increased sharply during the next A two months follow in g passage o f the Tax Reform Act. , A p ril 1,1986 The D ID C alleviated the interest rate ceilings on savings deposits. (See FRB, June 1986. Table 1.16). .......- . V Thursday Release @ 4:30 PM EDT Week ending Monday MARCH/APRIL 1994 60 1987 , J, F , M , A , M , J, J, A, S, O, N, D, MI A http://fraser.stlouisfed.org/ FEDERAL St. Louis Federal Reserve Bank ofRESERVE BANK OF ST. LOUIS (O c to b e r 19,1987 The D ow Jones Industrial Average plummeted 500 points sending other major stock exchanges into a significant decline as well. The effect on the monetary aggregates was to boost liquid components due to the increased volume o f transactions. Thursday Release @ 4:30 PM EDT Week ending Monday 61 1988 M M O N D MIA \11/M 1B M1 + M1B - shift adjusted M2 ► M3 ► M4 M5 + l February 18, 1988 Annual benchmark and seasonal review. Benchmarked to call reports through June 1987. Beginning on February 18, 1988, the H.6 included weekly estimates of M2 and M3 sea sonally adjusted and seasonally unadjusted on the same publication schedule as M 1. M 1 was redefined to make the treatment of thrift institutions identical with that of com mercial banks in the construction of the monetary aggregates. Under the new definitions, all vault cash held by thrift institutions was excluded from the currency component of M l, and all demand deposits and OCDs held by thrifts were excluded from the demand deposit and OCD components, respectively. Previously, only a portion of the vault cash and transactions deposits held by thrifts were excluded at the M 1 level— representing the estimated amount held to service their OCD liabilities— while the remainder was sub tracted at the M2 level. In addition to the redefinitions noted above, ATS accounts at credit unions— like those at commercial banks and all other thrift institutions— were now included in the OCD com ponent of M l, rather than in the savings deposit component of M2. The monetary aggregates M2 , M3 and L had no change in their definitions. (See H.6 release). y i i / March 10,1988 / Weekly seasonal factors for the nontransactions component of M2 / beginning with the week of March 28, 1988 were revised to / incorporate further analysis of certain holiday-related effects. / (See H.6 release). / Thursday Release ______________________________________________________________________________________@ 4:30 PM EDT Week ending Monday MARCH/APRIL 1994 62 1989 . J , f . m , a , m , j , j , a , s . o , n . d , MIA M l/M lB ir M ----------------- 1 ----------------------------------------------------------------------------------------------------------------------► M1 + M1B - shift adjusted RESERVE BANK OF ST. LOUIS FEDERAL Thursday Release @ 4:30 PM EDT Week ending Monday 63 1990 MIA M1/M1B M1 + M1B - shift adjusted M2 -► M3 -► M4 M5 L February IS, 1990 Annual benchmark and seasonal review. Benchmarked to call reports through June 1989. M2 was revised to include overnight RPs issued by thrift institutions, formerly included with term RPs in the non-M 2 component of M3. This redefinition had no effect on the levels of M l , M3 or L. (See H.6 release). ............................................................................................. 1 ' ^ Thursday Release @ 4:30 PM EDT Week ending Monday MARCH/APRIL 1994 64 1991 I __ J . I f , m _ ___ _______L , a , m . j , j . a , s , o , n , d , MIA M1/M1B I---------- M1+ M1B - shift adjusted M2 l-------- M3 ► l-------- M4 M5 L February 7,1991 Annual benchmark and seasonal review. Benchmarked to call reports through June 1990. (See H.6 release). FEDERAL RESERVE BANK OF ST. LOUIS I October 3,1991 There was a change in the format of the H.6 release. The change is necessary because, on September 17, 1991, depository institutions began reporting to the Federal Reserve only their combined savings deposits and MMDAs, rather than reporting them separately, owing to changes in the deposits reports (FR2900). In order to calculate consistent seasonally adjusted data, the new seasonal factors are equal to the inverse of the weighted average of the inverses of the seasonal factors for savings deposits and MMDAs, where the weights are defined as the ratio of each component to the sum of the components during the month of August. In other words, the total of savings and MMDAs was split into its two components, ’savings’ and ’MMDAs’ for both commercial banks and thrifts. Then its old seasonal factors (published in February 1991) continued to be used, namely, the seasonal factors for bank savings, bank MMDAs, thrift savings, and thrift MMDAs. (See H.6 release). Thursday Release @ 4:30 PM EDT Week ending Monday 65 1992 I i J i MIA F A i M , A , M , J , J , A, S, O, N, D, .............................................................................................. .................. M1/M1BI ^ I w M1 + _________ I. . L _____________________________________________________________________ M1B - shift adjusted M2 1 M3 ^ 1 .................. I . . M5 M4 . .................... L . . ^ L ! 1 1 1 1 1 February 13,1992 / Annual benchmark and seasonal review. Benchmarked to call reports through September 1991. (See H.6 release).________________________________ I March 5, 1992 I The release dated March 5, 1992 incorporates further / revisions to historical data. The change was due to the / reclassification of some brokered deposits from large time / to small time deposits in addition to those reported in the / annual benchmark on February 13, 1992. / (See H.6 release). Thursday Release @ 4:30 PM EDT Week ending Monday MARCH/APRIL 1994 66 1993 , J , F , M . A , M , J , J . A , S . O , N . D . MIA M1/M1B I" M ----------------- 1 ---------------------------------------------------------------------------------------------------------------------- ► M1+ .................. I ......................................................................................................................................... M1B - shift adjusted FEDERAL RESERVE BANK OF ST. LOUIS Thursday Release @ 4:30 PM EDT Week ending Monday 67 Charles W. Calom iris Charles \N. Calomiris is an associate professor a t the Universi ty o f Illinois, Urbana-Champaign, a nd is a co-director for the university’s Office for Banking Research. He is also a faculty research fellow, N ational Bureau o f Econom ic Research, and a visiting scholar at the Federal Reserve B ank o f St. Louis. Com m entary M. N THEIR PAPER, Anderson and Kavajecz pro vide the rare public service of a careful exami nation of the construction of m onetary data. The paper is im portant because the data on m onetary aggregates are central to academic and policy research in macroeconomics. I expect that by the m etric of the percentage of m one tary economists who will have this paper in their file cabinets 10 years from now, this will be one of the most successful works in m one tary economics. Data are forever. Like the Feder al Reserve Board’s Banking and Monetary Statistics and All Bank Statistics, Friedman and Schwartz’s Monetary Statistics o f the United States, Capie and Wood’s Monetary Statistics o f the United Kingdom, and Eisner’s How Real is the Federal Deficit?, Anderson and Kavajecz pose and answ er descriptive questions of lasting interest to macroeconomists. Judging from the paucity of this kind of work, its im portance seems to be underestim ated. I found little to quibble w ith in the way the authors organized their description. In my dis cussion I will focus on the question of why researchers doing empirical m onetary economics should care about the details of how m onetary aggregates are m easured and how those m eas urem ents have changed over time. Five sets of issues seem central in motivating the potential usefulness of this exercise. First, and most obviously any attem pt to construct m onetary aggregates for long stretches of time m ust do its best to ensure comparability of measures. This m eans coming to grips not only w ith financial innovations that affect the range of definitions of money, but also with changes in sampling procedures, seasonal adjustment, and other choices made by the data con structors. Second, the Fed’s procedure of revising data retrospectively to m aintain consistent definitions and seasonal adjustm ent factors—which some times have produced large retrospective revi sions of the aggregates—makes it difficult to compare empirical research of different vin tages. For example, two studies of Ml money de m and over the same period, perform ed at different dates, may differ not only because of specification, b ut because of the vintage of data used in each. It would be worthwhile to ask how m uch of the differences across studies of money dem and can be attributed to retro spective revision of data, as opposed to the incorporation of additional periods of data, or differences in specification. Third, there should be an objective outside evaluation of the Federal Reserve Board's choices of definitions of money and methods of seasonal and benchm ark adjustment. Prior to this study, this was not feasible because relatively little was known about the Fed’s procedures. Anderson and Kavajecz suggest that the Fed’s decisions MARCH/APRIL 1994 68 regarding w hat to include in the m onetary aggregates are often influenced by w hether adding a new com ponent helps to stabilize the relationship betw een money and economic activity. While this procedure may make sense, in general, given the Fed's desire to use m one tary aggregates as targets, it would be interest ing to describe clearly how the Fed decides (and how it should decide) that the improvement in stability will persist (that is, it reflects a lasting behavioral change rath er than a tem porary statistical coincidence). How long should the Fed wait before incorporating new (apparently stabilizing) elements of money into its defini tions of aggregates? Would an increased em pha sis on Divisia indices be w arranted in light of the difficulties posed by having to make an all-or-nothing decision about w hether to include financial assets in one or m ore of the aggregates? Regarding seasonal adjustment, it would be in teresting to consider how the Fed should deter mine w hen a change in seasonal factors has occurred—and how far back retrospective changes in seasonality should be made. W hat is the optimal choice of the period over which de term inistic seasonals should be estimated? How much less relevant is distant information for estimating seasonal com pared to recent inform a tion? Anderson and Kavajecz have provided researchers interested in these questions with a wealth of detail that will allow them to con struct counterfactual rules for defining m one tary aggregates, and to compare these with those adopted by the Fed. Fourth, if the Fed attem pts to keep m onetary aggregates "on track” relative to economic activi ty (by altering definitions and adjustm ent fac tors), then this makes the reported aggregates unsuitable for perform ing hypothesis tests about the stability of money demand. Researchers in terested in w hether money dem and is stable, therefore, should perform sensitivity analysis to examine w hether reasonable counterfactual defi nitions of the aggregates lead to different con clusions about the stability of money demand. Finally, there is a problem I will label the "ex pectations erro r effect.” The essence of this problem is that (unforecastable) errors in m eas urem ent, which affect expectations of agents in "real time,” may weaken the apparent connec tion betw een money and output using ex post (corrected) m onetary data. Assume for the sake of argum ent (unrealistically) that the current retrospective data on the m onetary aggregates FEDERAL RESERVE BANK OF ST. LOUIS are "correct.” That is, assume that all definitions, seasonal adjustm ents and revisions that have been made so far are perfectly accurate, and that no fu rth e r revisions will be made in the series. Fur therm ore, assume th at we can agree upon an econometric procedure for m easuring the close ness of the relationship betw een money and output (using, for example, a "structural VAR” model of money, output and other variables). Even under these ideal circumstances, the m easured relationship betw een "true” money (measured ex post) and economic activity will be biased toward zero if money is initially m easured w ith error. The reason is that "true” money, as well as errors in m easuring aggregate money, will elicit responses that affect economic activity and money subsequently. Money and output are linked through “fundam ental” struc tural links and through "expectational” effects. For example, an increase in an individual’s hold ings of money may lead him to rebalance his portfolio (putting pressure on interest rates to fall and output to rise in the standard IS-LM model). Second, estimates of m onetary aggregates (which include initial m easurem ent error) will also be taken into account by the public in economic activity if aggregates are used as economic indicators. In the absence of m easurem ent error, the in dividual agent can observe w ithout e rro r not only his own m oney b u t also the aggregate. In the presence of m easurem ent error, the ag gregate is observed w ith error, and these errors will elicit real responses from agents. So, both announced and tru e money will be linked to output. Neither will be as strongly linked to out put as tru e money in the absence of m easure m ent error, and empirical analysis using ex post data (after removing errors) may underestim ate the link betw een money and output. Thus, tem porary inaccuracies in m onetary ag gregate estimation (which elicit real responses) might explain weak correlations betw een money and output using ex post (accurate) data. How can one come to grips w ith this problem em piri cally to decide w hether bias arising from "expec tations erro r effects” is im portant for conclusions about the role of money in the economy? One simple first step is to compare various m easures of m onetary aggregates (and m onetary grow th rates) for a given period reported at different dates. If the differences among these m easures are small, then the problem of potential bias is of little practical importance. If the differences 69 Figure 1 M2 Against HM2 Billions of dollars 1981 82 Quarterly data 83 84 85 86 are large, then one would have to take on the m uch harder job of m easuring the extent of the bias by gauging the reaction of the public to m easurem ent errors. As Anderson and Kavajecz point out, there are several types of potential error, and each involves a different correction horizon. First-published num bers (which appear one or two weeks after the fact) are updated w ithin a m onth or so be cause of the arrival of new data. They are changed (roughly) annually to adjust for changes in benchm arks and seasonals, and change with new definitions of the aggregates as well. As an illustrative exercise, I chose the easiest case—M2 from January 1981 to January 1993. I chose this period because, as Anderson and Kavajecz show, there was no im portant change 87 88 89 90 91 92 1993 in the definition of M2 during this period, so th at one can focus on the role of revisions from new data, benchm ark changes, and changes in seasonal factors as sources of error. I construct ed m easures for this period using three differ ent timings of m easurem ent. I used the first date of publication of m onthly M2 in the H.6 statistical release as my definition of the initial m easure of M2. This was released roughly two weeks after the end of the m onth. My second date of m easurem ent is the M2 figure reported in the Federal Reserve Bulletin, which appears with a two-month lag. My third m easure is the retrospective series as of January 1993.1 Using seasonally adjusted data from these sources for January, April, July and October, I constructed m easures of the level of M2 and of 1Table 3 in Anderson and Kavajecz decomposes revisions in money into three different adjustments and expresses them in absolute terms. This is interesting for some purposes, but not for my purpose. I am interested in whether errors coming from all sources are potentially large relative to the actual number. MARCH/APRIL 1994 70 Figure 2 DM2 Against DHM2 Quarterly data 1981 82 83 84 85 86 87 88 89 90 91 92 1993 Figure 3 DM2 Against DOM2 Quarterly data 1981 82 83 FEDERAL RESERVE BANK OF ST. LOUIS 84 85 86 87 88 89 90 91 92 1993 71 that q u arter’s grow th in M2 m easured at the time the M2 num ber was reported. For example, M2 grow th for the first q uarter of 1982 accord ing to the H.6 release is the log difference be tw een the first H.6 num ber for M2 in April and the January 1982 num ber reported in that same release. Figures 1-3 compare these definitions of money and money growth using these three m easurem ent horizons. The level and growth data from the 1993 series are labeled M2 and DM2; the data from H.6 are labeled HM2 and DHM2; and the data from the Bulletin are la beled OM2 and DOM2. conclusions would be reached for M l in the 1980s, or for these and other aggregates during other periods, rem ains an open question. As An derson and Kavajecz note, Depository Institution Deregulation and M onetary Control Act (DIDMCA) improved the accuracy of m onetary statis tics in the 1980s, and M2 tends to be a sm oother series than M l. Thus, my results may understate the im portance of m easurem ent erro r for other aggregates and earlier periods. These figures indicate that revision of M2 has been trivial in the 1980s, and so I conclude that for these series over this period, "expectations e rro r effects” were not important. To the extent revisions did matter, long-term retrospective changes (the difference betw een M2 and 02, or DM2 and DOM2) are m ore im portant than those occurring w ithin two m onths of initial publi cation. Board of Governors of the Federal Reserve System. All-Bank Statistics 1986-1955. Board of Governors of the Federal Reserve System, 1959. One conclusion to draw from these findings is that, if there has been a breakdown in the rela tionship betw een M2 and output during the past decade, it cannot be attributed to tem porary m ism easurem ent of money. W hether similar REFERENCES ________ . Banking a nd M onetary Statistics 1914-1941. Board of Governors of the Federal Reserve System, 1976. ________ . Banking a nd M onetary Statistics 1941-1970. Board of Governors of the Federal Reserve System, 1976. ________ . Federal Reserve Bulletin. ________ . M oney Stock, L iq u id Assets, a n d D ebt Measures, supplement to the Federal Reserve statistical release, H.6. Capie, Forrest, and Alan Webber. M onetary H istory o f the United Kingdom, 1870-1982. Allen & Unwin, 1985. Eisner, Robert. How Real is the Federal Deficit? Free Press, 1986. Friedman, Milton, and Anna J. Schwartz. M onetary Statistics o f the United States. Columbia University Press, 1970. MARCH/APRIL 1994 73 K. Alec Chrystal an d R on ald M acDonald K. Alec Chrystal is N ational Westminster Bank professor of econom ics at City University o f London. Ronald M acD onald is professor o f econom ics at the University o f Strathclyde, Glasgow. E m pirical E viden ce on the R ecen t B eh avior an d U sefulness o f Sim ple-Sum a n d W eighted M easures o f the M oney S tock "We m ust have a good definition o f Money, For if we do not, then what have we got, But a Quantity Theory o f no-one-knows-what...” Boulding (1969, p. 555) .M.HE FEDERAL RESERVE BANK of St. Louis has been, for the last th ree decades or so, at the center of an approach to macroeconomic policy which became universally known as “M onetarism.” Indeed, the very term entered the public domain through an article in the Federal Reserve Bank of St. Louis’ Review by Karl Brunner in 1968. The central tenet of m onetarism was that there is a stable dem and function for something called "money.” Policy advice came down to recom m ending th at the m onetary authorities should deliver a steady rate of the grow th of m oney within some target range. The 1970s w ere a good time for monetarists. Velocity in the United States appeared to be on a stable trend, and the adoption of floating exchange rates generated a need for independent m easures of m onetary stance in most of the in dustrial countries. M onetary targeting was widely adopted and m onetarism became a worldwide credo. Since the end of the 1970s, however, life has been m uch h a rd e r for m onetarists. The stability of empirical m onetary relationships became m uch m ore difficult to maintain, and governm ent after governm ent has given up even the notional attem pt to target m onetary aggregates. The allegedly m onetarist governm ent of M argaret Thatcher abandoned m onetary targets in the United Kingdom in 1985. The Chairman of the Federal Reserve Board has recently announced that the Fed has ceased to m onitor M2 and, instead, will be using the real interest rate as an indicator of m onetary stance. Only the Bundesbank appears to be retaining any faith in the significance of m onetary aggregates, though they have been widely criticized for so doing. (Norbert W alter, the chief econo- MARCH/APRIL 1994 74 mist of Deutsche Bank, has, for example, been quoted as saying that "...M3, the broad money supply indicator targeted by the central bank, was obviously distorted and devalued as an indi cator.” Financial Times, August 10, 1993, p. 2). The standard explanation for w hy previously stable m onetary relationships have broken down is financial innovation. In particular, liberalization and competition in banking have generated shifts in dem and betw een compo nents of money which have underm ined earlier empirical regularities. Interest payments on transaction deposits have made it m ore difficult to distinguish money held for transaction from money held for savings. Robert Rasche (1993) in his paper to the St. Louis Fed conference 12 m onths ago identi fied the beginning of the 1980s as a time of a critical regime change. This structural change, he claimed, had destroyed the validity of the traditional St. Louis reduced-form methodology as a m eans for explaining and forecasting the course of GNP. Policy m akers around the w orld have clearly also been convinced that m onetary aggregates provide little useful inform ation to guide m acro policy. Presumably, nobody would argue that no guide to m onetary policy was necessary. How ever, the advocates of a simplistic policy based upon any traditional m easure of money as the sole guide are disappearing rapidly. At the theoretical level, the significance of ex ogenous m onetary shocks as a cause of business cycles has been under threat from the so-called Real Business Cycle school. For them, m onetary disturbances are not the trigger to cycles but, rath er, are an endogenous response to shocks em anating in the real economy. While this ap proach does not necessarily eliminate the validity of countercyclical m onetary policy, it certainly reduces the significance of the traditional m one tarist line that m onetary shocks are the prim ary trigger to the cycle. Several recent empirical studies have apparently produced evidence to support the contention that money does not have any explanatory pow er—at least for real economic activity. (De Long and Summers, 1988; Friedm an and Kuttner, 1992, 1993.) The consensus view emerging from all of this appears to be th at trying to target and control money is no longer a very sensible thing for policy m akers to do. M onetary policy is now mainly about setting short-term interest rates, FEDERAL RESERVE BANK OF ST. LOUIS despite all the well-known difficulties that choosing the “correct” interest rate entails (Friedman, 1959). This paper follows an alternative line of reasoning, for which th ere is an overwhelming theoretical case. T here has been a m ajor m eas urem ent e rro r in virtually all of the previous literature on money. Instability in empirical rela tionships has been prim arily due to the fact that simple-sum m easures of money are not admissi ble aggregates on index-theoretic grounds. This e rro r has been especially im portant in a period w hen characteristics of com ponents w hich are added together have been changing. We do not claim that correction of this m eas urem ent e rro r salvages entirely the role of money as a m acroeconomic indicator (though such may still be the case). Rather, ou r prim ary focus is to see w hether acceptable indexes of money outperform traditional money m easures in conventional tests. As is often the case in applied studies, the evidence tu rn s out to be mixed but leaning in favor of the superiority of weighted over simple-sum aggregates. Before presenting our own empirical evidence, we shall first review briefly the evolution of the concept of m oney and then the case for an ap propriately constructed index. W h at Is M o n ey ? The definition of money has not been static over time. The first identifiable m easure of money was undoubtedly the stock of the physi cal commodity which served as currency— typically precious metal. At some point, certain ly by the 18th century in England, it was clear that bank notes had become a major elem ent of the money stock so th at a m onetarist at that time would have had to extend the definition of money to include notes plus specie in the hands of the public. By the 19th century, financial in novation had moved things a stage fu rth e r and the relevant concept of money had expanded yet again to include bank deposits, which could be used on dem and and could be tran sferred by w riting a check. In recent times, the issue has been: Which of the other highly liquid assets held by the public should be included? The Radcliffe view in the United Kingdom and the view of Gurley and Shaw in the United States was that the bound ary betw een money and other liquid assets was impossible to draw because so many close sub stitutes w ere available. This contention was 75 countered successfully for a while by the evi dence that elasticities of substitution w ere relatively small (Chetty 1969) and also by the evidence that predictions of m onetarist ap proaches w ere fairly robust to m inor definitional changes. In other words, the general message of the evidence was not so different if one used M l or M2, or even M3. Such a defense would be m uch harder to m aintain today than it was 15 years ago. The introduction of interest paym ents on checking accounts in the United States led to a major reversal of the velocity tren d —at least for M l in about 1980. In the United Kingdom, abolition of quantitative ceilings on bank interm ediation, also in 1980, led to a period of rapidly rising broad money coinciding w ith very slow narrow money grow th. The innovations which followed w ere clearly associated w ith big movements of deposits from non-interest bearing to interestbearing accounts. In such circumstance, neither narrow nor broad money proved to be reliable indicators—at least in the short term . It would be a mistake to believe that the composition changes of the 1980s w ere a new phenom enon. In Volume I of A Treatise on Money, Keynes argued that an unchanged quan tity of m oney could conceal im portant changes in circulation as holders tran sferred money be tw een cash and savings deposits, and betw een income and business accounts. In Volume II, he reported the statistical finding that the p ropor tion of deposit (savings) accounts to total ac counts had risen in Britain from 38 percent to 46 percent betw een 1920 and 1926. According to Keynes, "... The continual transference from cu rren t to deposit accounts ... [acted as] a con cealed m easure of deflation...” (Keynes, 1930b, p. 10) sufficient to explain a drop in the price level of 20 percent over the period. There is nothing rem arkable about the fact that these composition changes have been noticed before. W hat is rem arkable is that so m any economists w ere happy to ignore them for so long in the post-W orld W ar II period. Partly, this was because the regulatory regime in most countries (interest ceilings and/or quan titative controls on intermediation) limited for some time the significance of the interface be tw een checking and savings accounts, as well as the significance of nonbank competitors. Fisher, 1989, Chapter 1, for a survey). At the risk of oversimplifying, it is sufficient for present purposes to note that the traditional reason for regarding money as critically different from other assets is that it has a direct role in tra n s actions and, hence, has a direct role in the tra d ing activity of a m arket economy. According to the Quantity Theory, the money stock will de term ine the general level of prices (at least in the long term) and, according to m onetarists, it will influence real activity in the short run. For this reason, empirical m easures of the money stock have tried to identify as compo nents of money those instrum ents which can be used directly in transactions. The problem of our time is that a whole range of types of deposits which can be spent, m ore or less, directly also yield an interest rate and could, thus, be chosen as a form of savings as well. From a m icro-demand perspective, it is hard to justify adding together assets which have different and varying yields (Barnett, Fisher and Serletis, 1992). It has long been know n that only things that are perfect substitutes can be com bined as one commodity. T here is ample evi dence that the assets which are commonly combined in money m easures are not in fact perfect substitutes. From a micro-foundations perspective this leaves only tw o alternatives. The first is to res trict attention to a very narrow definition of money, w hich only needed non-interest bearing components. The alternative is to construct an index num ber of "m onetary services” which could, in principle, capture the transactions services yielded by a wide range of financial as sets in a superlative way (Diewert 1976, 1978). Two potential index num bers are the Divisia index proposed by Barnett (1980) and the Cur rency Equivalent (CE) index proposed by Botemberg (1991) at the St. Louis conference in 1989. M o n e y M e a su re m e n t The attraction of both of these m onetary serv ices indicators is that they internalize the substi tution effects betw een components of a potential aggregate and, thus, solve the problem of com position changes w hich was discussed above. They do not in themselves guarantee the weak separability of any chosen aggregate, but they do approxim ate optimal aggregator functions for those collections of aggregates which have been found "admissible” on separability grounds (Belongia and Chalfant, 1989). A substantial am ount of literature discusses the concept of money and its m easurem ent (see The theoretical case for weighted m onetary aggregates is overwhelm ing—at least to anyone MARCH/APRIL 1994 76 w ith a training in microeconomics and/or index num ber theory. The only objection could be on the grounds that it does not make an im prove m ent over flawed simple-sum aggregates in practice. There has been a significant accum ula tion of evidence, however, to suggest that Divisia aggregates outperform their simple-sum equiva lents. For example, Barnett (1980) showed that some apparent shifts in money dem and in the United States w ere rem oved w hen Divisia m eas ures replaced simple sum. Barnett and Spindt (1979) showed the informational superiority of Divisia over simple-sum measures. Belongia and Chalfant (1989) find Divisia MIA to have superior informational content to other admissible ag gregates. Barnett, O ffenbacher and Spindt (1984) also find evidence for the superiority of Divisia. F urther support is provided by Serletis (1988). Lindsey and Spindt (1986) is one of the few papers which have looked at this comparison to come out against Divisia, though Fisher and Ser letis (1989) is inconclusive. Belongia (1993) has recently discovered that using weighted, as opposed to simple-sum, m onetary aggregates alters significantly the con clusions th at should have been reached by several recent influential studies. These studies have, on the whole, adduced evidence that money is not a "cause” of cycles in real activity. Hence, this suggests th at the problem s w ith tests of money in the economy in recent years may be m ore due to bad m easurem ent theory rath e r th an to an instability in the link betw een the tru e m oney and the economy. Rather than a problem associated w ith the Lucas Critique, it could instead be a problem stemming from the "Barnett Critique.” The idea of weighted m onetary aggregates has spread outside the United States. Studies include Horne and M artin (1989) for Australia; Cockerline and M urray (1981) and Hostland, Poloz and Storer (1987) for Canada; Ishida (1984) for Japan; Yue and Fluri (1991) for Switzerland; and Belongia and Chrystal (1991) and Drake and Chrystal (forthcoming) for the United Kingdom. A recent Bank of England study in the United Kingdom context concludes: "A Divisia m easure of money appears to have some leading indica tor properties for predicting both nominal out put and inflation...a case can clearly be made for including Divisia in the range of indicators analyzed by the authorities w hen form ing their judgments on m onetary conditions.” (Fisher, Hudson and Pradhan, 1993, p. 63). FEDERAL RESERVE BANK OF ST. LOUIS A variation on the traditional “closed economy” tests is provided by Chrystal and MacDonald (1993). They point out that exchange rate models have been just as dependent upon money m easures as have dem and for money studies or reduced form tests of m onetary policy. It is no coincidence th at exchange rate equations started to misbehave at the same time as velocity trends appeared to shift (in the early 1980s). By replacing simple-sum aggregates in an exchange rate model by Divisia aggregates, for the dollarpound rate, they show th at a simple, flexible, price m onetary model can be salvaged as a long-run proposition. They also find that, w hen Divisia m easures are used, the short-run fore casting perform ance is far superior on out-ofsample tests. We now tu rn to some empirical results of our own. The results we shall present fall into two distinct sections. In the first section, we rep o rt comparisons of simple-sum and weighted m eas ures of the money supply in the context of St. Louis Equations. The dependent variable is ac cordingly nominal GNP. We are aw are of the problems encountered in the past w ith such m ethods (Rasche, 1993). However, it is a simple, familiar and well-known context w ithin which to com pare money m easures. We are not con cerned w ith the absolute validity of the results b u t only w ith the relative perform ance of different m easures. Non-nested testing tech niques w ere used to distinguish betw een vari ous indicators of money. In the second section, we use the m ore so phisticated m odern time-series methodology to test for the existence of short-run and long-run causal links betw een money and real activity. It is this latter question which has dom inated the recent literature. We add to this literature both by including alternative money m easures and by providing international comparisons. EMPIRICAL RESULTS WITH ST. LOUIS EQUATIONS In this section, we rep o rt results of com pari sons betw een traditional simple-sum aggregates, Divisia m easures and the Rotemberg Currency Equivalent (CE) m easure. We use the environ m ent of a modified St. Louis Equation as a vehi cle for these comparisons, and we use non nested testing m ethods to identify superior inform a tional content. We are well aw are w ith all the 77 difficulties associated w ith the St. Louis Equa tion methodology. If we cannot use this struc tu re at a St. Louis Fed conference, however, w here else can we? More seriously, this m ethod offers simplicity and transparency. It does at least give us a feel for the properties of the data we are dealing with. A methodology m ore acceptable to the econometric purist will be reported in the following section. None of the data we used was original to this study. The bulk of it was made available to us by Michael Belongia at the Federal Reserve Bank of St. Louis, though the U.K. Divisia series (post-1977) was constructed by the Bank of En gland (Fisher, Hudson and Pradhan, 1993). It should be noted that the time period of the study differs for each country, depending upon data availability. Data definitions also vary from country to country, b u t space does not perm it an extensive discussion of such differences. Seasonally adjusted data w ere used in all cases. The dependent variable is taken as the first difference of the log of nominal GDP or GNP. The first difference of the log of nominal governm ent spending (federal in the United States case) on goods and services is used as a fiscal variable in all cases. A w orld trade varia ble was tested as an external dem and variable but was not found to add explanatory pow er in the countries tested. Also tested was an interest rate variable. This was found to be im portant in this context only for the United States. Hence, the U.S. Equation includes the first difference of the Treasury bill rate. The original St. Louis Equation contained lags of order 0-3. On quarterly data, most economists would expect to use at least 0-4, so, given the short data series for some countries, this is the standard lag length we adopted. In parallel to the simple St. Louis Equation form at, we also rep o rt tests in a version of the equation which includes the lagged dependent variable, lagged 1-4 periods. Additionally in this latter context we rep o rt an F test on the exclu sion of money from the equation entirely. This provides useful information, not only about the relative inform ational content of different money m easures b u t also about w hether money m at ters at all. In some cases Divisia money m atters but simple-sum m oney does not. The reverse is never true. The basic test is to use the same equation in one case w ith simple-sum money and in another case Divisia or CE money. Three test statistics are reported for comparisons betw een the two form ulations—the Davidson and MacKinnon J-test, the Fisher and McAleer JA-test and the Akaike Inform ation Criterion (AIC). O ther tests have been m onitored, including the NT test of Pesaran and Godfrey and the Wald-type test. These other tests differ in detail but they do not alter the overall picture produced. Accord ingly, they are not reported here. We refer to the J-test and the JA-test as being inconclusive w hen both form ulations reject each other and indeterm inate if neither rejects the other. The results are reported in Tables 1 to 7 for the United States, the United Kingdom, A ustra lia, Germany, Switzerland, Canada and Japan, respectively. Let us consider each. U n ited S ta te s The U.S. results are sum m arized in Table 1. Simple-sum aggregates M l and MIA in general dominate their Divisia equivalents. From M2 on w ards to broader aggregates, however, the domination is reversed. This is clear for M2 and M3, though the difference betw een Divisia L and simple-sum L is probably not significant. This general picture is not altered by the inclu sion or exclusion of the lagged dependent variable. From the F-tests it is clear th at simple-sum MIA, Divisia MIA and Divisia M l do not add significant explanatory pow er to the equation at norm al significance levels. However, Divisia M2 has the greatest informational content of all the aggregates tested, though it is only marginally m ore significant than simple-sum M2. The CE aggregate holds its ow n against M l and MIA, though never establishing statistically significant domination in either direction. It loses out to the broader simple-sum aggregates, however, and also to the broader-based Divisia m easures (the latter result is implied but not shown). Overall, the M2 level of money aggregation seems superior, though the Divisia aggregate at this level does not dominate its simple-sum equivalent sufficiently to make an overwhelming case for preferring one to the other. U n ited K in g d o m The U.K. results appear in Table 2. There are far few er aggregates to choose from in the U.K. case. The Bank of England even stopped report- MARCH/APRIL 1994 78 Table 1 St. Louis Equations for the United States: Simple-Sum vs. Weighted Money_________________________________________ Dependent variable: first difference of the natural log of nominal GNP. Independent control variables: first difference of the natural log of federal spending on goods and services; first difference of the T-bill rate; the current period value and four lags of each variable are included as regessors. Part 1: no lagged dependent variable in regression M1 vs. Divisia M1 Akaike Information Criterion (AlC) J-test JA-test favors M1 favors M1 favors M1 (4.42) (-.1 ; 3.06) (-.7 2 ; 2.26) M1 vs. Rotemberg Currency Equivalent (CE) AlC J-test JA-test favors CE inconclusive indeterminate (-2 4 ) (2.74; 2.64) (1.56; 1.39) M1A vs. Divisia M1A AlC J-test JA-test favors M1A favors M1A favors M1A (3-78) (-.5 9 ; 2.7) (-.67; 2.6) M1A vs. CE AlC J-test JA-test favors CE inconclusive indeterminate (-.5 5 ) (2.7; 2.4) (1.66; 1.16) M2 vs. Divisia M2 AlC J-test JA-test favors Divisia M2 favors Divisia M2 favors Divisia M2 (-1.1) (2.42;1.8) (2.04;1.4) M2 vs. CE AlC J-test JA-test favors M2 inconclusive favors M2 (8.75) (2.4; 4.9) (.92; 3.6) M3 vs. Divisia M3 AlC J-test JA-test favors Divisia M3 favors Divisia M3 favors Divisia M3 (-1.76) (2.4; 1.5) (1.9; 1.2) M3 vs. CE AlC J-test JA-test favors M3 inconclusive inconclusive (6.45) (3.07; 4.7) (2.05; 2.72) L vs. Divisia L AlC J-test JA-test favors Divisia L inconclusive inconclusive (-.3 1 ) (2.35; 2.18) (1.6; 1.34) L vs. CE AlC J-test JA-test favors L inconclusive inconclusive (7.27) (2.9; 4.8) (2.2; 3.98) Note: The Akaike Information Criterion is an adjusted difference between two values of the likeli hood function. It indicates the direction of informational advantage but has no critical bounds. The J and JA tests are f-statistics for the rejection of one model over the other and then the reverse. "Inconclusive" = both significant; “ Indeterminate” = neither significant. Data period is 60:1-92:4. FEDERAL http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis 79 Table 1 (continued) Part 2: four lags of the dependent variable included M1 vs. Divisia M1 AIC J-test JA-test favors M1 favors M1 indeterminate (3.74) (.27; 2.8) (-.4 5 ; 1.78) M1 vs. CE AIC J-test JA-test favors M1 inconclusive indeterminate (.46) (2.49; 2.59) (1.6; 1.15) M1A vs. Divisia M1A AIC J-test JA-test favors M1A favors M1A favors M1A (3.2) (-.4 2 ; 2.44) (-.5 5 ; 2.3) M1A vs. CE AIC J-test JA-test favors CE inconclusive indeterminate (-.8 4 ) (2.7; 2.3) (1.4;.62) M2 vs. Divisia M2 AIC J-test JA-test favors Divisia M2 inconclusive indeterminate (-.7 4 ) (2.4; 2.1) (1.87; 1.56) M2 vs. CE AIC J-test JA-test favors M2 inconclusive favors M2 (5.76) (2.4; 4.3) (.62; 3.4) M3 vs. Divisia M3 AIC J-test JA-test favors Divisia M3 favors Divisia M3 indeterminate (-1 .5 ) (2.24; 1.5) (1.63; 1.14) M3 vs. CE AIC J-test JA-test favors M3 inconclusive favors M3 (3.4) (3.05; 3.99) (1.32; 2.26) L vs. Divisia L AIC J-test JA-test favors Divisia L inconclusive indeterminate (-.4 2 ) (2.38; 2.21) (1.42; 1.29) L vs. CE AIC J-test JA-test favors L inconclusive favors L (3.93) (2.9; 4.0) (1.85; 3.4) MARCH/APRIL 1994 80 Table 1 (continued) Part 3: F-tests on exclusion of money from St. Louis Equation probability M1 M1A M2 M3 L Divisia Divisia Divisia Divisia Divisia CE F(5,107) M1 M1A M2 M3 L ” ” ” ” = = = = = = = = = = = 2.36 1.88 4.43 3.49 3.69 1.01 0.73 4.73 4.09 3.86 2.19 [0.045] [0.103] [0.001] [0.006] [0.004] [0.418] [0.600] [0.001] [0.002] [0.003] [0.060] Note: Exclusion test conducted in equation including lagged dependent variable shown in Part 2 of the table. This is equivalent to the concept of Granger causality tests, but includes contem poraneous observations on independent variables. ing M l and M3 in 1989 because it considered the data too distorted by financial innovation. Hence, the only choice using official statistics is betw een MO (the m onetary base) and M4. The results show a clear domination of Divisia M4 over simple-sum M4 both w ith and w ithout the presence of lagged GDP. The non nested tests, however, make it impossible to choose betw een Divisia M4 and MO. Also, while the Akaike In form ation Criterion favors MO over simple-sum M4, the J-test and the JA-test are inconclusive and indeterm inate, respectively. On the other hand, the F-test gives informational advantage to MO, w ith Divisia M4 running second. Simplesum M4 has no significant explanatory pow er at norm al probability levels. This suggests that Divisia M4 should replace simple-sum M4 as an indicator of the course of broad money in the United Kingdom. A u stra lia Results for Australia appear in Table 3. They show comparisons betw een M2, M3 and their Divisia equivalents. The inform ation criterion is always in favor of Divisia, and the significant J-tests favor Divisia. More dram atic perhaps are the F-tests which show that neither simple-sum aggregate m atters at anything close to norm al probability levels, while both Divisia aggregates do have significant inform ational content. This is probably the clearest case available w hich il lustrates the domination of Divisia over simple FEDERAL http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis sum aggregates—especially for broad money measures. G e rm a n y Table 4 contains the results for Germany. The inform ation criterion generally favors Divisia measures over their simple sum counterparts, with the exception of M3 in the absence of the lagged dependent variable. All the J- and JA-tests are indeterm inate w ith the exception of the J-test which shows dominance of Divisia M2 over M2 (in the presence of the lagged dependent varia ble). The same test for M3 is very close to ac cepting Divisia M3 as dominating M3. The overwhelm ing impression of the German results, however, is th at conveyed by the F-tests, which show the very low informational content of all money measures. In this respect, only Divisia M3 is significant at even the 10 percent level and the simple-sum aggregates do not obvi ously m atter at all. This is a surprising result for a country which has a reputation for sound m onetary policy and adheres to a simple-sum M3 target. It is possible that the very success of m onetary policy is responsible for a low varia tion of nominal income grow th, which makes it hard to establish statistical relationships. However, it is also possible th at Divisia money m easures do a better job in tracking nominal GDP than their simple-sum equivalents. 81 Table 2 St. Louis Equations for the United Kingdom: Simple-Sum vs. Weighted Money Dependent variable: first difference of the natural log of nominal GDP. Independent control variables: first difference of the natual log of government spending on goods and services. Part 1: no lagged dependent variable included in regression Divisia M4 vs. M4 AIC J-test JA-test favors Divisia M4 favors Divisia M4 favors Divisia M4 (3.136) (1.05; 3.1) (-.4 5 ; 2.2) Divisia M4 vs. MO AIC J-test JA-test favors Divisia M4 inconclusive indeterminate (.168) (2.77; 2.72) (1.52; .96) M4 vs. MO AIC J-test JA-test favors MO inconclusive indeterminate (-3 .2 ) (3.5; 2.5) (■49; .77) Part 2: four lags o f dep end ent variable included Divisia M4 vs. M4 AIC J-test JA-test favors Divisia M4 favors Divisia M4 indeterminate (2.32) (1.3; 2.7) (-.1 6 ; 1.82) Divisia M4 vs. MO AIC J-test JA-test favors MO inconclusive indeterminate (-.7 8 ) (3.17; 2.8) (15; .73) M4 vs. MO AIC J-test JA-test favors MO inconclusive indeterminate (-3 .6 9 ) (3.8; 2.7) (-.1 8 ; -.3 6 ) Part 3: F-tests on exclusion o f money from St. Louis Equation probability M4 Divisia M4 MO F(5, 78) ” = 1.45 [0.215] = 2.33 [0.050] = 2.83 [0.021] Note: Test is done in equation from Part 2 including lagged dependent variable. Data period is 1968:3-1992:4. S w itz e r la n d In Switzerland, (Table 5) Divisia aggregates dominate on inform ation grounds, though the JA-test is always indeterm inate and the J-test only gives clear dominance to Divisia on one oc casion (Divisia M2 beats M2 in the presence of lagged GDP). The F-tests suggest that M l and Divisia M l are very similar in informational con tent (with a tiny advantage to Divisia). Simple sum M2, by contrast, is overwhelmingly domi nated by its Divisia counterpart. This confirms the simple (and obvious) conclusion from other countries that Divisia clearly dominates w hen it comes to broad m oney measures, but at the narrow money level it does not make m uch difference. This is clearly almost a tautology w hen narrow aggregates have minimal interestbearing components. MARCH/APRIL 1994 82 Table 3 St. Louis Equations for Australia: Simple-Sum vs. Weighted Money Dependent variable: first difference of the natural log of nominal GDP. Independent control variable: first difference of the natural log of government spending. Part 1: no lagged dependent variable included in regression M2 vs. Divisia M2 AIC J-test JA-test favors Divisia M2 inconclusive indeterminate (-3 .7 4 ) (3.63; 2.67) (-.7 ; -.1 9 ) M3 vs. Divisia M3 AIC J-test JA-test favors Divisia M3 favors Divisia M3 indeterminate (-3 .9 ) (3.2; .9) (1.1; -.8 2 ) Part 2: four lags of dependent variable included M3 vs. Divisia M3 AIC J-test JA-test favors Divisia M3 indeterminate (-5 .6 ) (3.5; 1.1) I (-6 .6 ) (4.4; 3.3) (-1 .5 ; -.0 7 ) C O favors Divisia M2 inconclusive indeterminate (J) M2 vs. Divisia M2 AIC J-test JA-test Part 3: F-tests on exclusion of m oney from St. Louis Equation probability M2 M3 Divisia M2 Divisia M3 F(5,43) ” = = = = 0.99 0.82 3.46 2.84 [0.430] [0.540] [0.010] [0.027] Note: Data period is 1974:2-1989:4. C an ada Japan The Canadian results (Table 6) confirm the general p attern established above. M l has a m arginal edge over Divisia M l (in the presence of lagged GNP, but not otherwise) but for b roader aggregates the Divisia m easure domi nates w here any discrimination is possible. Divisia L sweeps the board w ith its simple-sum equivalent and both Divisia M2 and Divisia M3 exhibit obvious domination. The F-tests confirm this general story. Simple-sum M l has the greatest inform ational content, b ut it is closely followed by Divisia M l. Simple-sum M2, M3 and L do not have significant informational content at the 5 percent level, though all of their Divisia equivalents do so. Japan does not fit in at all w ith the p attern of all the other countries in the sample (Table 7). On the basis of the Akaike Inform ation Criteri on, all of the simple-sum aggregates marginally dom inate their Divisia counterparts. However, none of the J- or JA-tests are able to discriminate and the F-tests make it clear than none of the money m easures has any explanatory pow er at all. In this context it makes no sense to try to distinguish betw een sets of num bers, none of which m atter. http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS FEDERAL Federal Reserve Bank of St. Louis Japan’s m onetary aggregates differ from m any others at the M2 and M3 level, because they in clude negotiable CDs. However, this would not 83 Table 4 St. Louis Equations for Germany: Simpie-Sum vs. Weighted Money Dependent variable: first difference of the natural log of nominal GDP. Independent control variable: first difference of the natural log of government spending. Part 1: no lagged dependent variable included in regression M2 vs. Divisia M2 AIC J -te s t J A -test favors Divisia M2 indeterminate indeterminate (-.8 5 ) (1.84; 1.54) (.25; .93) M3 vs. Divisia M3 AIC J -te s t J A -test favors M3 indeterminate indeterminate (1.3) (.57; 1.73) (-.0 6 ; .92) Part 2: four lags of dependent variable included M 2 vs. Divisia M2 AIC J -te s t JA -test favors Divisia M2 favors Divisia M2 indeterminate (-1 .8 ) (2.9; 1.89) (-.3 6 ; -.014) M3 vs. Divisia M3 AIC J -te s t JA -test favors Divisia M3 indeterminate indeterminate (-1.17) (1.88; 1.12) (.91; .24) Part 3: F-tests on exclusion of m oney from St. Louis Equation probability M2 M3 Divisia M2 Divisia M3 F(5,41) ” ” = = = = 0.71 1.66 1.77 2.08 [0.620] [0.167] [0.140] [0.087] Note: Data period is 1975:1-1990:1. explain the poor perform ance of M l. Also, this should give an advantage to Divisia M2 and Divisia M3 which is not supported by the data. Either the Japanese economy behaves very differently from the others studied o r there are serious data errors underlying this evidence. We now tu rn to tests of the causal links be tw een money and real activity using m odern time-series methods. TIME SERIES TESTS OF MONEY/ REAL ACTIVITY CAUSALITY In this part of the paper, we consider Granger causality tests for a selection of Divisia and simple-sum money aggregates for each of the countries referred to in our St. Louis tests. The causality tests are based on vectors consisting of real GDP, the GDP deflator, a Treasury bill rate and the relevant m easure of the money supply. Defining our causality vectors in this way facili tates separate modelling of the effect th at differ ent m onetary impulses may have (particularly in the short run) on the real and price compo nents of GDP. The Treasury bill rate is also in cluded in the vector because of the well-known spurious effect m oney can have on output if an interest rate effect is excluded (Sims, 1980). Our causality tests have a num ber of other features, some of w hich are novel to this paper. First, for reasons w hich are now widely ac cepted, it is extrem ely im portant that the varia bles entering the causality vector should be stationary and that any indication of cointegratability should be determ ined (see, Engle and Granger, 1987; MacDonald and Kearney, 1987). The latter aspect of the time-series properties of MARCH/APRIL 1994 84 Table 5 St. Louis Equations for Switzerland: Simpie-Sum vs. Weighted Money Dependent variable: first difference of the natural log of nominal GDP. Independent control variable: first difference of the natural log of government spending. Part 1: no lagged dependent variable included in regression M1 vs. Divisia M1 AIC J-test JA-test favors Divisia M1 indeterminate indeterminate (-.0 6 4 ) (1.37; -1.31) (1.35; -1.3 3 ) M2 vs. Divisia M2 AIC J-test JA-test favors Divisia M2 inconclusive indeterminate (-1 .8 4 ) (2.99; 2.76) ( -.6 ; .76) Part 2: four lags of dependent variable included M1 vs. Divisia M1 AIC J-test JA-test favors Divisia M1 indeterminate indeterminate (-0 5 ) (.93; -.87 ) (.91; -.8 8 ) M2 vs. Divisia M2 AIC J-test JA-test favors Divisia M2 favors Divisia M2 indeterminate (-4 .0 2 6 ) (3.1; 1.5) (.35; -.2 7 ) Part 3: F-tests on exclusion of m oney From St. Louis Equation probability M1 M2 Divisia M1 Divisia M2 F(5,39) ” ” = 2.9 [0.026] = 0.41 [0.840] = 2.92 [0.025] = 2.11 [0.085] Note: Data period is 1975:2-1989:4. the vector is im portant, since if there is one (or more) cointegrating relationships among the vec tor, then it is inappropriate to test for causality among a vector of first-differenced variables, because the Granger representation theorem asserts th at such a vector will be misspecified; it will exclude im portant "long-run” information contained in the levels of the variables. (This was a point recognized by Friedm an and Kuttner, 1992, in th eir causality tests on U.S. data [see their footnote 19], b ut they did not include such long-run elements in their testing framework.) A second im portant aspect of any causality test is that it should be robust to non-norm al errors. Holmes and Hutton (1992) suggest handling this issue using a non-param etric rank F-test (instead of the standard F-test used in conventional causality studies). In this paper, we argue that since most departures from normality arise http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS FEDERAL Federal Reserve Bank of St. Louis from heteroskedasticity, this issue may be dealt w ith using the Hansen-White non-param etric correction for heteroskedasticity. The general class of causality tests employed in this section of the paper have come in for some criticism in the literature (see: Zellner, 1979, 1988; Basmann, 1963; and Cooley and LeRoy, 1985). In particular, it is argued th at to be interpreted as indicating causality from, say, money to output, Granger-type causality tests have to be em bedded in a structural setting and appropriate identifying restrictions imposed (see Holmes and Hutton, 1992, for a partial rebuttal). However, given our purpose is not to examine causality for a single m easure of money, but rath e r to determ ine which m easures from a range of simple-sum and Divisia money magni tudes have the greatest informational content, 85 Table 6 St. Louis Equations for Canada: Simple-Sum vs. Weighted Money Dependent variable: first difference of the natural log of nominal GNP. Independent control variable: first difference of the natural log of government spending. Part 1: no lagged dependent variable included in regression Divisia M1 vs. M1 AIC J-test JA-test favors Divisia M1 indeterminate indeterminate (.34) (.45; .78) (.09; .46) Divisia M2 vs. M2 AIC J-test JA-test favors Divisia M2 favors Divisia M2 favors Divisia M2 (6.0) (1.7; 4.8) (-.8 4 ; 2.5) Divisia M3 vs. M3 AIC J-test JA-test favors Divisia M3 favors Divisia M3 favors Divisia M3 (3.67) (1.29; 3.5) (-.3 9 ; 2.4) Divisia L vs. L AIC J-test JA-test favors Divisia L favors Divisia L favors Divisia L (8.43) (-.4 3 ; 4.3) (-.9 9 ; 3.72) Part 2: four lags of dependent variable included Divisia M1 vs. M1 AIC J-test JA-test favors M1 indeterminate indeterminate (-.1 9 ) (.81; .56) (.44; .25) Divisia M2 vs. M2 AIC J-test JA-test favors Divisia M2 favors Divisia M2 indeterminate (3.34) (1.26; 3.3) ( -.8 ; 1.04) Divisia M3 vs. M3 AIC J-test JA-test favors Divisia M3 favors Divisia M3 indeterminate (2.87) (.55; 2.71) (-.7 4 ; 1.53) Divisia L vs. L AIC J-test JA-test favors Divisia L favors Divisia L favors Divisia L (3.49) (.49; 2.8) (-.4 9 ; 2.17) Part 3: F-tests on exclusion of m oney from St. Louis Equation probability M1 M2 M3 L Divisia Divisia Divisia Divisia F(5,55) M1 M2 M3 L ” ” ” ” = = = = = = = = 7.05 2.03 1.37 1.25 6.95 3.34 2.43 2.53 [0.000] [0.088] [0.250] [0.299] [0.000] [0.010] [0.047] [0.039] Note: Data period is 1968:3-1987:1. MARCH/APRIL 1994 86 Table 7 St. Louis Equations for Japan: Simple-Sum vs. Weighted Money Dependent variable: first difference of the natural log of nominal GNP. Independent control variable: first difference of the natural log of nominal government spending. Part 1: no lagged dependent variable included in regression M1 vs. Divisia M1 AlC J-test JA-test favors M1 indeterminate indeterminate (.32) (-1.57; 1.86) (-1 .7 ; 1.72) M2 vs. Divisia M2 AlC J-test JA-test favors M2 indeterminate indeterminate (■5) ( - 7 ; 1.45) (-1.09; 1.05) M3 vs. Divisia M3 AlC J-test JA-test favors M3 indeterminate indeterminate (.72) (-.6 2 ; 1.53) (-1.01; 1.17) Part 2: four lags of dependent variable included M1 vs. Divisia M1 AlC J-test JA-test favors M1 inconclusive inconclusive (.503) (-1 .9 6 ; 2.23) (-2 .0 3 ; 2.16) M2 vs. Divisia M2 AlC J-test JA-test favors M2 indeterminate indeterminate (.44) (-.4 5 ; 1.49) (-1.19; .72) M3 vs. Divisia M3 AlC J-test JA-test favors M3 indeterminate indeterminate (.77) (-.6 4 ; 1.56) (-1 .0 8 ; 1.15) Part 3: F-tests for exclusion of m oney from St. Louis Equation probability M1 M2 M3 Divisia M1 Divisia M2 Divisia M3 F(5,42) ” ” ” = = = = = = .79 [0.559] .24 [0.942] .38 [0.861] .63 [0.675] .11 [0.990] .14 [0.981] Note: Data period is 1976:1-1991:2. we do not believe that the standard criticisms of our fram ew ork have as m uch im port as they may have for m ore conventional studies. We also take encouragem ent from the fact that even in recent papers which only address the causality properties of a single money m easure (see, for example, Friedm an and Kuttner, 1993), Granger-type tests have still been employed (although, we would argue, incorrectly since such tests only involve a vector of differenced variables). FEDERAL RESERVE BANK OF ST. LOUIS U nit R o o t A n d M u ltiv a rite C o in te g rtio n R e s u lts We begin the empirical analyses of this sec tion by testing for unit roots in the variables entering ou r causality vector. Although the coin tegration m ethod we employ below is due to Johansen and is, therefore, a m ultivariate test for the num ber of unit roots in a given vector, we nevertheless thought it w orthw hile to exa mine some simple univariate unit root rests for 87 motivational purposes, and also to guide us in the appropriate ord er of differencing for the variables entering the cointegrating tests. There have been, in fact, a variety of pro posed m ethods for implementing univariate unit roots tests (for example, Dickey and Fuller, 1979; Phillips and Perron, 1988; Stock, 1990; and Park and Choi, 1988) and each has been used in the applied macroeconomics literature. Since, however, th ere is now a growing consen sus that the earliest, unit root test—due to Dickey and Fuller (1979)—has superior small sample properties com pared to its com petitors (see Campbell and Perron, 1991, for a discussion), we employ it. In particular, we estimate the fol lowing regression equation for the series en ter ing our causality vector: (1) A*, = ^ + fir + 7r^(1 + S YA!<;-i + u,> w here * is the variable of interest, /x and t denote determ inistic regressors (a constant and a time trend, respectively). Equation 1 represents a reparam eterization of an auto regression of in levels, w here the length of the autoregression is set to ensure that ut is serially uncorrelated. In this context, a test for a unit root in the series ^ am ounts to a f-test of 7r = 0 (that is, the sum of the autoregressive param eters in the levels autoregression is unity). The alternative hypothesis of stationarity re quires that 7r be significantly negative. Since u n der the null hypothesis of non-stationarity the calculated t-ratio will not have a student’s t-distribution, critical values calculated by Fuller (1976) m ust be used instead. In estimating equation 1 for so many coun try/variable combinations, we initially used a common lag length, q, of 4 for all variables. However, given the sensitivity of Augmented Dickey-Fuller (ADF) tests to the chosen lag length we also experim ented w ith shorter lag lengths in instances w here the estimated f-ratio on 7 was close to its critical value, this being r particularly so w hen a variable appeared to be 1(2). (In particular, and following the recom m en dation of Hall, forthcoming, and Campbell and Perron, 1991, we sequentially deleted insignifi cant lags of the dependent variable until we a r rived at a parsimonious relationship which satisfied the non-autocorrelation criterion.) In the reported tables that follow, a shorter lag length than 4 is denoted by the num ber in parenthesis after the variable mnemonic. W here the default value of 4 is reported for the ADF statistic, it m eans that either all four lags are sig nificant or, in instances w here some lags are in significant, reducing the lag length from 4 would not have m ade a qualitative difference to the interpretation. In Tables 8 through 14, we present our estimates of the f-ratio for the estimated coeffi cient 7 in equation 1 for the levels and first and r second differences of each series in question. This procedure facilitates a test for one and two unit roots, respectively. The f-ratio has been calculated w ith the time tren d included in the regression equation, as in equation 1 (referred to as fT ), and the tren d excluded (referred to as t ). This follows the sequential testing strategy recom m ended by, for example, Perron (1988): If a de term inistic com ponent is excluded from a unit root test b u t such a com ponent features in the data generation process (DGP) of the series, the resulting test will have low pow er. However, if the deterministic com ponent is absent from the DGP, greater pow er may be obtained by esti mating p w ithout the trend component. In our unit root tests, all variables, apart from the in terest rate series, have been transform ed into natural logarithms. In order to capture any rem aining seasonality in the variables, three seasonal dummies have been incorporated into our estimated version of equation 1. * A num ber of findings em erge from Tables 8 to 14. First, th ere are only two variables which appear to be stationary around a deterministic trend, namely the Australian Treasury bill rate and the Rotemberg C urrency Equivalent m easure —all the other variables appear to contain stochastic trends. As is common in many other studies of the time series properties of macroeconomic series, the level of the price deflator for a num ber of countries appears to be an 1(2) process; that is, inflation in these countries is an 1(1) process. Interestingly, it is also the case that some of ou r m onetary series appear to be 1(2) processes. In general we found that this result (but not the result for the deflator) was particu larly sensitive to the lag length specified in the estim ated equation. For example, in the case of the United States, all of the simple-sum money m easures appeared to be 1(2) w hen q was set equal to 4 (DM1A and DM1 also appeared 1(2) with four lags). However, in these instances it appeared that this lag length resulted in an overparam eterized regres sion equation and the deletion of a single lag MARCH/APRIL 1994 88 Table 8 Unit Root Tests for the United States L t SSM1 DM1 SSM1A DM1A SSM2 DM2 SSM 3 (1) DM3 SSL DL GDP DEF TB RCE BCE A *T — 1.61 -1.97 0.03 0.77 1.40 0.80 1.66 1.17 1.40 0.95 -1.7 3 -1.01 -2.31 -1.31 0.67 -2 .0 8 -2 .1 6 -2 .7 0 -2 .0 5 -0 .2 3 -1.78 0.44 -1 .2 7 -0.61 -1.6 6 -2 .7 3 -1 .9 4 -2 .0 3 -3 .7 0 -2 .5 9 A2 t tT -3 .8 0 -3.61 -3.61 -4 .1 3 -3 .2 0 -4.1 9 -2 .9 3 -3 .9 3 -1.7 3 -3 .6 8 -4.51 -1.74 -5 .4 9 -4 .7 3 -3 .5 8 -4 .3 5 -4 .3 4 -3.61 -4.21 -3.4 7 -4 .2 2 -3 .3 6 -4 .0 6 -2.01 -3.7 4 -4.71 -1 .5 9 -5 .6 4 -4 .7 6 -3 .6 8 t /• *T -9.1 8 -8 .3 8 -9 .2 9 -9 .5 2 -8.3 7 -8.21 -9.1 4 -7.79 -7.05 -7.81 -7.69 -6 .5 2 -11.84 -5.1 6 -6 .3 8 -9 .1 6 -8 .3 5 -9 .2 9 -9 .4 9 -8.3 7 -8.1 9 -9.1 4 -7.79 -7.12 -7.8 0 -7.6 6 -6.1 0 -11.78 -5.1 4 -6 .3 6 Note: Unless otherwise noted, each ADF statistic was computed with a lag of 3. SS denotes a simple-sum monetary aggregate; D denotes a Divisia aggregate; M denotes money; L denotes liquidity; GDP denotes real Gross Domestic Product; DEF denotes the GDP deflator; and TB denotes a Treasury bill rate. L, A and A2 denote, respectively, the level and first and second differ ence of a variable, t^ and tT are augmented Dickey Fuller statistics with allowance for a constant mean and for a trend in mean, respectively. The 5 percent critical values for t and t are -2 .8 9 and -3.4 3 , respectively (Fuller, 1976). Table 9 Unit Root Tests for the United Kingdoim L t SSM4 (2) DM 4 (1) GDP DEF TB A » • tT -0 .4 4 -0 .7 2 -1 .0 6 -2 .0 7 -2 .7 8 -1.6 0 -1 .2 2 -1.9 2 -0 .6 6 -3.12 Note: See Table 8. http://fraser.stlouisfed.org/RESERVE BANK OF ST. LOUIS FEDERAL Federal Reserve Bank of St. Louis t A2 » < tT -3.07 -3.64 -3.81 -2.71 -5.26 -3 .0 6 -3 .6 4 -3 .8 5 -3 .3 5 -5 .2 5 t * < -7.3 9 -10.42 -8 .2 2 -5 .0 2 -8.07 t -7.35 -10.36 -8.1 8 -5.0 7 -8 .0 3 89 Table 1 0 Unit Root Tests for Australia L t SSM2 DM2 SSM 3 (2) DM3 (2) GDP DEF TB A t m 0.97 0.28 2.55 2.39 -0 .2 2 -1.15 -1.7 3 -2 .4 4 -1.7 3 -0 .2 6 -0 .4 3 -2 .4 8 -1.19 3.46 t A2 t -3.21 -3 .0 2 -3.1 8 -3.01 -3 .6 8 -3 .3 4 -3 .3 3 -3 .3 0 -3 .0 2 -4 .2 4 -3 .8 8 -3 .6 4 -3 .4 6 -3.31 t t -4 .7 6 -5 .4 5 -4 .5 6 -5.17 -5.0 7 - 4 .2 4 -5.17 -4 .7 3 -5 .4 0 -4 .4 9 -5.1 5 -4 .9 7 -4.41 -5.12 Note: See Table 8. Table 1 1 Unit Root Tests for Germany L t SSM2 (1) DM 2 (1) SSM 3 (2) DM3 (1) G DP (1) DEF (2) TB (2) A tT -1.7 2 -1.9 7 -2 .7 2 -2 .6 2 -1.0 8 -0 .9 2 -1.9 7 -0 .7 5 -0 .4 5 -2 .0 2 -1.01 0.24 -1 .9 9 -1 .9 9 t A2 tT -2 .9 7 -3 .6 9 -3 .2 3 -3 .8 4 -6.21 -3.21 -3 .5 8 -3.11 -3 .7 4 -2 .9 6 -3 .7 8 -6 .2 3 -2 .7 6 -3 .6 3 t /* »T -9 .3 8 -8 .2 7 -6.81 -7 .6 5 -11.18 -7 .2 4 -6 .9 8 -9 .4 3 -8 .2 0 -6 .7 5 -7.57 -11.08 -7.17 -6 .9 3 Note: See Table 8. Table 12 Unit Root Tests for Switzerland______________________ L t SSM1 DM1 SSM2 (1) DM2 (1) GDP DEF TB -1.17 -1.18 0.15 -1.19 0.95 -1.0 6 -1.7 6 A tT -2 .3 2 -2 .3 5 -1 .4 4 -1.91 -1.5 7 -0 .9 4 -1 .9 9 t A2 t » • -3 .4 3 -3 .4 4 -3 .3 0 -3.19 -2 .1 0 - 2 .2 2 -3 .0 3 -3 .4 0 -3 .4 0 -3.31 -3 .1 8 -2 .7 8 -2 .0 3 -2 .9 9 » * *T -5.11 -5 .1 0 -7.16 -7 .2 5 -4.81 -4 .7 6 -5 .8 4 -5 .0 7 -5 .0 6 -7.07 -7 .2 2 -4 .7 5 -4 .8 7 -5.81 Note: See Table 8. MARCH/APRIL 1994 90 Table 13 Unit Root Tests for Canada L t SSM1 SSM 2 SSM 3 SSL DM1 DM 2 DM 3 DL GDP DEF TB A tT -1.21 -1.15 -0 .7 0 1.48 -1.18 -0 .7 4 -1 .4 0 -0 .7 9 -1.87 -2.2 1 -1 .3 6 -2 .6 4 -1 .6 5 -2 .0 4 -1 .8 0 -2 .7 2 -2 .0 6 -2 .6 3 -2 .2 7 -1 .2 8 -1 .2 3 -1 .4 4 t A2 tT -3 .0 2 -2 .2 1 -1.7 2 0.32 -2 .9 3 -2 .6 8 - 2 .2 5 -2 .5 3 -2 .9 9 -1.79 -3 .2 4 -3 .7 3 -2.61 -2 .3 7 -0 .3 6 -3 .6 6 -3.21 -3 .2 0 -3 .2 6 -3.11 -1.9 4 -3 .2 6 t tT i* -5 .7 7 -4 .8 5 -5.6 7 -1 .9 9 -5 .2 3 -5 .4 4 -5 .4 7 -5 .0 6 -5 .7 2 -4 .1 3 -5 .8 6 -5 .7 2 -4 .8 6 -5 .6 6 -2 .3 8 -5 .1 8 -5 .4 5 -5 .5 2 -5 .1 2 -5 .6 6 -4 .1 8 -5.81 Note: See Table 8. Table 14 Unit Root Tests for Japan A L t tT t A2 tT t b tT SSM1 -1.18 -2 .7 8 -3 .0 8 -3.19 -4 .7 2 -4 .6 7 DM1 -1.19 -2 .7 4 -3.01 -3.1 3 -4 .6 4 -4 .6 0 SSM2 -1.14 -2.31 -2 .5 5 -2 .6 5 -3 .2 6 -3.19 DM2 -0 .8 5 -2.31 -2 .0 8 -1.9 6 -3 .5 2 -3 .5 3 SSM 3 -2.11 - 2 .2 2 -2 .4 4 -2.41 -2 .6 9 -2 .5 9 DM 3 -1 .3 4 -2 .5 8 -1.7 0 -1 .9 2 -3 .2 8 -3 .2 8 -9 .2 9 0.38 -1.16 -5.1 5 -5 .1 2 -9.41 DEF -2 .0 7 -2 .4 7 -2 .4 5 -2 .3 5 -4 .8 7 -5 .1 0 TB - 2 .2 3 -3 .3 6 -3 .9 0 -3 .6 4 - 2 .8 3 -2.7 1 GNP (1) Note: See Table 8. FEDERAL RESERVE BANK OF ST. LOUIS 91 made a dram atic difference to the estimated fratio on ir (without significantly affecting the non-autocorrelatedness properties of the residu als). Indeed, w ith three lags all of the money m easures w ith the exception of simple-sum M3 (SSM3) and simple-sum L (SSL) appear to be 1(1); the form er variable appears 1(1) w hen q = l, while SSL appears 1(2) at all lag lengths (again, the residuals in each of these cases w ere non autocorrelated). The country with the greatest preponderence of m onetary aggregates being 1(2) is Canada, in which six out of the eight chosen m onetary ag gregates appear to have tw o unit roots. The finding that the level of a country’s price series is an 1(2) process finds confirmation in a num ber of other empirical studies (see, for example, Johansen and Juselius, 1990). Furtherm ore, the finding that m onetary aggregates are 1(2) has also been reported by other researchers (Rasche, 1993), although this finding does not appear to be as robust as that for price defla tors. We now tu rn to an analysis of the cointegra tion properties of a vector of variables for each of our chosen countries. In particular, for each country w e use the m ethods of Johansen (1988, 1991) to estimate the num ber of cointegrating vectors in y ' = [^m, gdp, def, tb], w here m denotes the money supply, ^ is either ss (for simple-sum) or d (for Divisia), gdp is real output, d e f is the deflator corresponding to output, and tb denotes the relevant interest rate (usually a Treasury bill rate). The fact that the variables entering y ' may for any one country be a mix of 1(1) and 1(2) processes has to be taken into account in our implem entation of the Johansen procedure, since the latter is only appropriate for 1(1) variables and driftless 1(0) variables. We therefore use the inform ation from our unit root tests to reduce the order of integration of any 1(2) variables to 1(1), by entering the first difference of the level of such a variable. Thus, if a country's price level is 1(1), we enter the change in the price level (equivalent to the infla tion rate, since the price level is transform ed logarithmically) and/or if the m oney m easure is also 1(2), it is also entered in differences. In the context of estimating a conventional money dem and function (which has the same set of variables as are contained in our y vec tor), Johansen (1991) has suggested dealing with the two unit roots in m and p by respecifying the y vector to consist of (m-p), y, i and Ap. However, given the nature of the c u rren t exer cise, and also since, in m any instances it is only p that appears to be 1(2), we do not believe that such a specification is as attractive as the one adopted here. To determ ine the num ber of unit roots in y ' we use the following method, due to Johansen (1988, 1991). This m ethod may be thought of as the m ultivariate equivalent of (1). It is assumed that y t has the following autore gressive representation w ith Gaussian e rro rs £,: (2) y t = n ^ , . , + n2 + . . . + n k y ,.k + £, y,_2 t = 1,2, ...,T. Equation 2 may be reparam eterized as (20 Ay, = n + Uyt_k + J j T Ay,.,. + u,, k w here q = k - 1, II = S B. - /, Bj is an (n x n) m atrix from the lag polynomial in the (levels) k VAR and T. = -T , B for i=l,...q. The key 1 j-i+1 ^ difference betw een 1 and 2 is th at in 2 there is no allowance for a deterministic trend (or that the series are driftless). The long-run static equilibrium corresponding to 2 is1 (3) n * = 0. The m atrix II is the m ultivariate analogue of 7r in equation 1. Assuming that the variables en tering the y vector do not have an order of in tegration greater than 1, then the right-hand side of equation 2 can only be stationary if the components of Ily, k are stationary. This, in turn, may be determ ined by the rank, r, of the m atrix II, and, in particular, w hether 0 < r < n, w here n denotes the num ber of variables in y. If r=n (that is, II has full rank) then Ily, t can only be stationary if all n linearly indepen dent combinations of y t_k form ed using II are stationary: A standard VAR analysis in levels is appropriate here. If, at the other extreme, r= 0 (and 11 = 0) then there are no linear combina tions in y, which are stationary, and (2) th ere fore becomes a VAR in first differences (this is the kind of VAR specification used in the 1Dynamic steady-state equilibrium simply involves the addi tion of a term in the constant vector of steady-state growth rates to equation 2, which we omit here for expositional purposes; this does not affect the subsequent discussion. MARCH/APRIL 1994 92 majority of traditional G ranger causality tests). If, how ever 0 < r < n , II will be of reduced rank and there m ust exist (n x r) m atrices a and /J such that II = ap', and for IIyt_t to be station ary p'y,_k m ust be stationary. The ft matrix therefore contains the cointegrating vectors and a represents the m atrix of adjustm ent vectors. For example, if P\ is the ith row of /?' then: (4) P \yt ~ /(0). Johansen (1988, 1991) has proposed a maxi mum likelihood m ethod of estimating all of the cointegrating vectors contained in II and sig nificance tests to determ ine how many of the vectors are statistically significant. Since the Jo hansen technique is now well-known, we do not present it here. Instead, we simply note the two test statistics used to determ ine the num ber of significant cointegrating vectors. In our application the likelihood ratio, or trace, test statistic (LR1), for the hypothesis that there are at most r distinct cointegrating vec tors, is (5) Lfll = T t ln( 1 -A), l-P +1 w here A r+1,...,An are the n - r smallest squared ca nonical correlations betw een the y t k and Ay, corrected for the effect of the lagged differ ences of the y( process (for details of how to ex tract the A/s, see Johansen 1988). Additionally, the likelihood ratio statistic for testing at most r cointegrating vectors against the alternative of r + 1 cointegrating vectors is given by equation 4 (6) LR2 = T/n(l-A r+1). Johansen (1988) shows that equations 5 and 6 have a non-standard distribution under the null hypothesis. He does, however, provide approxi m ate critical values for the statistic, generated by Monte Carlo methods. (The critical values recorded in Johansen’s 1988 paper are for a VAR w ithout an intercept term or seasonal dummies. Since these w ere included in our em pirical analysis, we used the critical values for 5 and 6 reported in Johansen and Juselius, 1990.) In Table 15, our estim ated values of LR1 and LR2 are presented, and the critical values and relevant null hypothesis are reported at the bot tom of the table. Consider first the results for the United States. Interestingly, th ere is no evi dence of cointegration for any of the narrow m onetary m easures (i.e. M l and MIA). However, w ith the exception of SSM3, th ere is clear evi dence of one unique cointegrating vector for all FEDERAL RESERVE BANK OF ST. LOUIS m onetary m easures which are broader than M l. It follows from this that it is the introduction of these broader m onetary m easures that produces a cointegrating set (and not the income, interest rate or inflation rate). Since the Rotemberg cu r rency equivalent m easure appears to be station ary around a determ inistic trend, it would appear not to be an ideal candidate for the Johansen methodology. However, for complete ness, and also since it is often difficult to dis crim inate betw een a variable w hich is stationary around a tren d and one which has a stochastic unit root, we also test for the num bers of coin tegrating vectors in a y vector defined for RCE. Interestingly, this also gives strong evidence of one cointegrating vector, as does the BCE m ea sure. (BCE is a variety of currency equivalence which uses Divisia weights.) The evidence for other (non-U.S.) countries in Table 15 is also suggestive of there being long-run relationships contained in different specifications of the y ’ vectors: The vast majority of m onetary m ea sures produce at least one cointegrating vector and m any produce two. Again, th ere does not appear to be any split betw een Divisia and simple-sum m onetary m easures in term s of the production of cointegrating relationships. The broad picture to em erge from Table 15 is that there is strong evidence of at least one cointegrating vector for most country/m oney combinations. It also seems that, at least in this long-run modelling context, th ere is no sharp distinction betw een Divisia and simple-sum money. It may be, however, th at one or other m onetary m easures produce m ore "sensible” estimates of the cointegrating vector and we re tu rn to this point in a later section (where we also examine sample specific issues which may be im portant for the United States). However, for the im plem entation of our causality tests, the main implication to be draw n from Table 15 is that a causality relationship specified simply in differences will be misspecified for the vast majority of country/m oney combinations. We therefore propose estimating the vector e rro r correction models implied by our cointegration results and subjecting them to exclusion tests on the lags of each of the differenced (either first or second differenced, depending on the out come of the results reported in Tables 8 to 14) and also on the lagged cointegrating term s. Since we correct the coefficient variance-covariance matrix for heteroskedasticity (using the m ethods of Hansen, 1980; and White, 1978), the exclusion tests are perform ed using linear Wald statistics, 93 Table 15 Estimated Trace and AMax Statistics U nited States Trace (LR1) SSM1 DM1 SSM1A DM1A SSM2 DM2 SSM3 DM3 SSL DL 0.02 5.19 18.55 40.17 0.10 0.07 18.05 38.13 0.07 7.66 17.83 33.73 0.14 7.73 18.69 40.25 3.80 10.38 23.34 57.92 5.13 12.80 23.34 58.85 3.45 8.14 20.54 47.67 5.08 11.59 21.91 52.35 2.31 8.05 20.65 54.99 4.77 10.99 21.53 53.27 United States AMAX (LR2) SSM1 DM1 SSM1A DM1A SSM2 DM2 SSM3 DM3 SSL DL 0.02 5.18 13.35 21.61 0.01 5.06 12.88 20.08 0.07 7.59 10.16 15.90 0.14 7.58 10.97 21.56 3.80 6.58 12.94 34.58 5.13 7.66 10.63 35.41 3.45 4.66 12.39 27.13 5.08 6.51 10.32 30.44 2.31 5.73 12.60 34.34 4.77 6.22 10.53 31.74 United Kingdom T race (LR1) AMax (LR2) SSM4 DM4 SSM4 OM4 1.89 10.20 38.34 88.45 3.85 12.17 37.53 64.20 1.89 8.31 28.13 50.12 3.85 8.32 25.36 26.67 MARCH/APRIL 1994 94 Table 15 (continued) Estimated Trace and AMax Statistics Australia T race (LR1) AMax (LR2) SSM2 DM2 SSM3 DM3 SSM2 DM2 SSM3 DM3 0.01 10.51 21.89 53.96 0.75 5.74 26.93 55.67 0.61 10.77 31.97 67.61 0.12 6.23 23.12 56.41 0.01 10.50 11.38 32.07 0.74 4.99 21.18 28.74 0.06 10.71 21.19 35.64 0.12 6.12 16.88 33.79 G erm any T race (LR1) AMax (LR2) SSM2 DM2 SSM3 DM3 SSM2 DM2 SSM3 DM3 0.02 11.11 29.62 63.38 0.28 12.62 39.21 77.27 0.81 17.08 40.02 70.38 0.68 18.44 43.56 81.55 0.01 11.10 18.51 33.75 0.03 12.34 26.58 38.06 0.81 16.26 22.94 30.37 0.68 17.76 25.11 37.98 Sw itzerland Trace (LR1) AMax (LR2) SSM1 DM1 SSM2 DM2 SSM1 DM1 SSM2 DM2 0.59 15.54 35.55 56.74 0.60 15.50 35.57 57.08 0.49 12.41 33.91 64.82 1.53 13.29 33.71 60.78 0.59 14.93 20.01 21.18 0.60 14.89 20.07 21.51 0.49 11.91 21.49 35.91 1.53 11.76 20.41 27.08 FEDERAL http://fraser.stlouisfed.org/RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis 95 Table 15 (continued) Estimated Trace and AMax Statistics Canada Trace (LR1) AMax (LR2) SSM1 DM1 SSM2 DM2 SSM1 DM1 SSM2 DM2 2.35 15.06 41.85 79.71 2.67 15.00 43.42 81.55 0.77 8.28 21.36 53.38 0.16 6.66 17.74 55.68 2.35 12.71 26.79 37.86 2.67 12.33 28.41 38.13 0.77 7.51 13.07 32.03 0.16 6.50 11.08 37.95 Japan Trace (LR1) AMax (LR2) SSM1 DM1 SSM2 DM2 SSM3 DM3 SSM1 DM1 SSM2 DM2 SSM3 DM3 3.09 12.74 32.26 65.14 30.08 12.97 32.19 67.00 2.42 16.28 38.76 71.17 0.62 15.03 34.97 69.63 3.46 16.91 39.39 67.29 3.03 15.41 35.25 64.82 3.04 9.64 19.51 32.88 3.08 9.89 19.21 34.81 2.42 13.85 22.47 32.41 2.62 12.41 19.95 34.65 3.46 13.45 22.48 27.90 3.03 12.37 19.83 29.56 Null hypotheses and 5 percent critical values for T race and AMax statistics. Trace AMax Null Hypothesis 5 % Critical Value r< 3 r< 2 r< 1 r= 0 8.18 17.95 31.53 48.28 Null Hypothesis r= 3 r=2 r= 1 r= 0 | r= 4 | r= 3 jr=2 | r= 1 5 % Critical Value 8.18 14.90 21.07 27.14 Note: Variables are defined in Table 8. The Trace and AMax statistics are defined in the text. which have a central chi-squared distribution. C a u sa lity T e sts The exclusion tests for each country are reported in Tables 16 through 22. Consider first the results for the United States, reported in Ta ble 16. Since there is some ambiguity regarding the stochastic properties of the Rotemberg cu r rency equivalent m easure (see discussion above), we present two systems for this variable: one in w hich the variable enters as a level and a deter ministic time tren d is included in each equation of the system (the system with RCE1), and one in which it enters as a first difference and the ECM term from the Johansen estimates re p o rt ed in Table 15 is also included in each equation of the system (the system w ith RCE2). In term s of the U.S.’ real output relationship, there is a very clear, significant, short-run in fluence of the Treasury bill rate. This influence is repeated in all of the other equations as well, (apart from the deflator equation w hen SSM1 and DM1 are used). This confirms the findings of m uch other research on the im portance of including an interest rate in the causality specification (see: Sims, 1980; and Friedm an and Kuttner, 1993). In equations in which an ECM term appears, the majority of significant impacts tend to occur in equations w hich feature the deflator or money (Divisia or simple sum) as the dependent variable. W hat then of the potential short-run differential impact of simple-sum and Divisia money? Interestingly, and in contrast to our initial discussion of the long-run impact, MARCH/APRIL 1994 96 Table 16 Causality Tests for the United States SSM1 SSM1 GDP DEF TB 87.21 1.22 2.85 4.00 (0.00) (0.87) (0.58) (0.00) DM1 DM1 GDP DEF TB 37.88 0.17 1.39 17.32 (0.00) (0.99) (0.84) (0.00) SSM1A SSM1A GDP DEF TB 68.74 2.73 4.90 21.86 (0.00) (0.60) (0.29) (0.00) DM1A DM1A GDP DEF TB 57.30 3.45 3.77 16.81 (0.00) (0.48) (0.44) (0.00) SSM2 SSM2 GDP DEF TB ECM 48.63 13.67 20.93 27.36 17.23 (0.00) (0.00) (0.00) (0.00) (0.00) DM2 DM2 GDP DEF TB ECM 23.71 10.09 6.72 95.24 18.39 (0.00) (0.04) (0.15) (0.00) (0.00) SSM3 SSM3 GDP DEF TB ECM 89.12 11.88 12.32 10.52 8.50 FEDERAL RESERVE BANK OF ST. LOUIS (0.00) (0.02) (0.02) (0.03) (0.00) GDP 4.19 28.67 5.55 27.45 (0.38) (0.00) (0.23) (0.00) GDP 5.38 28.08 5.86 28.33 (0.25) (0.00) (0.21) (0.00) GDP 1.07 22.40 5.16 21.22 (0.89) (0.00) (0.27) (0.00) GDP 1.12 26.46 5.59 21.17 (0.89) (0.00) (0.23) (0.00) GDP 7.34 9.73 8.51 13.97 2.25 (0.12) (0.04) (0.07) (0.00) (0.13) GDP 9.97 7.78 6.95 18.35 4.52 (0.04) (0.09) (0.14) (0.00) (0.03) GDP 5.04 15.23 6.84 22.28 0.68 (0.28) (0.00) (0.14) (0.00) (0.41) DEF 4.22 4.63 56.56 6.28 (0.38) (0.33) (0.00) (0.18) DEF 5.04 4.53 52.46 6.81 (0.28) (0.34) (0.00) (0.15) DEF 12.78 2.37 64.57 9.45 (0.01) (0.00) (0.00) (0.05) DEF 12.54 3.30 61.65 9.29 (0.01) (0.51) (0.00) (0.05) DEF 4.82 1.50 23.76 14.23 8.29 (0.31) (0.83) (0.00) (0.00) (0.00) DEF 10.74 2.45 26.71 18.51 12.19 (0.03) (0.65) (0.00) (0.00) (0.00) DEF 8.74 2.61 25.13 13.14 13.25 (0.07) (0.63) (0.00) (0.01) (0.00) TB 17.43 10.61 13.44 19.59 (0.00) (0.03) (0.00) (0.00) TB 5.66 10.85 11.85 10.01 (0.23) (0.03) (0.02) (0.04) TB 40.14 11.39 10.51 24.78 (0.00) (0.02) (0.03) (0.00) TB 17.12 13.79 12.39 23.56 (0.00) (0.00) (0.01) (0.00) TB 6.75 7.53 14.57 12.42 0.60 (0.15) (0.11) (0.00) (0.01) (0.44) TB 9.23 18.64 6.73 14.47 4.74 (0.05) (0.00) (0.15) (0.00) (0.03) TB 1.04 9.80 10.49 9.91 2.67 (0.90) (0.04) (0.03) (0.04) (0.10) 97 Table 16 (continued) Causality Tests for the United States DM3 DM3 GDP DEF TB ECM 38.22 9.20 6.45 78.27 14.58 (0.00) (0.05) (0.17) (0.00) (0.00) SSL SSL GDP DEF TB ECM 92.18 24.22 13.82 20.53 17.05 (0.00) (0.00) (0.00) (0.00) (0.00) DL DL GDP DEF TB ECM 33.11 12.87 6.47 68.23 14.45 (0.00) (0.01) (0.16) (0.00) (0.00) RCE1 RCE1 GDP DEF TB 37.52 17.22 3.55 17.83 (0.00) (0.00) (0.47) (0.00) RCE2 RCE2 GDP DEF TB ECM 26.14 16.07 2.86 16.66 7.64 (0.00) (0.00) (0.58) (0.00) (0.00) BCE BCE GDP DEF TB ECM 12.78 7.41 4.09 62.27 0.27 (0.01) (0.12) (0.39) (0.00) (0.60) GDP 8.64 9.37 6.58 18.73 3.02 (0.07) (0.05) (0.16) (0.00) (0.08) GDP 7.16 15.34 6.94 27.57 0.75 (0.13) (0.00) (0.14) (0.00) (0.38) GDP 12.01 10.37 7.07 23.47 2.76 (0.02) (0.03) (0.13) (0.00) (0.09) GDP 11.78 9.47 7.18 15.47 (0.02) (0.05) (0.13) (0.00) GDP 8.57 13.65 7.11 16.36 1.72 (0.07) (0.00) (0.13) (0.00) (0.18) GDP 5.66 9.83 6.05 17.52 4.46 (0.23) (0.04) (0.19) (0.00) (0.03) DEF 10.82 3.21 29.61 18.79 13.77 (0.03) (0.52) (0.00) (0.00) (0.00) DEF 7.96 3.17 16.05 13.11 13.69 (0.09) (0.53) (0.00) (0.01) (0.00) DEF 7.79 2.91 26.05 16.64 11.96 (0.09) (0.57) (0.00) (0.00) (0.00) DEF 5.27 0.87 53.72 4.92 (0.26) (0.93) (0.00) (0.79) DEF 10.33 1.69 42.38 8.02 4.34 (0.04) (0.79) (0.00) (0.09) (0.04) DEF 5.61 1.46 30.72 12.19 5.79 (0.23) (0.83) (0.00) (0.02) (0.02) TB 7.16 18.27 6.61 15.62 5.66 (0.13) (0.00) (0.16) (0.00) (0.02) TB 2.54 10.07 10.64 11.04 1.17 (0.64) (0.04) (0.03) (0.03) (0.27) TB 7.39 14.26 7.05 15.57 3.94 (0.12) (0.00) (0.13) (0.00) (0.04) TB 32.10 11.08 6.94 10.16 (0.00) (0,02) (0.14) (0.04) TB 28.56 13.13 7.11 10.04 4.74 (0.00) (0.01) (0.13) (0.04) (0.03) TB 7.02 20.56 8.51 12.95 6.44 (0.11) (0.00) (0.07) (0.01) (0.01) Note: The variables are as defined in Table 8. The variable at the column head is the dependent variable. The numbers not in parentheses are linear Wald statistics, while the numbers in paren theses are marginal significance levels. MARCH/APRIL 1994 98 Table 17 Causality Tests for the United Kingdom SSM4 SSM4 GNP DEF TB ECM 39.89 11.20 28.74 4.44 4.26 (0.00) (0.02) (0.00) (0.35) (0.12) GNP 0.26 2.39 2.79 3.97 8.46 15.58 1.43 16.30 6.58 4.54 (0.00) (0.84) (0.00) (0.16) (0.10) (0.99) (0.66) (0.59) (0.41) (0.01) GNP DM4 DM4 G NP DEF TB ECM DEF 4.88 1.91 2.17 3.58 0.92 1.90 9.08 5.28 11.99 24.38 (0.75) (0.06) (0.26) (0.02) (0.00) DEF (0.30) (0.75) (0.70) (0.46) (0.63) 14.59 6.18 12.23 5.74 27.67 (0.00) (0.19) (0.02) (0.22) (0.00) TB 17.59 10.22 3.97 8.66 10.71 (0.00) (0.04) (0.41) (0.07) (0.00) TB 13.52 7.60 8.15 2.11 2.03 (0.00) (0.11) (0.08) (0.72) (0.36) Note: See Table 16. th ere is a clear differential impact. For example, in term s of the output equation, the Divisia m onetary m easure is significant at the 5 percent level in th ree cases (namely, DM2, DL and RCE1) and at the 7 percent level in two in stances (that of DM3 and RCE1), but none of the simple-sum money term s enters significantly even at the 10 percent level. It is also interesting to note th at among the tw o currency equivalent m easures, it is only the RCE m easure which fea tures significantly in the real output equation (confirming the significant influence for this m onetary m easure noted by Belongia, 1993). The significance of these Divisia m easures is repeated in the deflator equations (apart from RCE1, although, additionally, DM1A is also sig nificant), although in these equations one of the simple-sum m easures is also significant (for SSM1A). W ith respect to m onetary causality in the United States, TB equations, both simple sum and Divisia seem to do equally well in that each m easure has significant strikes. The U.K. evidence, reported in Table 17, con trasts sharply w ith that for Switzerland. Neither M4 nor Divisia M4 affects real GNP. However, Divisia M4 does influence the inflation rate. Both m oney m easures influence interest rates. Thus, the superiority of Divisia M4 over M4 is confirm ed (at least so far as inflation is con cerned), but the lack of causality from money to real activity is notew orthy. http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis The Australian results, recorded in Table 18, differ from the U.S. results in that the TB rate does not have a significant short-run influence in any of the real output equations or in the price equations. However, in common w ith the U.S. results, Divisia money is significant—both M2 and M3—in the real output equation, w hereas the simple-sum m easures are not. In contrast, however, it is the simple-sum m easures which have a significant short-run impact in the TB equations rath e r than the Divisia measures. There are also significant long-run influences in all of the equations, although these do not seem to be confined to any particular m easure of money. For Germany, none of the m onetary impulses —neither Divisia nor simple-sum—appears with a significant influence in the real output equa tions, although there would appear to be an in terest rate effect in this equation for the two sum m easures of money. Real GDP has a signifi cant influence in all of the money equations, apart from SSM2. The joint effect of the TB rate is significant in all of the money equations and inflation, in turn, has a significant impact on in terest rates. Both simple-sum and Divisia m onetary meas ures have also a significant influence on infla tion and the TB rate in the Swiss case (Table 20), although in contrast to the German case 99 Table 18 Causality Tests for Australia SSM2 SSM2 GDP DEF TB ECM 18.76 3.21 5.40 21.29 0.68 (0.00) (0.52) (0.25) (0.00) (0.41) DM2 DM2 GDP DEF TB ECM 1 9 .24 (0 .0 0) 4.07 (0.39) 6.08 (0.19) 4.49 (0.34) 4.07 (0.04) SSM3 SSM3 GDP DEF TB ECM 17.40 15.78 10.38 12.91 18.76 (0.00) (0.00) (0.03) (0.01) (0.00) DM3 DM3 GDP DEF TB ECM 27.15 4.24 7.99 5.05 0.13 (0.00) (0.37) (0.09) (0.28) (0.72) GDP 9.05 16.06 17.95 2.51 2.32 (0.06) (0.00) (0.00) (0.64) (0.12) GDP 9.15 15.68 16.68 4.24 8.07 (0.00) (0.00) (0.00) (0.37) (0.00) GDP 7.27 11.46 12.67 2.67 10.14 (0.12) (0.02) (0.01) (0.61) (0.00) GDP 14.94 24.13 24.23 4.99 5.41 (0.00) (0.00) (0.00) (0.28) (0.02) DEF 7.06 30.92 19.35 6.18 0.07 (0.13) (0.00) (0.00) (0.18) (0.79) DEF 4.24 22.32 19.61 4.75 3.59 (0.37) (0.00) (0.00) (0.31) (0.06) DEF 2.91 38.99 18.73 6.21 7.54 (0.57) (0.00) (0.20) (0.18) (0.02) DEF 2.53 30.65 15.79 4.88 0.02 (0.64) (0.00) (0.00) (0.29) (0.88) TB 18.18 3.45 5.33 13.15 8.68 (0.00) (0.48) (0.25) (0.01) (0.00) TB 3.87 4.91 4.01 11.81 0.70 (0.42) (0.79) (0.40) (0.02) (0.40) TB 9.72 4.66 5.74 19.21 6.52 (0.04) (0.32) (0.22) (0.00) (0.04) TB 1 .4 7 (0 .8 3 ) 5.45 (0.24) 6.71 (0.15) 11.33 (0.02) 6.78 (0.00) Note: See Table 16. MARCH/APRIL 1994 10 0 Table 19 Causality Tests for Germany SSM2 SSM2 GDP DEF TB ECM 6.92 1.49 3.99 10.56 3.91 (0.14) (0.83) (0.41) (0.03) (0.05) DM2 DM2 GDP DEF TB ECM 8.92 18.82 5.15 24.47 50.64 (0.06) (0.00) (0.27) (0.00) (0.00) SSM3 SSM3 GDP DEF TB ECM 4.71 17.41 37.78 21.18 38.10 (0.32) (0.00) (0.00) (0.00) (0.00) DM3 DM3 GDP DEF TB ECM 16.73 30.32 50.59 41.28 55.30 Note: See Table 16. FEDERAL http://fraser.stlouisfed.org/RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis (0.00) (0.00) (0.00) (0.00) (0.00) GDP 3.17 31.97 1.62 16.55 6.41 (0.53) (0.00) (0.81) (0.00) (0.01) GDP 1.54 9.08 3.28 5.05 5.54 (0.82) (0.05) (0.51) (0.78) (0.06) GDP 3.79 24.94 1.52 10.68 6.46 (0.43) (0.00) (0.82) (0.03) (0.04) GDP 2.37 19.03 1.08 1.68 1.22 (0.66) (0.00) (0.89) (0.79) (0.54) DEF 9.08 9.39 24.15 7.68 11.79 (0.00) (0.06) (0.00) (0.10) (0.00) DEF 9.88 6.52 33.66 12.75 21.10 (0.04) (0.16) (0.00) (0.01) (0.00) DEF 0.04 1.58 9.92 12.55 3.79 (0.99) (0.81) (0.04) (0.01) (0.19) DEF 14.54 4.63 7.29 7.14 30.37 (0.00) (0.33) (0.12) (0.13) (0.00) TB 0.81 3.90 9.33 14.35 0.85 (0.74) (0.42) (0.05) (0.00) (0.36) TB 6.93 6.87 10.16 5.02 1.82 (0.14) (0.14) (0.04) (0.78) (0.40) TB 6.32 3.22 3.37 9.99 25.81 (0.18) (0.52) (0.49) (0.04) (0.00) TB 7.16 4.64 11.17 9.15 0.53 (0.13) (0.33) (0.02) (0.06) (0.76) 101 Table 20 Causality Tests for Switzerland SSM1 SSM1 G NP DEF TB ECM 7.47 22.21 14.71 10.35 25.22 (0.11) (0.00) (0.00) (0.03) (0.00) DM1 DM1 GNP DEF TB ECM 7.26 20.11 15.04 9.75 24.11 (0.12) (0.00) (0.00) (0.04) (0.00) SSM2 SSM2 GNP DEF TB ECM 7.69 5.70 11.74 30.06 14.15 (0.10) (0.22) (0.02) (0.00) (0.00) DM2 DM2 GNP DEF TB ECM 10.17 4.54 5.07 5.86 3.66 (0.04) (0.33) (0.28) (0.21) (0.16) GNP 18.96 50.69 13.43 26.68 10.65 (0.00) (0.00) (0.00) (0.00) (0.00) GNP 18.83 50.48 12.91 26.38 10.66 (0.00) (0.00) (0.02) (0.00) (0.00) GNP 5.59 46.40 17.95 49.45 32.73 (0.23) (0.00) (0.00) (0.00) (0.00) GNP 10.64 76.37 8.85 25.41 19.53 (0.03) (0.00) (0.06) (0.00) (0.00) DEF 24.09 18.80 21.44 7.15 13.21 (0.00) (0.00) (0.00) (0.13) (0.00) DEF 24.75 19.68 21.76 7.59 13.95 (0.00) (0.00) (0.00) (0.11) (0.00) DEF 13.10 12.01 35.34 15.47 10.53 (0.01) (0.02) (0.00) (0.00) (0.00) DEF 30.98 23.46 44.67 12.71 50.06 (0.00) (0.00) (0.00) (0.01) (0.00) TB 6.56 25.46 10.05 12.74 2.54 (0.16) (0.00) (0.04) (0.01) (0.28) TB 6.59 24.96 10.18 12.75 2.67 (0.16) (0.00) (0.04) (0.01) (0.26) TB 29.64 13.26 14.81 23.78 11.94 (0.00) (0.01) (0.01) (0.00) (0.00) TB 18.57 26.24 19.76 15.15 19.04 (0.00) (0.00) (0.00) (0.00) (0.00) Note: See Table 16. MARCH/APRIL 1994 102 Table 21 Causality Tests for Canada SSM1 SSM1 GNP DEF TB ECM 3.48 6.32 2.09 8.94 15.08 (0.48) (0.18) (0.72) (0.06) (0.00) GNP 10.22 7.11 15.89 1.91 11.46 DM1 DM1 GNP DEF TB ECM 5.39 11.20 2.34 7.88 16.96 (0.25) (0.02) (0.67) (0.09) (0.00) TB ECM 22.25 5.24 10.13 49.24 0.04 (0.00) (0.76) (0.04) (0.00) (0.83) (0.03) (0.13) (0.00) (0.75) (0.00) GNP 8.89 7.09 34.00 9.18 46.18 6.65 2.75 7.57 5.11 9.85 4.79 3.27 10.53 3.98 1.80 (0.31) (0.51) (0.03) (0.41) (0.40) DEF (0.06) (0.13) (0.00) (0.05) (0.00) GNP SSM2 SSM2 GNP DEF DEF 5.16 5.95 16.58 1.39 0.68 (0.27) (0.20) (0.00) (0.84) (0.71) DEF (0.16) (0.60) (0.11) (0.28) (0.00) 12.36 4.74 29.68 6.06 0.68 (0.01) (0.31) (0.00) (0.19) (0.41) DM2 GNP DEF DM2 GNP DEF 30.89 (0.00) 8.97 (0.06) 8.61 (0.07) 1.24 (0.84) 2.02 (0.73) 11.33 (0.02) 5.46 (0.24) 2.89 (0.68) 19.24 (0.00) TB 2 5 .8 5 (0 .00) 2.41 (0.66) 3 .5 4 (0 .47) 2.81 (0.09) 7.53 (0.00) 0 .6 0 (0 .44) ECM both Divisia m easures appear significant in the output equation, as does SSM1. In common with a num ber of other countries, the TB rate has a statistical influence in all of the output equa tions and in three of the m oney equations. It is notew orthy that the joint effect of inflation is statistically significant in three out of four of the output equations. The Canadian results (Table 21) portray little significant impact of money on any variable (the exceptions being SSM3 and DM3 in the TB equa tion). Interest rates also do not have the same significant role to play as they did in the U.S. case for real output, although they do feature FEDERAL RESERVE BANK OF ST. LOUIS TB 2.77 11.55 5.54 16.95 9.02 (0.54) (0.02) (0.23) (0.00) (0.01) TB 3.82 9.65 7.17 16.68 5.01 (0.43) (0.04) (0.12) (0.00) (0.08) TB 4.62 15.11 19.07 11.16 0.78 (0.33) (0.00) (0.00) (0.02) (0.38) TB 6.97 12.25 19.59 9.88 0.11 (0.14) (0.01) (0.00) (0.04) (0.74) in the majority of m oney equations. The effects of price (or, m ore correctly, inflation) feature prom inently in almost all of the TB equations. The Japanese causality p attern (reported in Table 22) is in many ways similar to th at for Germany. Thus, neither simple-sum nor Divisia money enters significantly into the output equa tion, although there is a significant impact of both types of money in the inflation and TB equations. The TB rate also features significantly in all of the Japanese real output equations but, in contrast to the German case, only enters sig nificantly into one other equation (apart from its ow n lags)—that for DM3. 103 Table 21 (continued) Causality Tests for Canada GNP SSM3 SSM3 GNP DEF TB ECM 64.96 8.25 7.34 16.58 0.00 (0.00) (0.08) (0.12) (0.00) (0.95) 6.30 4.25 8.13 7.00 13.33 22.32 12.82 26.96 13.52 2.96 (0.00) (0.01) (0.00) (0.00) (0.23) 1.27 4.67 6.08 1.88 12.18 SSL SSL GNP DEF TB ECM 15.36 3.51 3.98 7.50 1.15 (0.00) (0.48) (0.41) (0.11) (0.78) 1.63 1.67 11.92 1.78 11.48 59.15 14.29 15.84 9.34 1.97 (0.00) (0.00) (0.00) (0.05) (0.16) (0.86) (0.32) (0.19) (0.76) (0.00) (0.12) (0.12) (0.00) (0.06) (0.56) 5.27 4.01 18.74 3.00 1.20 (0.26) (0.40) (0.00) (0.56) (0.55) DEF (0.80) (0.79) (0.02) (0.86) (0.00) GNP 1.38 2.25 9.52 2.04 8.55 7.25 7.24 20.25 9.04 0.34 DEF GNP DL DL GNP DEF TB ECM (0.17) (0.37) (0.08) (0.14) (0.00) GNP DM3 DM3 GNP DEF TB ECM DEF 0.97 4.97 20.64 3.87 0.51 (0.91) (0.29) (0.00) (0.42) (0.48) DEF (0.84) (0.68) (0.04) (0.73) (0.00) 1.38 3.98 21.27 3.78 0.58 (0.84) (0.41) (0.00) (0.44) (0.44) TB 13.12 15.93 27.09 10.08 1.02 (0.01) (0.00) (0.00) (0.04) (0.31) TB 10.69 13.26 21.38 7.74 0.52 (0.03) (0.01) (0.00) (0.10) (0.77) TB 6.39 11.61 20.11 14.04 0.57 (0.17) (0.02) (0.00) (0.00) (0.45) TB 4.19 (0.38 9.38 (0.05) 16.82 (0.00) 10.11 (0.04) 0.09 (0.76) Note: See Table 16. We may summ arize the results reported in this section in the following way. First, there appear to be countries in which Divisia money has greater informational content than simplesum money and this is most clear in the U.S. and Australian cases. This is possibly because the pace of financial innovation has varied across countries. Such differential impacts across countries also holds tru e for other variables. In particular, the widely cited effect that the in terest rate has in U.S. causality tests does not seem to carry over to other countries. Also, although Divisia m oney does not have a signifi cant impact in all countries, the evidence for the United Kingdom suggests th at this, at least in part, may be attributable to the sophistication w ith which the Divisia m easure is constructed. Thus, although none of the U.K. Divisia m easures has a significant effect on real output, the Bank of England Divisia m easure (BOED) does have significant inform ational content for inflation. This m easure is widely regarded as being superi or to the other m easures, perhaps because the Bank economists had access to detailed data on asset compositions w hich are not publicly available. MARCH/APRIL 1994 104 Table 22 Causality Tests for Japan SSM1 SSM1 GNP DEF TB ECM 6.82 4.81 2.56 3.84 10.66 (0.15) (0.31) (0.63) (0.43) (0.00) DM1 DM1 GNP DEF TB ECM 11.93 5.25 2.21 4.57 11.36 (0.02) (0.26) (0.69) (0.33) (0.00) SSM2 SSM2 GNP DEF TB ECM 39.02 6.02 7.75 7.13 9.15 (0.00) (0.19) (0.10) (0.13) (0.01) DM2 DM2 GNP DEF TB ECM 57.64 4.36 9.19 8.84 6.83 (0.00) (0.36) (0.05) (0.06) (0.03) SSM3 SSM3 GNP DEF TB ECM 47.24 4.06 6.97 7.03 7.74 (0.00) (0.39) (0.14) (0.13) (0.02) DM3 DM3 GNP DEF TB ECM 66.52 3.19 9.70 9.41 6.07 Note: See Table 16. FEDERAL RESERVE BANK OF ST. LOUIS (0.00) (0.53) (0.04) (0.05) (0.05) GNP 1.69 25.08 7.61 9.83 12.33 (0.79) (0.00) (0.11) (0.04) (0.00) GNP 1.46 24.64 8.04 9.17 12.17 (0.83) (0.00) (0.08) (0.06) (0.00) GNP 1.10 21.36 6.37 12.22 14.14 (0.89) (0.00) (0.17) (0.02) (0.00) GNP 1.04 22.02 8.15 14.81 13.31 (0.90) (0.00) (0.08) (0.00) (0.01) GNP 0.74 19.25 6.39 10.49 12.69 (0.94) (0.00) (0.17) (0.03) (0.00) GNP 0.73 20.91 8.13 13.93 13.05 (0.94) (0.00) (0.08) (0.00) (0.00) DEF 4.30 3.82 32.52 7.39 6.55 (0.37) (0.43) (0.00) (0.12) (0.04) DEF 3.97 3.75 34.01 7.17 6.54 (0.41) (0.44) (0.00) (0.13) (0.04) DEF 10.41 4.33 35.58 7.10 3.61 (0.03) (0.36) (0.00) (0.13) (0.16) DEF 11.72 3.92 34.41 6.40 3.78 (0.02) (0.42) (0.00) (0.17) (0.15) DEF 8.53 3.44 34.03 6.97 3.57 (0.07) (0.48) (0.00) (0.14) (0.16) DEF 12.16 3.23 33.71 6.40 3.87 (0.02) (0.52) (0.00) (0.17) (0.14) TB 8.37 3.47 7.85 10.65 10.94 (0.07) (0.48) (0.09) (0.00) (0.00) TB 8.89 3.39 7.92 19.25 11.16 (0.06) (0.49) (0.09) (0.00) (0.00) TB 14.06 6.16 5.18 17.47 10.97 (0.00) (0.18) (0.27) (0.00) (0.00) TB 11.08 4.48 8.04 21.34 15.57 (0.03) (0.34) (0.09) (0.00) (0.00) TB 13.92 7.37 6.49 17.58 10.03 (0.01) (0.11) (0.16) (0.00) (0.00) TB 10.93 4.59 8.20 21.08 14.56 (0.02) (0.33) (0.08) (0.00) (0.00) 105 Table 23 Estimated Trace and AMax Statistics : United States Sub-Samples TRACE (LR1) SSM1 DM1 1.02 11.23 29.31 53.90 0.86 12.42 30.18 53.92 SSM1A DM1A 0.02 12.96 30.33 52.67 0.57 13.50 29.34 51.00 SSM2 DM2 SSM3 DM3 SSL DL RCE BCE 3.25 18.55 36.89 62.29 0.02 10.29 27.23 49.45 0.83 12.11 27.47 54.19 0.01 11.13 27.33 50.09 1.44 14.31 34.77 63.07 0.06 1.44 28.69 50.37 6.07 20.99 41.70 67.90 0.16 9.01 30.04 56.69 AMAX (LR2) SSM1 DM1 1.02 10.22 18.07 24.60 0.86 11.56 17.76 23.74 SSM1A DM1A SSM2 DM2 SSM3 DM3 SSL DL RCE BCE 0.57 13.44 15.84 21.66 3.25 15.30 17.34 26.40 0.02 10.28 16.94 22.21 0.83 11.29 15.35 26.72 0.01 11.12 16.19 22.76 1.44 12.86 20.46 28.91 0.06 9.38 20.46 28.91 6.08 9.38 20.46 26.28 0.16 8.84 21.03 26.66 0.02 12.93 17.37 22.34 S u b -S a m p le R e s u lts f o r th e U n ited S ta tes: th e P o st-1 9 7 9 R e g im e C h ange The causality results reported in the previous section are for the longest span of data for w hich consistent simple-sum and Divisia data are available for each country. W ithin each country-specific data sample, there may be one or two changes in the way m onetary policy has been implemented. Thus, some countries have switched from targeting one particular aggregate to another or have switched from m onetary ta r geting to interest rate targeting, or vice versa. Therefore, it is of interest to inquire if the results reported in the previous section carry through for sub-samples corresponding to specific m onetary regimes. One of the possible examples of a regime change arises in the Unit ed States around 1980, w hen a combination of reform s (including a change in Fed operating procedure and a liberalization of deposit m ar kets) produced an apparent shift in previously stable m onetary relationships (Rasche 1993). Given this, and also since the U.S. data sample is one of the longest, we concentrate our sub sample tests on our U.S. data set. In particular, we have re-estimated our U.S. causality tests for the first q u arter of 1960 to the third qu arter of 1979 (lags being generated within this sample). In Table 23, the estim ated Trace and Amax statistics are reported for our chosen U.S. sub sample. In contrast to the full sample results, it is notew orthy that all of the m onetary m easures produce at least one cointegrating vector (for the full sample, none of the M l m onetary meas ures produced any cointegrating vectors). We therefore use the inform ation concerning the num ber of cointegrating vectors to set up ap propriate VECMs for each m onetary m easure. The sub-sample exclusion tests based upon these VECMs are reported in Table 24. The broad conclusion to em erge from this table is, perhaps not surprisingly, that the sub-sample produces a very different picture w ith respect to the relative m erits of simple-sum and Divisia money. More specifically, w e note th at the sig nificant impact of money in the real output equations occurs for the narrow M l m easures of money and not for the broader m easures (and, in term s of the M l m easures, simple sum seems to outperform Divisia since tw o of the sum m easures are significant at the 5 percent level against one Divisia m easure at this sig nificance level). Of the two currency equivalent measures, RCE is insignificant in the GDP equa tion, while BCE is significant (albeit at the 6 p er cent level of significance) the reverse of our findings for the full sample. O ther notable fea tures of the sub-sample results, which are dis MARCH/APRIL 1994 106 Table 24 Sub-Sample Causality Tests for the United States SSM1 SSM1 GDP DEF TB ECM 57.23 5.87 8.87 15.20 5.55 (0.00) (0.21) (0.06) (0.00) (0.02) GDP 11.28 2.44 5.34 5.94 9.99 27.14 0.62 3.80 22.95 1.20 (0.00) (0.96) (0.43) (0.00) (0.27) 10.69 4.29 7.31 5.28 10.22 SSM1A SSM1A GDP DEF TB ECM 40.06 7.13 4.79 15.13 1.81 (0.00) (0.13) (0.31) (0.00) (0.18) 10.76 3.64 6.82 5.14 10.80 26.63 0.85 3.74 25.91 0.00 (0.00) (0.93) (0.44) (0.00) (0.97) tinct from the full sample results, include the finding of a strongly significant effect of money on the deflator for all m easures of money (ex cept RCE) and a m uch less im portant role for the interest rate in the output equation. (The TB rate is only significant in two instances, w hereas it was significant in all cases for the full sample.) CONCLUSION The evidence from the St. Louis equations is fairly straightforw ard: Divisia weighted ag gregates appear to offer advantages over broad simple-sum m onetary aggregates. The credibility of narrow simple-sum aggregates has universally been underm ined by the spread of financial innovation. Although results of our real income FEDERAL RESERVE BANK OF ST. LOUIS (0.03) (0.37) (0.12) (0.26) (0.00) (0.01) (0.74) (0.00) (0.16) (0.52) 15.16 3.53 11.99 8.33 3.29 (0.00) (0.47) (0.02) (0.08) (0.07) DEF (0.03) (0.46) (0.14) (0.27) (0.00) 16.73 4.18 20.98 8.94 2.96 (0.00) (0.38) (0.00) (0.06) (0.08) DEF GDP 8.14 4.69 8.36 5.31 9.33 13.04 1.98 39.38 6.52 0.42 DEF GDP DM1A DM1A GDP DEF TB ECM (0.02) (0.65) (0.25) (0.20) (0.00) GDP DM1 DM1 GDP DEF TB ECM DEF (0.08) (0.32) (0.08) (0.26) (0.00) 19.14 7.34 4.94 11.08 7.17 (0.00) (0.12) (0.29) (0.03) (0.00) TB 32.78 6.99 1.11 41.57 0.49 (0.00) (0.14) (0.89) (0.00) (0.48) TB 18.57 8.27 1.04 22.31 0.03 (0.00) (0.08) (0.90) (0.00) (0.87) TB 30.93 5.72 0.41 42.39 0.04 (0.00) (0.22) (0.98) (0.00) (0.83) TB 17.75 7.02 0.22 23.14 0.12 (0.00) (0.13) (0.99) (0.00) (0.73) causality tests are less persuasive. However, they still give a clear edge to Divisia aggregates over simple sum. The results are not so strong that we can conclude that Divisia money m at ters while simple sum does not. Nonetheless, it is clear from the U.S. evidence that the advan tages of Divisia are particularly strong after 1980, the period in which financial innovation is greatest. Pre-1980 data do not show any sup port for Divisia. It may well be that if we could base our tests on post-1980 data alone, we would find m uch stronger support for Divisia. Also, the existence of reverse causality (from real income to money) is not particularly sur prising given the fact that most authorities are pegging short-term interest rates or exchange rates. Superficially, this would support the “real business cycle” view or even the "money doesn’t 107 Table 24 (continued) Sub-Sample Causality Tests for the United States GDP SSM2 SSM2 GDP DEF TB ECM 45.44 3.53 5.73 46.46 8.08 (0.00) (0.47) (0.22) (0.00) (0.01) 6.54 7.48 9.99 2.02 15.69 DM2 DM2 GDP DEF TB ECM 45.70 5.31 4.53 76.87 3.63 (0.00) (0.26) (0.34) (0.00) (0.05) 74.22 5.00 4.35 15.43 4.23 (0.00) (0.29) (0.36) (0.00) (0.04) 6.06 9.10 5.14 2.69 0.16 1.01 5.27 9.62 10.61 10.22 98.38 8.00 9.88 38.56 6.14 (0.00) (0.09) (0.04) (0.00) (0.01) m atter” view. However, it may instead be the old problem of observational equivalence. The policy significance of these results may be limited. M onetary authorities can no m ore control Divisia aggregates than they can broad money. However, Divisia aggregates undoubtedly offer potential inform ation to m onetary authori ties about the relative ease or tightness of m onetary stance—m uch m ore so than do broad simple-sum aggregates. However, the body of research supporting Divisia is not yet sufficiently large or robust that we would wish to recom m end direct targeting at this stage. W hat is im portant, however, is that official credible Divisia index num bers should be produced so that researchers can test exhaustively the p erfo r m ance of these indicators. Only w hen a clear (0.19) (0.06) (0.27) (0.61) (0.69) (0.02) (0.57) (0.88) (0.17) (0.00) 17.51 4.16 20.15 8.02 3.26 (0.00) (0.39) (0.00) (0.09) (0.07) DEF (0.91) (0.26) (0.05) (0.03) (0.00) GDP 2.89 5.71 8.99 3.23 10.94 11.22 2.93 1.14 6.48 9.66 DEF GDP DM3 DM3 GDP DEF TB ECM (0.16) (0.11) (0.04) (0.73) (0.00) GDP SSM3 SSM3 GDP DEF TB ECM DEF 20.52 7.44 2.72 11.13 14.81 (0.00) (0.11) (0.61) (0.03) (0.00) DEF (0.57) (0.22) (0.06) (0.52) (0.00) 18.94 9.10 2.68 18.72 9.25 (0.00) (0.06) (0.61) (0.00) (0.00) TB 17.30 7.22 4.03 35.07 1.38 (0.00) (0.12) (0.40) (0.00) (0.50) TB 8.73 10.08 5.75 26.04 2.13 (0.07) (0.04) (0.22) (0.00) (0.14) TB 4.88 7.26 1.99 25.20 0.79 (0.29) (0.12) (0.74) (0.00) (0.37) TB 9.85 4.34 0.94 25.78 0.18 (0.04) (0.36) (0.92) (0.00) (0.67) consensus emerges should policy be directly linked to such indicators. Just because an indi cator does well in the 1980s does not m ean it will do well in the 1990s. Divisia aggregates did particularly well at handling the introduction of interest on checking accounts. They may be less useful in a period of, say, the w idespread adop tion of "sm art” cards. In short, while our results are encouraging enough to suggest th at m onetary authorities should commission fu rth e r w ork on Divisia, the picture w hich em erges is not sufficiently clearcut to lead to immediate policy recom m enda tions. However, the message for the economics profession is m uch clearer. All those who do applied research using money should take on board the fact that simple-sum m easures are MARCH/APRIL 1994 108 substantially distorted and a better m easure is likely to be provided by a m onetary services index constructed along something like Divisia lines. Rejections of the role of money based upon flawed money m easures are themselves easy to reject. Dickey, David A., and Wayne A. Fuller. “Distribution of the Estimators for Autoregressive Time Series with a Unit Root,” Journal o f the Am erican Statistical A ssociation (June 1979), pp. 427-31. Diewert, W.E. “Superlative Index Numbers and Consistency in Aggregation,” Econom etrica (July 1978), pp. 883-900. ________ “Exact and Superlative Index Numbers,” Journal of Econom etrics (May 1976), pp. 115-45. Drake, L., and K. Alec Chrystal. “Sectoral Money Demand,” Journal o f Money, Credit and Banking (forthcoming). REFERENCES Barnett, William A. “The Optimal Level of Monetary Aggrega tion,” Journal o f Money, Credit and Banking (November 1982, part 2), pp. 687-710. ________ “Economic Monetary Aggregates: An Application of Index Number and Aggregation Theory,” Journal o f Econo m etrics (September 1980), pp. 11-48. ________ , Douglas Fisher, and Apostolos Serletis. “Consumer Theory and the Demand for Money,” Journal o f Econom ic Literature (December 1992), pp. 2086-119. ________ , Edward K. Offenbacher, and Paul A. Spindt. “The New Divisia Monetary Aggregates,” Journal o f Political Econom y (December 1984), pp. 1049-85. ________ , and Paul A. Spindt. “The Velocity Behavior and In formational Content of Divisia Monetary Aggregates,” Eco nom ics Letters, vol. 4 (1979), pp. 51-7. Basmann, R.L. “The Causal Interpretation of Non-Triangular Systems of Economic Relations,” Econom etrica (July 1963), pp. 439-48. Belongia, Michael T. “ Measurement Matters: Recent Results from Monetary Economics Re-examined,” manuscript, 1993. ________ , and James A. Chalfant. “The Changing Empirical Definition of Money: Some Estimates from a Model of the Demand for Money Substitutes,” Journal o f Political Econom y (April 1989), pp. 387-97. ________ , and K. Alec Chrystal. “An Admissible Monetary Ag gregate for the United Kingdom,” The Review of Economics and Statistics (August 1991), pp. 497-502. Boulding, Kenneth. “ Minutes of Session V: How Does Monetary Policy Affect the Economy?” Journal o f Money, Credit and Banking (August 1969), p. 555. Campbell, John Y., and Pierre Perron. “Pitfalls and Opportu nities: What Macroeconomists Should Know about Unit Roots,” NBER M acroeconom ics A nnual (1991), pp. 141-201. Chetty, V. Karuppan. “On Measuring the Nearness of NearMoneys,” The Am erican Econom ic Review (June 1969), pp. 270-81. Chrystal, K. Alec, and Ronald MacDonald. “Exchange Rates and Financial Innovation: The Sterling Dollar Rate 1972-90.” Paper presented to Conference of American Statistical Association, held at San Francisco, August 1993. Cockerline, Jon P., and John D. Murray. “A Comparison of Al ternative Methods of Monetary Aggregation: Some Prelimi nary Evidence,” B ank o f Canada Technical Report #28 (November 1981). Cooley, Thomas F., and Stephen F. LeRoy. “Atheoretical Mac roeconomics: A Critique,” Journal o f M onetary Econom ics (November 1985), pp. 238-308. De Long, J. Bradford, and Lawrence H. Summers. “How Does Macroeconomic Policy Affect Output?” Brookings Papers on Econom ic Activity, 2:1988, pp. 433-80. FEDERAL RESERVE BANK OF ST. LOUIS Engle, Robert F., and C.W.J. Granger. “Co-integration and Er ror Correction: Representation, Estimation and Testing,” Econom etrica (March 1987), pp. 251-76. Fisher, Andrew. “Bundesbank criticized by chief economist of Germany’s largest commercial bank” and “Bundesbank told it should drop reliance on M3,” F inancial Times, Au gust 10, 1993. Fisher, Douglas. Money D em and a nd M onetary Policy. Harvester Wheatsheaf, 1989. ________ , and Apostolos Serletis. “Velocity and the Growth of Money in the United States 1970-85,” Journal of M acroeconom ics (summer 1989), pp. 3 2 3 -3 2 . Fisher, Paul, Suzanne Hudson, and Mahmood Pradhan. “Divisia Indices for Money: An Appraisal of Theory and Practice,” B a n k o f England Working Paper Series #9 (April 1993). Friedman, Benjamin M., and Kenneth N. Kuttner. “Another Look at the Evidence on Money-lncome Causality,” Journal o f Econom etrics (May/June 1993), pp. 189-203. ________ , a n d ________ “Money, Income Prices and Interest Rates,” The Am erican Econom ic Review (June 1992), pp. 472-92. Friedman, Milton. A Program for M onetary Stability. Fordham University Press, 1959. Fuller, Wayne A. “ Introduction to Statistical Time Series.” John Wiley and Sons, 1976. Hall, Alastair. “Testing for a Unit Root in Time Series with Pretest Data Based Model Selection,” Journal o f Business a nd Econom ic Statistics (forthcoming). Hansen, Lars Peter. “Large Sample Properties of Generalized Method of Moments Estimators,” Econometrica (July 1982), pp. 1029-54. Holmes, James M., and Patricia A. Hutton. “A New Test of Money-lncome Causality,” Journal o f Money, Credit and Banking (August 1992), pp. 338-55. Horne, Jocelyn, and Vance L. Martin. “Weighted Monetary Aggregates: An Empirical Study Using Australian Monetary Data, 1969-87,” Australian Econom ic Papers (December 1989), pp. 181-200. Hostland, Doug, Stephen Poloz, and Paul Storer. “An Analy sis of the Information Content of Alternative Monetary Ag gregates,” B ank o f Canada Technical Reports #48 (October 1987). Ishida, Kazuhiko. “Divisia Monetary Aggregates and Demand for Money: A Japanese Case,” B ank o f Japan M onetary a nd Econom ic Studies (June 1984), pp. 49-80. Johansen, Soren. “ Estimation and Hypothesis Testing of Cointegration Vectors in Gaussian Vector Autoregressive Models,” Econom etrica (November 1991), pp. 1551-80. ________ “Statistical Analysis of Cointegration Vectors,” Jour n a l o f Econom ic Dynam ics a nd Control (June/Sept. 1988), pp. 231-54. ________ , and Katarina Juselius. “Maximum Likelihood Esti mation and Inference on Cointegration - with Applications to the Demand for Money,” Oxford Bulletin o f Economics and Statistics (May 1990), pp. 169-210. 109 Keynes, John Maynard. A Treatise on Money, volume I (The Pure Theory of Money). Harcourt, Brace and Company, 1930a. ________ . A Treatise on Money, volume II (The Applied Theory of Money). Harcourt, Brace and Company, 1930b. Rotemberg, Julio J. “Commentary: Monetary Aggregates and their Uses,” in Michael T. Belongia, ed., M onetary Policy on the 75th Anniversary of the Federal Reserve System (Proceedings of the 14th Annual Economic Policy Confer ence of the Federal Reserve Bank of St. Louis). Kluwer Academic Publishers, 1991, pp. 223-31. Lindsey, David E., and Paul Spindt. “An Evaluation of Mone tary Indexes,” Special Studies Paper #195. Board of Gover nors of the Federal Reserve System, Division of Research and Statistics, March 1986. Serletis, Apostolos. “The Empirical Relationship between Money, Prices and Income Revisited,” Journal o f Business and Econom ics Statistics (July 1988), pp. 351-8. MacDonald, Ronald, and Colm Kearney. “On the Specifica tion of Granger-Causality Tests Using the Cointegration Methodology,” Econom ics Letters, vol. 25 (1987), pp. 149-53. Sims, Christopher. “Comparison of Interwar and Postwar Bus iness Cycles: Monetarism Reconsidered,” The Am erican Econom ics Review (May 1980), pp. 250-7. Park, Joon Y., and Buhmsoo Choi. “A New Approach to Test ing for a Unit Root,” Cornell University, Center for Analytic Economics, Working Paper #88-23 (May 1988). Perron, Pierre. “Trends and Random Walks in Macroeconomic Time Series: Further Evidence from a New Ap proach,” Journal o f Econom ic Dynam ics and Control (June-September 1988), pp. 297-332. Phillips, Peter C.B., and Pierre Perron. “Testing for a Unit Root in Time Series Regression,” Biom etrika (June 1988), pp. 335-46. Plosser, Charles I. “ Money and Business Cycles: A Real Bus iness Cycle Interpretation,” in Michael T. Belongia, ed., M onetary Policy on the 75th Anniversary o f the Federal Reserve System (Proceedings of the 14th Annual Economic Policy Conference of the Federal Reserve Bank of St. Louis). Kluwer Academic Publishers, 1991, pp. 245-74. Rasche, Robert H. “ Monetary Aggregates, Monetary Policy and Economic Activity,” this Review (March/April 1993), pp. 1-35. Stock, James H. “A Class of Tests for Integration and Coin tegration,” mimeo, Harvard University, May 1990. ________, and Mark W. Watson. “Interpreting the Evidence on Money-income Causality,” Journal o f Econom etrics (January 1989), pp. 161-81. White, Halbert. “A Heteroskedasticity-Consistent Covariance Matrix and a Direct Test for Heteroskedasticity,” Econom etrica (May 1980), pp. 817-38. Yue, Piyu, and Robert Fluri. “Divisia Monetary Services In dexes for Switzerland: Are They Useful for Monetary Targeting?” this Review (September/October 1991), pp. 19-33. Zellner, Arnold. “Causality and Causal Laws in Economics,” Journal o f Econom etrics Annals 1988-3, p. 7-21. ________ . “Causality and Econometrics,” in Three A spects of Policy a nd Policy M aking: Knowledge, Data a nd Institutional Conference (Carnegie-Rochester Conference Series on Public Policy, April 1978), 1979, pp. 9 -5 4 . MARCH/APRIL 1994 110 Charles R. Nelson Charles R. Nelson is professor o f econom ics a t the University o f Washington. Com m entary W ARE INDEBTED TO K. Alec Chrystal and Ronald MacDonald (1994) for assembling a valuable body of evidence on the relative ex planatory pow er of simple-sum and Divisia ver sions of the money supply aggregates across a range of countries. This contribution comes at a time w hen the usefulness of money supply m easures is called into question by economists across the policy spectrum . I am dismayed at the wide agreem ent among macroeconomists ranging from Alan Blinder to Robert Rasche th at the money to income relationship is broken and that our conventional understanding of money dem and is at a loss to e x p la i n the decline in velocity that has occurred during the past decade. Has the quickening pace of financial innovation rendered old relationships obsolete, as m any are suggesting? Somehow this all sounds too familiar. Those of you who w ere around in the 1970s may recall “The Case of the Missing Money.” Then it was a puzzling rise in velocity, and one explanation put forw ard was the quickening pace of financial innovation (see: Enzler, and others, 1976; Goldfeld, 1976; and Ham burger, 1977). Economists, neverthe less, continued to think that m onetary aggregates w ere im portant, enough so th at they w ere dis appointed again a decade later w hen their models seemed to go off track. Even the “m onetarists” are in disarray among themselves on the issue of w hich aggregate to watch. At a Federal Reserve Bank of San Fran cisco conference last spring, the Bank’s presi dent, Robert Parry, quipped that Milton Friedm an FEDERAL RESERVE BANK OF ST. LOUIS had told him that M2 was growing m uch too slowly while Allan M eltzer had told him that M l was growing m uch too fast, so he figured that m onetary policy m ust be just about right. In the face of that kind of disagreem ent, it becomes difficult at best to explain to skeptical colleagues or the public w hy they should take any m one tary aggregate seriously as an indicator of m onetary policy. But as one of the few teachers of introductory macroeconomics (perhaps the only one) who bases their course on the Quanti ty Theory of Money rath e r th an the Keynesian Expenditure Model, I can’t afford to take such a pessimistic view. The first and perhaps prim ary set of results presented in this paper uses regressions of the grow th rate of nominal income on the grow th rate of a money aggregate, the grow th rate of federal spending on goods and services, and, in the case of the United States, the change in the yield on Treasury bills. Lags of zero to four quarters are included. A second set of results adds four lags of the dependent variable to the regression. Test statistics com pare each simplesum (SS) aggregate w ith its Divisia (D) counter part. The Akaike Inform ation Criterion (AIC) statistic com pares the likelihoods of the two regressions, and tw o other tests each produce a pair of /-statistics, one that can reject SS in favor of D and another that can reject D in favor of SS. Finally, F-statistics for the exclusion of all the money term s in each regression address the question, “Does money m atter at all?” In thinking about these regressions, I found it 111 useful to keep in mind the distinction betw een a money dem and equation and a reduced form equation by Leonall C. A ndersen and Jerry L. Jordan (1968) and Leonall C. Andersen and Keith M. Carlson (1970) in their landm ark papers. A simple money dem and equation might be of the form: (1) M = K(i) Y e£ , w here Y is nominal income and K{i ) a function of the nominal interest rate i. Taking logs, denoted by lower case letters, and rearranging we have: (2) y = m - k(i) - t. This is a structural equation and to get to the reduced form w e need a model for the interest rate, something like: (3) i = pe + rigov, real shocks), w here pe is expected inflation and r(gov, real shocks) is the real rate as a function of govern m ent fiscal variables denoted gov, say the deficit as a fraction of GDP, and real shocks which may not be directly observed. In form ing pc, economic agents will presum ably use inform a tion in the past history of m and gov. Substitut ing for pe and then i, we have the reduced form equation: (4) y = y[m with lags, gov with lags) + E(£, real shocks), which is akin to the equation estim ated by Andersen and Carlson. The U.S. regressions run by Chrystal and MacDonald, w hich include an interest rate, are structural and therefore have no obvious role for governm ent spending, while regressions w ithout the interest rate for other countries are in reduced form. In the latter case, I would expect the deficit ra th e r than spending on goods and services to be the m ore appropriate governm ent fiscal variable. I w ant to m ention in passing that there is nothing here that says that velocity m ust be constant, or deterministically trended, or sta tionary, or even cointegrated w ith the interest rate for there to be a useful and predictable relationship betw een money and income. The e rro r processes £ and E may be integrated processes like random walks and indeed coeffi cients may also be stochastic processes w ithout destroying our ability to estimate models and make predictions, although the kinds of processes involved will affect the accuracy of predictions and the deterioration of accuracy with forecast horizon. Certainly the fact that velocity did not continue to move along its upw ard tren d of the 1970s does not in itself imply that money-income models are invalid, as some people seem to be saying. Tw enty years ago, John P. Gould and I (1974) noted th at velocity has experienced trend reversals in the past, behaving m uch like a ra n dom walk. Neither the characterization of velocity as a random walk, no r its link to nominal inter est rates should have lead us to expect the veloc ity trend of the 1970s to continue indefinitely. Turning to the results of the U.S. regressions, I am struck by how weak the evidence is for using Divisia in comparisons with the simplesum aggregates. I expected that the superiority of the D versions would increase w ith aggrega tion since the idea is to extract the transactions part of the aggregate. Indeed, according to the Akaike Inform ation Criterion which looks at the difference in log likelihoods, SSM1 is favored over DM1; DM2 has a slight edge over SSM2; and DM3 is strongly favored over SSM3, although the progression fails with L. One reason that I am surprised how small the AIC statistics are for D aggregates is th at in a sense they are al ready fitted to the data. It would be interesting to see a comparison betw een SSM1 and DM2 to see w hether most of the benefits of purging SSM2 are captured by SSM1, which is more readily available to most of us in real time. Similarly, it would be interesting to see direct comparisons betw een the relatively simple Cur rency Equivalent (CE) aggregate proposed at this conference last year by Rotemberg (1993) and the D aggregates. It would be helpful to the reader to have goodness-of-fit m easures and log likelihoods reported for all the regressions so that other comparisons could be made easily. The (-tests give very puzzling results, frequently giving inconclusive results in which each ver sion is rejected, in tu rn , in favor of the other. The character of the results does not change w hen lags of the dependent variable are included. Does money m atter? Does it m atter if money matters? Perhaps less than I might have thought in the context of structural regressions which include an interest rate. Certainly, variation in velocity, proxied by the interest rate, may ac count for a considerable variation in nominal income, so money is not the only m onetary vari M ARCH/APRIL 1994 112 able in the model. Indeed, it is not surprising then that in the results for the U.S. reported in Table 1 SSM2 m atters, in the sense of the F-test for inclusion of that variable, m uch more than SSM1, since the velocity of SSM1 varies m uch m ore than the velocity of SSM2. W hat puzzles me is that DM1 m atters so m uch less than SSM1, in fact not at all, and results are even w orse for DM1A. Has the money-to-income relationship broken down since the early 1980s? It might be in ter esting to see if the regressions using the D aggregates are m ore stable th an those for SS aggregates. For the rem aining countries, the regression does not include the interest rate, so for the non-U.S. countries, we are looking at a reduced form. As explained above, however, I might have expected the fiscal variable to be the budget deficit ra th e r than spending on goods and serv ices. The message I get from these countries overall is th at DM2 w orks better than SSM2, b ut it is not im portant to use the D version of M l. Japan, of course, is a special case. I say “of course” because Japan seems to be different in m any economic studies, a fact often pointed out w ith pride in my experience by Japanese economists. In the case of money aggregates, not only does Divisia not m atter, but nothing about the aggregates m atters. It would be inter esting to see how the time series for Japan differs from the other countries to see w hat ac counts for this result. I suspect it reflects lack of variation ra th e r than lack of a relationship. The unit root tests are of particular interest to me because we have a chance here to compare across countries. It w arm ed my heart to see only one variable th at is apparently stationary in levels less than expected by chance out of the 54 variables if these series w ere all unrelated. And that one variable is a T-bill rate (for Aus tralia), w hich is already first differenced because it is a grow th rate. W hat is perhaps m ore su r prising is how few other variables are station ary in grow th rates. For Australia, stationary inflation and grow th rates for GDP and SSM3 go along w ith stationarity in the level of the Tbill rate. But only 13 of the 54 series are sta tionary in first differences at the .05 level. In deed, countries as seemingly regular as Switzerland have non-stationary grow th rates, and for Japan it is only the T-bill rate that is stationary in first differences. Evidently, we live in an 1(2) world. FEDERAL http://fraser.stlouisfed.org/RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis Chrystal and MacDonald draw on the technol ogy of cointegration to try to detect long-run relationships among the variables. Since the variables are generally 1(2) while the VAR model used for detecting cointegrating vectors is to be estimated in first differences of 1(1) variables, it is grow th rates which become the relevant “levels” for this analysis. The authors report finding one or two cointegrating vectors for all the countries, implying that there is a long-run relationship among grow th rates of the varia bles. The M l aggregates for the United States are an im portant exception. In general, though, we would be missing some long-run inform ation if we looked only at relationships among the stationary second differences of these variables. It would be interesting to see w hat those coin tegrating relationships look like, w hether they resemble a money-demand function or are something quite unexpected. The VAR is then combined w ith the e rro r cor rection term implied by cointegration (where applicable) so that each variable (in turn, the change in the grow th rates of money, real GDP, the deflator and the T-bill rate) is predicted by four lagged values of itself and each of the other variables, as well as by the e rro r correc tion mechanism (ECM). As in so many VAR studies, it tu rn s out th at the strongest predictor is simply the lagged value of the variable being predicted. For the United States, lags of other variables are generally not useful in predicting GDP or the inflation rate, except that the T-bill rate helps to predict—and, in turn, is predicted by—GDP inflation and M. The ECM also helps to predict inflation in the case of the broader ag gregates. As we look across countries, the most striking regularity is the pow er of awkward, unclear lags in predicting each variable. O ther wise there is little regularity in the pattern of results w hich range from Switzerland, w here almost every variable helps to predict every other variable, to the United Kingdom, w here only the ECM seems to m atter for GNP. Why the great differences? If there is one variable that m oney should be able to predict, it is inflation. If the Divisia ag gregates are superior m easures of money, then one might expect them to be superior predic tors of inflation. There is, however, very little difference in the significance of lags of Divisia aggregates verses simple sum, and no clear m ar gin in favor of the form er. However, the ECM also presum ably includes the m oney aggregate, so differences in the contribution of the ECM 113 m ust be attributed to the distinction betw een the aggregates. In the case of Australia, for ex ample, lags of SSM2 are m ore significant in ex plaining inflation than are lags of DM2, but the e rro r correction term that appears in the DM2 equation is m ore significant. Since the two equa tions differ only in the choice of the money ag gregate, one m ust credit DM2 w ith the greater predictive pow er of the ECM in that equation. In fact, lags of the aggregate may not m atter at all given the ECM, and yet the aggregate may be playing an essential role in the ECM. There are m any examples in the tables w here the money aggregate itself is not significant b u t the ECM is. We cannot conclude in these cases that money does not m atter, and for that reason I would not call these causality tests. A nother reason to be cautious in concluding that money does not m atter if the lags of it are not significant is that the VAR is a restrictive fram ew ork in which to detect dynamic relation ships. A few lags of a noisy variable will contain little information if the variable operates with a long lag. The interest rate is a pow erful leading indicator probably because it smoothes m uch of the inform ation contained in the very noisy money-growth series. I think that this limitation of VARs is one of the main reasons why we have learned so little from the large volume of w ork based on them . Perhaps it is time to take seriously again distributed-lag modelling, which allows for differing lag structures on different variables. I would like to conclude with a plea for visual presentation of data. Economists are traditionally afraid to look at their data—it is considered cheating. I find, on the contrary, that plotting the data is an invaluable tool for understanding models, why they w ork or do not work, and how specification might be improved. I am u n comfortable w ith a statistical result that I can not see in the data. Often, plotting the data reveals w hy a relationship we expected to find does not show up in formal tests and w here it has gone off track. In this spirit, I have p re pared a few charts that may be very familiar to many, but which I found helpful in putting in perspective the notion of a long-run relationship betw een money and income. In Figure 1, I have plotted the velocities of M l and M2 along with the T-bill rate. I did not have ready access to the Divisia counterparts. It makes clear the huge difference betw een the stability of the M2 velocity and the great varia tion in M l velocity. Clearly, in a model of the money-income relationship, it will be very im portant to be able to explain the latter but rela tively unim portant to explain the form er. It also makes clear the fact that M l velocity reflects long-term variation in the short-term interest rate but not short-term variation, as Rotemberg (1993) and others have noted. It is by no means obvious to me that the decline in M l velocity since the early 1980s is in any way inconsistent w ith the decline in interest rates. M l velocity and the interest rate are plausibly cointegrated; that is, they appear to track over a long time period, although they move apart over shorter periods. These dynamics are evident in Figure 2, which is a scatter plot with the log of the velocity of M l on the vertical axis and the log of the T-bill rate on the horizontal. There is a clear differ ence betw een the small, short-run response of velocity to a change in the T-bill rate and the large, long-run response. The last several points represent the period since the recession w hen the sharp decline in short-term interest rates has been accompanied by only a modest decline in M l velocity. But it is not clear that this slug gish short-term response is out of line with experience. Since the velocity of M l evidently responds to the T-bill rate with a lag, I have smoothed the T-bill rate by replacing it in Figure 3 w ith the T-bond yield. While the long-term bond m arket may not provide the optimal sm oother for this purpose, it is free and was not contrived. Now the scatter follows a smooth curve and recent experience is indistinguishable from past ex perience, a fact noted by Poole (1988) and others. I fail to see why we should abandon the idea that there is a stable, long-run relationship in levels betw een money, interest rates and nomi nal income. I w onder w hether the substantial changes in param eters associated in this paper with the 1979 change of m onetary regime would hold if the bond rate replaced the bill rate. I do w ant to call your attention to the scatter plot for M2 velocity and the T-bond yield in Figure 4, because this presents m ore of a puzzle in its recent behavior. Keep in mind that we are looking at relatively little variation in the velocity, b u t certainly the bond yield accounts for little of it. Indeed, the recent rise in the velocity of M2 runs counter to the decline in both shortand long-term interest rates. W hat gives? Perhaps it is the beginning of the end for M2 as Higgins MARCH/APRIL 1994 114 Figure 1 Percent Velocity Figure 2 Log of velocity of M1 FEDERAL http://fraser.stlouisfed.org/RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis Log of Treasury bill yield 115 MARCH/APRIL 1994 116 (1992) and others have suggested. My own view is that this is a tem porary phenom enon related to the discovery of equity mutual funds by traditional holders of CDs. Even relatively sophisticated individuals have been explaining to me recently how m utual funds pay 15 percent com pared to only 3 percent at the bank. There is an expected opportunity cost to holding M2 that we do not m easure. My expectation is that M2 velocity will again fall into line after the public is awakened, perhaps rudely, to the fact that m utual fund shares are not CDs. Chrystal, K. Alec, and Ronald MacDonald. “Empirical Evi dence on the Recent Behavior and Usefulness of SimpleSum and Weighted Measures of the Money Stock,” this Review (March/April 1994). Enzler, Jared, Lewis Johnson, and John Paulus. “Some Problems of Money Demand,” Brookings Papers on Econom ic Activity, 1976:1, pp. 261-80. Goldfeld, Stephen M. "The Case of the Missing Money,” Brookings Papers on Econom ic Activity, 1976:3, pp. 683-730. Gould, John P., and Charles R. Nelson. “The Stochastic Structure of the Velocity of Money,” The Am erican Econom ic Review (June 1974), pp. 405-18. Hamburger, Michael. “Behavior of the Money Stock: Is There a Puzzle?” Journal o f M onetary Econom ics (July 1977), pp. 265-88. Higgins, Byron. “Policy Implications of Recent M2 Behavior,” Federal Reserve Bank of Kansas City Econom ic Review (third quarter 1992), pp. 21-36. REFERENCES Andersen, Leonall C., and Keith M. Carlson. “A Monetarist Model for Economic Stabilization,” this Review (April 1970), pp. 7-25. ________ , and Jerry L. Jordan. “Monetary and Fiscal Actions: A Test of Their Relative Importance in Economic Stabiliza tion,” this Review (November 1968), pp. 11-24. FEDERAL http://fraser.stlouisfed.org/RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis Poole, William. “ Monetary Policy Lessons of Recent Inflation and Disinflation,” Econom ic Perspectives (summer 1988), pp. 73-100. Rasche, Robert H. “Monetary Aggregates, Monetary Policy and Economic Activity,” this Review (March/April 1993), pp. 1-35. Rotemberg, Julio J. “Commentary,” this Review (March/April 1993), pp. 36-41. 117 Douglas Fisher and Adrian Fleissig Douglas Fisher is professor o f econom ics a t North Carolina State University. Adrian Fleissig is assistant professor of econom ics at the University o f Texas-Arlington. The authors wish to acknowledge the assistance o f Ron Gallant, Douglas Pearce, Walter Thurman a nd M ichael Belongia, who m ade im portant suggestions for the revision o f this paper. M oney D em an d in a Flexible D yn am ic Fourier E xpenditure S ystem I n WELL-KNOWN SURVEYS of the growing literature on expenditure systems, Deaton and M uellbauer (1980) and Poliak and Wales (1992) describe many of the shortcomings of the exist ing work in this genre. Among the problems they list that inhibit the acceptance of these methods, the ones that seem most critical to us are (1) the failure to link theory to application, (2) im proper aggregation techniques, (3) im pre cise estimation of partial derivatives, (4) the failure of locally integrable models at some data points and (5) the misspecification of the dynamics. We can address several of these problem s by extending the Fourier Flexible Form of Gallant (1981). Most notably, his technique provides global flexibility and arbitrarily ac curate estimates of partial derivatives. In fact, the technique is capable of approximating the unknow n function (an aggregator function, for example) to any desired degree of accuracy. The version of the Fourier model in current use, however, is static in nature, which inhibits its application to time-series data; in particular, studies by Gallant (1981), Ewis and Fisher (1985), and Fisher (1989, 1992), all employ the static model and all produce residuals that are not white noise for each share; see also Barnett, Fisher, and Serletis (1992). This may be due to inadequately modeled dynamics; in fact, there are no examples of a dynamic Fourier in the literature. The task of this paper is to produce and evaluate two dynamic alternatives in the context of the Fourier model. In the traditional literature on consum er choice, the indirect utility function is approxi m ated by a specific functional form in order to obtain expenditure shares and estimates of the im portant own- and cross-elasticities. One might attem pt to estimate a param etric model, of course, but the results of such exercises have not been satisfactory. The chief problem has been model failure, partly related to the choice of specific (nonflexible) functional forms. To finesse this problem, a flexible functional form can be employed in order to estimate the un known indirect utility function. Diewert (1974) defines a flexible functional form as a secondorder approximation to an arbitrary twice con tinuously differentiable fu n ctio n /fc/ at any given point x*; the popular translog is an exam ple. The difficulty, however, is th at this defini tion, and the resulting approximation, fails to impose precision on the partial derivatives of the function. Indeed, it is well-known that away from the point of approximation, the translog can perform quite poorly in its task of tracking the unknow n function. The result is imprecise estimation of the expenditure shares. MARCH/APRIL 1994 118 THE TIME SERIES APPROACH Gallant (1981) developed the Fourier flexible form in order to approximate the unknow n indirect utility function and its first derivatives arbitrarily accurately within a Sobolov norm. The first derivatives are im portant since the expenditure shares are derived by differentia tion. The Fourier model, w ith its global p roper ties, can then provide integrability over a finite region for the estimated model, assuming con vergence. In particular, since integrability n o r mally implies a convex closure over a finite region, one can presum e desirable separability properties for data examined under the Sobolov norm. This contrasts, as noted, with the possible lack of closure on procedures th at provide an approximation only at a single point in the data space; in particular, it contrasts w ith locally integrable models (such as the Tl'anslog). Following Gallant (1981), the static Fourier flex ible form approximation of an indirect utility function h(v) may be w ritten as A J (1) hk(v,d) = a0+b'v+ i v 'C v+ YJ £ ajae ijk'av , a = l j= - J w here C — m a=J ao’ aoa and b are real-valued, and v is a vector of the expenditure-norm alized user costs of the particular assets involved in the exercise (Gal lant, 1981). In this expression the overbar denotes complex conjugation and / is the imaginary number. A multi-index ka, is an n-vector with integer com ponents and is used to denote p a r tial differentiation of the utility function (see the example in section four). The elem ents of a multi index can be considered to be the weights (when multiplied by v) of the norm alized price indexes. In this paper, we produce two versions of the dynamic Fourier expenditure system; these are then com pared w ith the static model in various ways. In section two we briefly discuss the static model before going into considerable detail over w hat we will be calling the "timeseries approach” to making the Fourier model dynamic. This basically follows the lead of Anderson and Blundell (1982, 1983), whose results are both well-known and have been applied in the literature on flexible functional forms (see Serletis, 1991). In section three, we continue w ith a second version of the dynamics, this time involving the construction of the dynamic Fourier utility function. We term this the "dynamic utility function approach." In sec tion four, we present examples of the two dynamic models in order to clarify the ideas and explain the notation. It is here possible to establish clear distinctions betw een the models in the context of the Fourier. In section five, we go over the procedures used to prepare the data, and in section six, finally, we discuss estimates of the two dynamic models th at utilize the U.S. data previously described. We also discuss how the two models perform in com parison w ith their static equivalents. O ur con clusions follow. In an empirical investigation, it is actually m ore convenient to work with a sine/cosine for mulation rath e r than the exponential just w rit ten and so the following form is generally employed: (2) h.iVfO) = u + b'v + — v'Cv A Uoa + 2 E [U,»CO ( K V ~ W Sj ] jaSin VKvH +z a =l j =1 in which C = -Y u ^ a =l (3) y j (v,0) = RESERVE BANK OF ST. LOUIS FEDERAL - E a oa k a k'. a After differentiating equation 2 and applying Roy’s identity, Gallant arrives at the following set of equations: A vP i aoak a a ana aja — a ja, k pnrl a - a J ( u oov ' K + 2 E j ^ jasm (jkv) + wJacos (jk'v)])ki vi = 1 _____________________ ___________________________________________________________________ A J b'v - £ (uo v'ka + 2 'E jlujasm(jk v) + wjacos (jk v)])k'v a=1 j =1 119 for i = 1, ..., n expenditure shares. This system is w hat is estimated w ith a vector of erro r term s appended. Equation (3) can be more compactly expressed as: A(L) = I + A,L + A 2L2 + ... + ALT B (L) = I + B,L + B2L2 + ... + B L q. Consider the following ARMA(1,1): (4) y„ = / M ' Note that we have attached a time subscript in order to emphasize the static nature of the equations. This completes the discussion of the static Fourier Flexible model. Consumption, m onetary and production th e ories use past variables—in the utility function, in the constraints, or by time-series m ethods— to model habit persistence, adjustm ent costs and/or expectations. In a demand systems ap proach, incorporating dynamics in any of these ways complicates the calculation of the restric tions, which still m ust hold. In the following ex ercises we present results for the time-series function and, in section three, for the utility function. We present the models first, including w ith each a discussion of the restrictions, before presenting examples of both. For the time series model, applying an ARMA (p,q) directly to equation (4) is one approach toward modeling the dynamic behavior of the consumer. This approach is taken by Anderson (1980) for the special case w hen f(vt 8) is linear in the expenditure-norm alized prices v, and the param eters 9. He shows that adding up, as the direct result of adopting the ARMA approach, implies four additional restrictions. Anderson and Blundell (1982, 1983) extend the results for the case in which f(v t,6) may be nonlinear in the parameters but linear in the normalized prices v; i.e., f(v t,d) = n(0)v(. W hen applying an Al\MA(p,q) to equation (4), they can extract a term , y n(6)vtq, the gap betw een the shares lagged p periods and normalized prices lagged q periods, representing the long-run structure for a system of simultaneous equations. This approach is not applicable w hen the m atrix tt(0) cannot be ex tracted, as is the case w ith the Fourier flexible functional form; as a consequence, we use an al ternative approach for analyzing the long-run structure. First, an ARMA(p,q) is applied to equa tion (4). The result is: (5) A(L)y, = B(L)f(vt,8). Here, w here L is the lag operator, the term s A(L) and B(L) represent the following distributed lags (6) y, = A*yt] +f(vl, 9) + B*f(vtl,9)+et. As in Anderson and Blundell (1982, 1983), the addingup restrictions require a transform ation A* of A, w here the columns of A* m ust sum to zero, and aJ* = a. - aln for i=l,...,n and V ’ ' t j= l,...,n-l. Similar restrictions for the matrix B, apply. In sum, then, the dynamics appear as lagged shares yM and lagged normalized prices THE DYNAMIC UTILITY FUNCTION APPROACH Individuals are unlikely, generally, to be able to adjust their consum ption plans instantaneous ly. Rather than apply an arbitrary lag to the shares derived from a static optimization exer cise, an attractive alternative is to allow past behavior to affect cu rren t decisions directly through the utility function. We can define the set of past decisions on a commodity to be an np.1 vector of shares (s) that are functions of all past values of v: (7) s = f ( v j fo r r=l,...,n-l. Here, each share depends on its own lagged normalized price and the lagged normalized prices of the rem aining n-1 shares. In this case, the representative consum er’s dynamic indirect utility function can be expressed as (8) U = U(v,s), w here v = P/M and s represents the dynamics. M is total "expenditures" on this class of assets. This is, in effect, a structural approach for obtaining dynamic shares since the dynamics are em bedded in the decision process rath er than appearing as dynamic extensions of the static shares (as in the time-series model). It produces a new version of the Fourier model, accordingly, lb begin with, we will let s = jct l, so that each share depends on its own lagged value as well as on lags from the rem aining n-1 shares. MARCH/APRIL 1994 120 The dynamic Fourier Flexible Form is defined as A (9) g^(z,0J = uo + b'z + |z 'C z + J y ajaeiJk'oz a = l j= -J and A z = C= - Z uo a W a a =l V ", v'. Parallel to equation 2, we may express the model as (10) g^(z,0J = uo + b'z + iz 'C z A ( , +a ? i ' J . " + 2j= j [uj«cos(j k '«z) ~ wi°sin ( k 'az)Il ’ ? j in which A C= - Z uo a W a =l In this formulation, a multi-index is now a 1 by (r+1) (n) vector w ith integer components; in the static case, it was 1 by (n). Here, r is the num ber of lags. The dynamic shares for this problem are obtained by applying Roy’s identity to equation 10: normalized prices. In the dynamic utility func tion model, the dynamics enter only as lagged normalized prices in each of the share equa tions. The dynamic models can be m ore clearly com pared w ith an example, w hich is w hat we now present. Note that we use w hat are term ed "multi-indices” in the process of estimating the Fourier model. This is a notational convenience, as we have explained, for expressing the partial differentiation of the indirect utility function and can be considered as weights (linear combi nations k a 'v ) of normalized prices. In this example we will be looking at four share equations, w ith A =4 and J=1 in the Fouri er model. The multi-indices used for the timeseries approach, assuming an ARMA(1,0), are: ka k-za Ka where k t k-4a1 i 0 1 i 0 K — l k, = 1 , 0 0 (l 0 0 1 t k2 — 1 0) V (v» V» with V - V„ v j A J v,A - £ (uo z 'k a + 2 £ j[ujasm(jk z) + vvocos (jk z)])ki z i (ID y, = * = 1__________ h i ______________________________ n A J Z b.v i ~ Z (uo z 'K + 2 £ i [u»sir# C z) + w^cos (jk z)))k'z a 1= 1 a =\ j= 1 w here i = expressed as ( This can be more compactly 12 ) y, = f l w j O ) . In this model, adding up is guaranteed, and no additional restrictions need to be applied at the estimation stage. EXAMPLES OF THE TWO MODELS In the two models just presented, the dynam ics are captured in quite different ways. For the time-series approach, the dynamics enter in the form of lagged shares and lagged expenditure FEDERAL RESERVE BANK OF ST. LOUIS Note th at V defines the four expenditurenormalized prices. The multi-indices are set up in the same way as in Gallant (1981) and one m ust be careful, w hen taking partial derivatives, to ensure that the corresponding k.a is used. In this example, the first element of each of the multi-indices, zero or one, corresponds to the first element in V this is the normalized price, ; Vu. Since the dynamics are modeled by adding lagged expenditure shares, the dimension of the multi-indices, which only appears in f(v t,6) in equation 5, stays the same w hen one moves from the static to the dynamic time series model. 121 On the other hand, in the dynamic utility ap proach, the inclusion of lagged normalized prices increases the length of each multi-index [see f(v t, vl t,d) in equation 12]; we use the fol lowing eight indices, accordingly: 0 II 1 0 1 0 0 0 0 0 1 1 0 0 0 ’ 0 01 ^3 1 0 1 1 0 0 ’ K 0 0 = 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 ' K = 1 , k 7 = 0 ’ ks = 1 0 1 1 1 1 0 2 3! In this case the vector of normalized prices is is based directly on economic theory. The Divisia index, indeed, is designed to internalize the substitution effects (at constant utility) that arise from relative price changes. In fact, the simple-sum index cannot produce this result u n less the com ponents of the proposed aggrega tion are perfect substitutes. We have reason to believe this is not the case for the m onetary aggregates in common use. Having a satisfactory procedure such as the Divisia does not, however, tell us exactly w hat set of assets to consider or how to group the subsets of the data for efficient estimation. A procedure that is available is the linear NONPAR program of Varian (1982, 1983), which is based directly on the Generalized Axiom of Revealed Preference (GARP). Satisfaction of GARP on a set of data implies that there exists a non-satiated, concave, monotonic utility function across that particular set. Such a set of data, if it exists, can be examined for logical groupings, again using the program NONPAR. If such groupings can be established—that is, if weak separability holds— then, according to the Leontief-Sono definition of separability, the marginal rates of substitution betw een any two commodities in the m onetary index are independent of changes in relative prices outside the m onetary group. This group is then available for (Divisia) aggregation. Z' = lVu'V*>V#V4l’VU-*V2,f-*V3J-l’V4,,JThe first four elem ents of each ka correspond to the static p art of the vector z and the last four elements of each ka to the dynamic elements of z. This separation of multi-indices enables one to test the static against the dynamic utility function because each multi-index has an as sociated parameter. THE CONSTRUCTION OF DIVISIA MONETARY AGGREGATES Most of the studies of money demand in the literature employ m onetary aggregates th at are simple sums of their components (for example, M l = Currency plus deposits) and are constructed essentially w ithout benefit of index-num ber th e ory. While simple-sum aggregation might serve policy makers well w hen interest rate fluctuations are relatively mild, it is at a disadvantage w hen the relative interest rates on the m onetary com ponents fluctuate significantly. A Divisia index is an alternative approach for aggregating data that On the quarterly U.S. data from 1970:1 to 1985:2, Swofford and W hitney (1987) have con structed a set of real per capita m easures of m onetary quantities and a set of related nominal user costs to represent the prices of these quan tities. With M l denoting narrow money (ex cluding the deposits of businesses); OCD, other checkable deposits; SD, savings deposits in finan cial institutions; and STD, small time deposits in financial institutions, they find that the follow ing arrangem ent passes the necessary and suffi cient conditions for the General Axiom of Revealed Preference: U[V(DUR, NONDUR, SERV, LEIS), M l, OCD, SD, STD], Here, the first three items in the equation refer to components of total consumption, while LEIS refers to leisure (evaluated at the wage rate). Note that SD and STD describe vectors of the liabilities of the various financial institutions (for example, SD = small time deposits in com MARCH/APRIL 1994 122 mercial banks, S&Ls, and so on).1 Also, notice th at in the arrangem ent just listed, the con sumption and leisure activities are separable from the financial assets but not the converse. This implies the existence of an aggregate utility function defined across these m onetary entities (for this time and place). Because of the failure to establish a sub grouping of the m onetary assets, it proves necessary to work w ith the following four aggregate commodities: M l, OCD SDCB, SDSL, SDSB, SDCU STDCB, STDTH, STDCU DUR, NONDUR, SERV, LEIS. A1 A2 A3 A4 Here, SDCB and so on are savings deposits at commercial banks, S&Ls, m utual savings banks and credit unions, while STDCB and so on are small time deposits at commercial banks, thrifts and credit unions. Tb attem pt to preserve the economic characteristics of this set of data up to a third-order rem ainder term , Divisia index num bers are constructed from the individual quantities and their associated user costs; these are designated as Al, ..., A4. Note th at M l and OCD are summ ed for convenience; this can be justified by fu rth e r noting that the correlation coefficient betw een the user costs of these two items is .994. Putting all the pieces together, then, we have m onetary data (and user costs) that satisfy an empirical test for revealed preference, we have aggregated the data in a way that is designed to preserve their economic characteristics in the face of changes in relative prices and, finally, we propose to estimate the elasticities using a model w hich can come arbitrarily close to the elasticities implied by the true (but unknown) aggregate indirect utility function known to be defined (by the GARP test) over these entities. Note, especially, that satisfaction of GARP im plies th at there is a firm link betw een the in 1The original variables were supplied by the Federal Reserve and appear in several publications by Farr and Johnson (1985a, 1985b). In this study, the monetary data are employed in per capita real form (where the latter is achieved by deflation with the CPI). SD represents savings deposits in commercial banks, S&Ls, mutual savings banks and credit unions, while STD represents the small time deposits of the same institutions. OCD is other checkable deposits and includes NOW accounts. See Swofford and Whitney’s two papers for more details on the construction of the data. As discussed in Swofford and Whitney (1987, 1988), the L RESERVE BANK O F ST. LOUIS FE D ER A direct utility that is actually estimated and the underlying utility function that actually gener ates these data. EMPIRICAL RESULTS In our empirical work, we compare the results of the estimation of the three systems: the stat ic, the time series dynamic and the utility fun ction dynamic. Because the static theory is nested in each of the two dynamic theories, we present the results in that form. The com pari sons are in term s of the significance of the coefficients, the characteristics of the residuals and the relevance of the dynamic form ulations using the results of the Gallant-Jorgenson (1979) chi-square test. Unfortunately, the two dynamic approaches are not nested, so that we cannot compute a Gallant-Jorgenson test statistic. We do, however, offer a comparison utilizing the other statistics just m entioned. As it tu rn s out, neither model has a clear advantage, although we do prefer the dynamic utility model in view of its economic properties and adequate perfor mance. We also offer some comparisons with earlier work that utilized the static Fourier model over the same data space (Fisher, 1992). Here, there are dramatic differences in the esti m ated elasticities of substitution; we believe the dynamic results (utilizing the estimates from the dynamic utility approach) are considerably more reasonable than the earlier static results. The share equations, w ith the across-equations restrictions, were estimated in the SAS system using PROC MODEL w ith nonlinear seemingly unrelated regression. The results for the dynam ic time-series model appear in T&ble 1. In this table, the Bs correspond to the quadratic term s in the Fourier Flexible Form, the Us and Vs to the Fourier series expansion, and the As to the lagged shares yM . These results describe reasonable fits, with 10 of the 12 adjustment param eters (AJ having tdata were prepared as follows. Each monetary asset is deflated by the consumer price index for urban areas. OCD includes super NOW accounts. The user cost is the concept defined by Barnett (1978). For leisure, the quantity is 98 hours less average weekly hours worked during the quarter (times 52). The wage rate measures the opportuni ty cost of time. The consumption figures are taken from Department of Commerce data that also provides the im plicit deflator for each category. A 10 percent depreciation rate is used in calculating the one-period holding cost of a durable good. 123 Table 1 Time Series Model: Dynamic Fourier Flexible Functional Form Nonlinear SUR sum m ary of residual errors Eqn. DF model DF error SSE MSE Root MSE square Adjust R-square SM1 SM2 SM3 9 9 9 53 53 53 0.00438 0.01650 0.02445 0.0000827 0.0003113 0.0004613 0.00909 0.01764 0.02148 0.915 0.848 0.923 0.902 0.825 0.912 R- Nonlinear SUR param eter estim ates Parameter B1 B2 B3 U01 U11 W11 A11 A12 A13 A14 U02 U12 W12 A21 A22 A23 A24 U03 U13 W13 A31 A32 A33 A34 U04 U14 W14 Estimate Approxim ate standard error “ T ” Ratio Approxim ate prob>|T| 0.175462 0.007578 -0.448512 -0.007955 -0.009762 0.023864 0.419255 0.259118 0.197632 0.190888 -0.014187 0.011554 -0.019444 -1.000371 0.958742 -0.0 3 4 4 8 0 0.766254 0.020581 -0.0 0 2 2 3 5 0.002259 0.572904 -0.381781 0.276142 -0.119175 -0.009973 -0.000714 0.000171 0.08184 0.25542 0.15285 0.02828 0.00710 0.02545 0.07229 0.03873 0.02071 0.04239 0.01355 0.01015 0.00809 0.13602 0.06774 0.03080 0.08967 0.00692 0.00216 0.00146 0.11232 0.07622 0.04249 0.08974 0.00554 0.00236 0.00287 2.14 0.03 2.93 0.28 1.37 0.94 5.80 6.69 9.54 4.50 1.05 1.14 2.40 7.35 14.15 1.12 8.55 2.97 1.03 1.55 5.10 5.01 6.50 1.33 1.80 0.30 0.06 0.0367 0.9764 0.0049 0.7796 0.1752 0.3526 0.0001 0.0001 0.0001 0.0001 0.3000 0.2599 0.0197 0.0001 0.0001 0.2680 0.0001 0.0044 0.3062 0.1272 0.0001 0.0001 0.0001 0.1899 0.0778 0.7638 0.9527 N = 62 Objective = 2.0164 Objective'N = 125.0495 The Aij represent the dynamics. values in excess of 2. Note th at it is the surfaces of (dld)0g*()d and {d2ld^cd^c') g*(z) that one aims to estimate accurately; it is not required that all param eters be significant. The coeffi cients capturing the dynamics tend to be the most significant param eters. We also calculated the autocorrelation and partial autocorrelations for each of the three share equations; the residuals here w ere white noise. In order to compare the dynamic and static results, we apply the Gallant-Jorgenson chi-square test to provide a comparison w ith the static equivalent of the time series model. The test statistic uses the value "objective*N” in the table. For the stat ic model (the estimates are not shown here), the value of this statistic is 527.9597; for the dynam ic it is 125.0495 as shown in the table. The value of the Gallant-Jorgenson statistic is then 527.9597-125.0495 = 402.9102 with degrees of freedom equal to the difference in the num ber of param eters, of 27-15 = 12. This calculation decisively rejects the static model. MARCH/APRIL 1994 124 Table 2 Utility Function Model: Dynamic Fourier Flexible Functional Form Nonlinear SUR sum m ary of residual errors Eqn. DF Model DF Error SSE MSE Root MSE RSquare Adjust R-square SM1 SM 2 SM 3 9 9 9 53 53 53 0.01057 0.01885 0.03803 0.0001995 0.0003556 0.0007175 0.01412 0.01886 0.02679 0.796 0.826 0.880 0.765 0.800 0.862 Nonlinear SUR param eter estim ates Parameter Estimate Approxim ate standard error B1 B2 B3 U01 U11 W11 U05 U15 W15 U02 U12 W12 U06 U16 W16 U03 U13 W13 U07 U17 W17 U04 U14 W14 U08 U18 W18 -0.163039 -1.187580 -1.028066 0.016757 -0.002319 0.027816 0.002500 0.009921 -0.064331 -0.111290 0.061322 0.022699 -0.0 0 8 5 5 9 0.006524 0.001616 -0.006461 0.006416 -0.113939 -0.013340 0.000898 -0.016214 -0.070695 -0.010984 -0.070924 0.123303 0.024230 0.020177 0.38557 0.32865 0.35163 0.03275 0.01498 0.02235 0.02955 0.01450 0.03009 0.03773 0.01864 0.01756 0.01160 0.00869 0.00993 0.01692 0.01338 0.00800 0.01127 0.00873 0.01118 0.01156 0.01223 0.00896 0.00850 0.01013 0.01199 “ T ” ratio 0.42 3.61 2.92 0.51 0.15 1.24 0.08 0.68 2.14 2.95 3.29 1.29 0.74 0.75 0.16 0.38 0.48 1.42 1.18 0.10 1.45 6.12 0.90 7.91 1.45 2.39 1.68 Approxim ate prob>|T| 0.6741 0.0007 0.0051 0.6110 0.8776 0.2187 0.9329 0.4968 0.0372 0.0047 0.0018 0.2023 0.4637 0.4561 0.8713 0.7041 0.6335 0.1604 0.2419 0.9185 0.1530 0.0001 0.3730 0.0001 0.1530 0.0204 0.0982 N = 62 Objective = 2.3732 Objective'N = 147.1362 The dynamic utility model features interaction among the asset choices over time. This charac teristic distinguishes the dynamic utility system from the time series approach. For this model the results are not quite as satisfactory as those just given. They follow in Table 2. Here, the Rsquares are slightly lower, the objective*N statis tic is higher, and there are fewer significant param eters. The static Fourier is nested within the dynamic utility function in term s of the multi-indexes (see section four). Consequently, we analyze the reduction in the residuals due to the dynamic specification (see Gallant, 1981). FEDERAL RESERVE BANK OF ST. LOUIS The residual sum of squares from the dynamic model is less than half the size of those ob tained from the static model. Quite often, the m ethods discussed to this point would be applied to systems of dem and equations, as they are here. While the estimated structural equations themselves might be of in terest, and for the dynamic versions presented here they could be used to generate forecasts, a typical concern is the elasticity of substitution among the assets. W hat the Fourier provides in this connection is precise estimates of a set of 125 own- and cross-elasticities of substitution (and income) at each data point. This can reveal the nature of the substitutability or com plem entari ty among the assets and the time-series behavior of this concept. While we do not wish to explore the fine points of the data set just examined, a fu rth er illustration, because it reveals an im portant characteristic of the dynamic models, is in ord er. For the m ore interactive dynamic utility function model, Table 3 presents the estimates of the Allen partial elasticities of substitution among the four commodity bundles studied here. In the table, Eij is the elasticity of substitu tion betw een Ai and Aj. The Fourier Flexible Form provides a global approximation and hence we can calculate the asymptotic standard errors for each elasticity (Eij) at each point in time and then evaluate the significance of the estimate. The Tij in the table are the corresponding tstatistics for Eij. Here, we show a complete set of substitution elasticities along with their associated f-values. Note that a positive value for the elasticity indi cates substitution, while a negative indicates a complementary relation. Several things stand out in Tkble 3. Most im portantly, the elasticity of substitution between cash and savings assets (E12 in the table) and betw een cash and time deposits (E13) are very precisely estimated at all data points. This was not the case for static estimates published else w here (Fisher, 1992). While we cannot say a pri ori w hat value of the elasticity of substitution is high, an elasticity over unity, as most are in the first column of the table, could be term ed “elas tic.” Note that the result here is that cash and savings accounts are substitutes, as many would expect on the basis of intuition. More provocatively, however, cash and time deposits appear to be "elastic” complements. This spells trouble for a simple-sum M3 defini tion of money, if these results are correct, since the simple-sum approach to aggregation treats all components as (perfect) substitutes. Clearly, we are not in a position to doubt our results. We have adopted a rigorous aggregation-theoretic approach and tied the empirical work to that as closely as our data would perm it. In fact, the very theory we are using can be invoked in our defense: Economic theory does not say w hether commodities will be substitutes or complements in practice. That is, in practice, economic agents decide w hat assets are substitutes and w hat are complements. Our results indicate that they use cash and time deposits as if they are comple ments, at least over this data sample. We also should point out that this is not an unusual finding in this literature (see the survey in Bar nett, Fisher and Serletis, 1992). Another interesting finding, and one that dem onstrates the power of the dynamic ap proach, is that the elasticities shown in Table 3 are m uch m ore stable than those obtained from the static model. For this comparison, we refer to the elasticities produced in the static Fourier from the same data set, as published in Fisher (1992). In Figure 1 we show the results for the substitution relation betw een cash and savings deposits. Note especially that the two series are scaled differently, an adjustm ent necessary be cause the static estimates fluctuate so wildly. While both series are generally positive (indicat ing that they are substitutes), the static esti mates are occasionally negative (although they were not significantly less than zero). This sort of result is not ruled out by economic theory, but is still hard to explain in term s of the known characteristics of these assets. In Figure 2 we present a comparison betw een the results for the static Fourier and the dynam ic utility model w here the form er results are, again, draw n from the earlier study. In this case we compare cash (A 1) and small time deposits (A3), a relation that is consistently that of com plem entarity in T&ble 3. Once again the dynamic elasticities are rela tively constant. In addition, the static elasticities are sometimes positive and sometimes negative (and statistically so, in both cases, at some dates). Clearly, then, the complementary rela tionship betw een cash and small time deposits is clearly established in the dynamic utility func tion results. We note that such results are quite common in this literature (see Barnett, Fisher and Serletis, 1992). CONCLUSIONS In the introduction to this paper, we listed five areas in which existing studies of expenditure systems frequently fall short, in Diewert’s opin ion. Obviously, the innovation of this paper is to convert a static system into a dynamic one; this deals w ith one of his concerns. Diewert is also concerned th at existing studies do not link the theory to the application firmly enough. This we have attem pted to address both by setting out MARCH/APRIL 1994 126 Table 3 Substitution Elasticities: Dynamic Utility Model E12 70:2 70:3 70:4 71:1 71:2 71:3 71:4 72:1 72:2 72:3 72:4 73:1 73:2 73:3 73:4 74:1 74:2 74:3 74:4 75:1 75:2 75:3 75:4 76:1 76:2 76:3 76:4 77:1 77:2 77:3 77:4 78:1 78:2 78:3 78:4 79:1 79:2 79:3 79:4 80:1 80:2 80:3 80:4 81:1 81:2 81:3 81:4 82:1 82:2 82:3 82:4 83:1 83:2 83:3 83:4 84:1 84:2 84:3 84:4 85:1 85:2 T12 E13 T13 E14 T14 E23 T23 E24 T24 E34 T34 1.291 1.276 1.255 1.167 1.163 1.166 1.115 1.101 1.077 1.052 1.010 1.033 0.995 1.041 1.008 0.920 1.152 1.284 1.184 1.231 1.216 1.199 1.190 1.169 1.137 1.107 1.055 1.041 1.030 0.997 0.977 1.003 1.031 0.982 0.994 0.992 1.000 1.067 1.139 1.194 1.299 1.267 1.265 1.296 1.323 1.382 1.329 1.319 1.338 1.395 1.349 1.460 1.479 1.485 1.483 1.483 1.493 1.465 1.470 1.471 1.463 2.983 3.163 3.389 3.669 3.744 3.896 4.050 3.896 4.083 4.269 4.451 4.524 4.672 4.739 4.726 4.791 4.725 4.358 4.615 4.094 3.977 4.134 4.117 4.137 4.249 4.347 4.499 4.487 4.515 4.611 4.651 4.625 4.612 4.718 4.692 4.653 4.632 4.658 4.448 4.103 4.479 4.593 3.983 3.675 3.688 3.333 3.513 3.653 3.774 4.130 4.221 3.505 3.353 3.344 3.396 3.421 3.241 3.392 3.640 3.447 3.319 -0 .2 5 0 -0 .2 5 9 -0 .2 7 0 -0 .2 7 2 -0 .2 7 0 -0 .2 7 8 -0 .2 7 9 -0 .2 6 5 -0 .2 6 6 -0 .2 7 3 -0 .2 8 2 -0 .2 9 7 -0 .3 3 5 -0 .4 4 6 -0 .4 6 5 -0.507 -0 .4 2 6 -0 .3 6 4 -0 .3 9 2 -0 .3 0 6 -0.281 -0 .2 9 4 -0 .2 8 7 -0 .2 8 2 -0 .2 8 6 -0.291 -0 .3 0 3 -0.291 -0 .2 9 5 -0.318 -0 .3 3 0 -0 .3 2 3 -0 .3 2 8 -0 .4 2 0 -0.5 2 7 -0 .5 6 9 -0.5 7 6 -0 .5 4 5 -0.618 -0 .6 3 4 -0 .5 0 3 -0.4 7 2 -0 .5 9 0 -0 .5 8 6 -0 .5 5 2 -0.581 -0 .5 7 3 -0 .5 6 2 -0 .5 3 2 -0 .4 3 5 -0 .3 8 0 -0 .3 2 3 -0 .3 2 8 -0.341 -0 .3 4 2 -0.3 4 7 -0 .3 2 6 -0 .3 8 8 -0.376 -0 .3 2 9 -0 .3 0 6 2.367 2.532 2.751 3.026 3.071 3.216 3.368 3.202 3.382 3.596 3.873 4.058 4.543 5.808 5.944 5.954 5.233 4.063 4.663 3.543 3.352 3.577 3.539 3.551 3.713 3.880 4.207 4.157 4.235 4.586 4.813 4.654 4.605 5.550 6.350 6.378 6.336 6.507 6.911 6.714 5.621 5.429 6.200 5.700 5.332 4.869 5.281 5.401 5.211 4.379 4.034 2.887 2.730 2.768 2.826 2.874 2.581 3.119 3.254 2.814 2.615 0.509 0.445 0.371 0.290 0.273 0.233 0.195 0.236 0.188 0.134 0.065 0.046 -0 .0 4 7 -0.150 -0.194 -0 .3 0 2 -0.074 0.090 -0.010 0.169 0.204 0.162 0.168 0.165 0.136 0.107 0.049 0.056 0.041 -0 .0 2 0 -0 .0 5 4 -0 .0 2 2 -0 .0 0 8 -0.159 -0 .2 7 4 -0 .3 2 9 -0 .3 3 5 -0 .2 6 0 -0.319 -0 .2 8 9 -0.074 -0.0 6 7 -0.164 -0.110 -0 .0 4 0 -0 .0 2 4 -0 .0 5 2 -0 .0 5 4 -0.011 0.102 0.117 0.358 0.410 0.401 0.385 0.373 0.438 0.318 0.287 0.379 0.429 1.394 1.329 1.218 1.107 1.057 0.932 0.832 0.987 0.825 0.623 0.328 0.230 0.261 0.875 1.167 2.083 0.386 0.375 0.049 0.669 0.790 0.659 0.685 0.683 0.588 0.482 0.240 0.273 0.206 0.109 0.299 0.118 0.044 0.944 1.722 2.097 2.119 1.529 1.806 1.456 0.332 0.313 0.707 0.409 0.140 0.071 0.172 0.191 0.038 0.362 0.443 0.951 0.996 0.953 0.940 0.917 0.978 0.775 0.782 0.971 1.067 -0.614 -0.601 -0 .5 9 2 -0.610 -0.610 -0 .5 9 5 -0 .6 0 2 -0 .6 2 5 -0.619 -0 .6 0 5 -0 .5 8 2 -0 .5 6 8 -0.5 2 7 -0.421 -0.412 -0.381 -0.419 -0 .4 4 4 -0 .4 6 0 -0 .5 4 8 -0.5 7 9 -0.5 6 7 -0 .5 7 6 -0.5 8 7 -0 .5 8 8 -0 .5 8 6 -0 .5 7 6 -0 .5 8 6 -0.581 -0 .5 5 5 -0 .5 3 9 -0 .5 4 9 -0 .5 4 6 -0 .4 6 3 -0 .3 5 7 -0.319 -0.313 -0 .3 3 5 -0 .2 3 7 -0.2 1 2 -0.301 -0 .3 5 3 -0 .2 2 4 -0 .2 0 8 -0 .2 2 3 -0.1 3 5 -0.1 9 3 -0.2 1 5 -0 .2 3 8 -0 .3 3 0 -0.416 -0 .4 4 4 -0.414 -0 .3 8 3 -0 .3 8 8 -0 .3 8 2 -0.381 -0.317 -0.3 6 7 -0.413 -0 .4 4 4 1.106 1.164 1.241 1.462 1.490 1.491 1.609 1.643 1.712 1.763 1.820 1.755 1.738 1.359 1.383 1.464 1.228 1.091 1.292 1.353 1.416 1.442 1.471 1.526 1.596 1.654 1.738 1.785 1.797 1.805 1.801 1.772 1.715 1.578 1.232 1.119 1.090 1.083 0.718 0.569 0.751 0.927 0.524 0.427 0.440 0.223 0.364 0.427 0.478 0.688 0.940 0.772 0.671 0.610 0.629 0.622 0.580 0.507 0.643 0.696 0.733 0.146 0.102 0.049 -0 .0 4 7 -0 .0 6 8 -0 .0 8 4 -0.1 2 9 -0.1 2 4 -0.1 6 5 -0.1 9 8 -0 .2 3 6 -0 .2 2 6 -0 .2 3 8 -0.1 2 5 -0.141 -0.176 -0.0 6 7 0.010 -0.079 -0.074 -0 .0 9 4 -0.111 -0.121 -0.1 3 9 -0.1 6 6 -0.1 8 9 -0 .2 2 4 -0 .2 3 9 -0 .2 4 5 -0 .2 5 7 -0 .2 6 2 -0 .2 4 8 -0 .2 2 7 -0 .2 0 2 -0.1 0 4 -0.079 -0.071 -0.051 -0 .0 6 2 0.113 0.128 0.052 0.202 0.268 0.289 0.422 0.328 0.292 0.283 0.207 0.081 0.237 0.324 0.359 0.340 0.339 0.382 0.388 0.296 0.294 0.291 1.153 0.857 0.438 0.491 0.718 0.890 1.480 1.436 1.979 2.450 3.052 2.766 3.071 1.304 1.538 2.098 0.666 0.083 0.811 0.702 0.898 1.085 1.195 1.413 1.761 2.083 2.632 2.827 2.935 3.201 3.253 2.987 2.663 2.430 1.061 0.788 0.716 0.492 0.464 0.782 0.907 0.427 1.035 1.128 1.212 1.295 1.244 1.221 1.286 1.252 0.599 1.454 1.823 1.881 1.832 1.814 1.855 1.744 1.676 1.714 1.733 -1 .2 8 6 -1.291 -1.301 -1 .2 6 9 -1 .2 8 0 -1.3 0 7 -1 .2 9 2 -1 .2 5 6 -1.271 -1 .2 8 2 -1 .2 8 2 -1.314 -1.3 2 0 -1.3 8 3 -1.375 -1 .3 2 3 -1 .4 3 4 -1 .4 6 0 -1.4 3 8 -1.397 -1.374 -1.381 -1.373 -1.361 -1 .3 5 2 -1 .3 4 6 -1 .3 3 3 -1.321 -1.319 -1.316 -1.313 -1.322 -1.3 3 8 -1.3 4 9 -1.3 8 8 14.957 14.973 14.906 14.076 14.219 14.273 13.771 13.693 13.523 13.165 12.547 12.436 11.475 9.225 8.904 8.123 10.784 13.096 12.212 14.327 14.565 14.093 14.101 13.984 13.559 13.145 12.371 12.366 12.199 11.491 10.997 11.359 11.573 9.671 7.894 7.546 7.581 8.127 7.332 7.133 9.679 10.522 7.546 7.396 7.778 7.238 7.567 7.760 8.378 11.357 13.092 15.850 15.791 15.192 15.251 15.054 15.085 12.801 14.393 15.821 16.458 FEDERAL RESERVE BANK OF ST. LOUIS -1 .4 0 1 -1.407 -1.437 -1 .5 2 8 -1.5 9 6 -1.5 4 5 -1.510 -1.6 2 6 -1.679 -1.677 -1.7 8 3 -1.707 -1.6 8 0 -1.6 5 9 -1 .5 7 7 -1.5 0 5 -1 .5 5 8 -1 .5 9 3 -1 .6 3 2 -1 .6 2 2 -1.6 2 8 -1 .6 5 4 -1.695 -1 .6 2 3 -1.579 -1.547 127 Figure 1 Substitution Elasticities Between Cash (A1) and Savings Deposits (A2), 1970-1985 Figure 2 Substitution Elasticities Between Cash (A1) and Time Deposits (A3), 1970-1985 MARCH/APRIL 1994 128 the theory and by dealing w ith two of his fu r th er concerns: aggregating in a consistent fashion and employing a system that provides arbitrarily accurate estimates of the partial derivatives of the system. W hat we did not do, w hich is in his list of concerns, is examine the model at every data point. In our results, the dynamic models derived and estimated appear clearly superior to the (nested) static models. We are not able to com pare the two dynamic form ulations directly, because they are not nested, but we find the statistical perform ance of the time series ap proach to be superior, while the dynamic-utility approach seems better able to capture the eco nomic interactions among the assets studied. Furtherm ore, most of the estimated share equa tions produced white noise residuals, and this is a characteristic that is not shared by the static estimates, w hether of the nested form in this paper or in the earlier (static) Fourier results that we have been using as a benchm ark. For the dynamic utility model, we have produced a set of elasticities of substitution and charted those betw een cash (Ml + OCD) and savings deposits and betw een cash and small time deposits. The form er are shown to be sub stitutes in the dynamic system, and, m ore im portant, to be m uch more stable than static estimates produced in an earlier study. The lat ter are actually complements, although the nega tive elasticity of substitution is generally less than m inus one, a finding th at is not w ithout foundation theoretically. We anticipate that fu rth e r study of the model and/or the U.S. data will provide fu rth e r observations on this phenomenon. REFERENCES Anderson, G. J. “The Structure of Simultaneous Equations Estimation: A Comment,” Journal o f Econom etrics (October 1980), pp. 271-76. ________ , and R. W. Blundell. “ Estimation and Hypothesis Testing in Dynamic Singular Equation Systems,” Econom etrica (November 1982), pp. 1559-72. ________ “ Testing Restrictions in a Flexible Demand System: An Application to Consumers’ Expenditure in Canada,” Review o f Econom ic Studies (July 1983), pp. 397-410. FEDERAL http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis Barnett, William A. “The User Cost of Money,” Econom ics Letters (vol. 1, no. 2, 1978), pp. 145-49. ________ , Douglas Fisher, and Apostolos Serletis. “ Consumer Theory and the Demand for Money,” Journal o f Econom ic Literature (December 1992), pp. 2086-119. Deaton, Angus, and John Muellbauer. Econom ics a n d Con sum er Behaviour. Cambridge University Press, 1980. Diewert, W. E. “Applications of Duality Theory,” in M.D. Intrilligator and D.A. Hendricks, Frontiers o f Quantitative Eco nomics. North-Holland, 1974, pp. 106-206. Ewis, Nabil A., and Douglas Fisher. “Toward a Consistent Es timate of the Substitutability between Money and Near Mo nies,” Journal of M acroeconom ics (spring 1985), pp. 151-74. Farr, Helen, and Deborah Johnson. “ Revisions in the Mone tary Services (Divisia) Indexes of Monetary Aggregates,” Special Studies Paper no. 189. Board of Governors of the Federal Reserve System, 1985a. ________ , a n d ________ “ Revisions in the Monetary Services (Divisia) Indexes of the Monetary Aggregates,” Staff Study no. 147. Board of Governors of the Federal Reserve Sys tem, 1985b. Fisher, Douglas. Money D em and a n d M onetary Policy. Univer sity of Michigan Press, 1989. ________ . “ Money-Demand Variability: A Demand-Systems Approach,” Journal o f Business & Econom ic Statistics (April 1992), pp. 143-51. Gallant, A. Ronald. “ On the Bias in Flexible Functional Forms and an Essentially Unbiased Form: The Fourier Flexible Form,” Journal o f Econom etrics (February 1981), pp. 211-45. ________ , and Dale W. Jorgenson. "Statistical Inference for a System of Simultaneous, Nonlinear, Implicit Equations in the Context of Instrumental Variable Estimation,” Journal of Econom etrics (October/December 1979), pp. 2 7 5-302. Poliak, Robert A., and Terence J. Wales. Dem and System Specification a n d Estimation. Oxford University Press, 1992. Serletis, Apostolos. “The Demand for Divisia Money in the United States: A Dynamic Flexible Demand System,” Jour n a l o f Money, Credit a nd Banking (February 1991), pp. 3 5 -5 2 . Swofford, James L., and Gerald A. Whitney. “ Nonparametric Tests of Utility Maximization and Weak Separability for Consumption, Leisure, and Money,” Review o f Econom ics a n d Statistics (August 1987), pp. 4 58-64. ________ , a n d ________ “A Comparison of Nonparametric Tests of Weak Separability for Annual and Quarterly Data on Consumption, Leisure, and Money,” Journal o f Business a nd Econom ics Statistics (April 1988), pp. 241-46. Varian, Hal R. “ The Nonparametric Approach to Demand Analysis,” Econom etrica (July 1982), pp. 945-73. ________ “ Nonparametric Tests of Consumer Behaviour,” Review o f Econom ic Studies (January 1983), pp. 99-110. 129 Jam es L. S w o fford James L. Swofford is a professor a nd chair o f the Departm ent o f Economics, University o f South Alabam a. Com m entary D OUG FISHER AND ADRIAN FLEISSIG develop and estimate a Dynamic Fourier Expen diture System in an attem pt to meet some criti cisms that have been raised against the literature on expenditure systems. First, I will discuss Fish er and Fleissig's model in term s of their own criteria set forth in their introduction. Then I will discuss their specification in term s of some criteria for an ideal model that Carl F. Christ proposed in his paper at last year’s Federal Reserve Bank of St. Louis Economic Policy Conference. Finally, I will make some general comments about research on money stock m easurem ent. FISHER AND FLEISSIG’S CRITERIA In their introduction, Fisher and Fleissig m en tion five of w hat they feel are the most telling shortcomings of the expenditure system litera ture. The first they m ention is a failure to link theory and application. On one level they have m et this criticism admirably. They chose to use the Fourier form as a flexible form specification that is able to approximate any unknow n in direct utility function. They have also modeled the dynamics in a way that makes economic sense with their dynamic utility function specifi cation. On another level they have not linked theory and application. Their discussion of the elasticities of substitution that they have estimat ed is fairly terse. While they show th at the elasticity of substitution betw een two assets is more stable w ith one dynamic specification, they do not discuss the sign, m agnitude o r eco nomic interpretation of their estimated elastici ties more than in passing. They say that their elasticities are typical for this literature, but is that good or bad? They do not cite specific previous studies nor do they m ention the size or sign of elasticities from other studies. Are we surprised that old consum er M l and other checkable deposits (OCD) and savings accounts are substitutes but old consum er Ml and OCD and small time deposits are complements? These results make sense to me b u t their implications should be explained in the paper. W hat about the size of these elasticities? W hat is their meaning? My view is that other economists may miss the im portance of the expenditure system literature, if those of us doing research using such systems continue to omit a thorough dis cussion of the elasticities that these systems produce and a comparison of these elasticities w ith those produced by previous research. The second shortcoming of the expendi tu re system literature that Fisher and Fleissig address is im proper aggregation over goods. In my view they have handled this problem in a very nice way. Expenditure systems like the Fou rier system are very parameter-intensive, and aggregation over goods is required to make them tractable. Fisher and Fleissig have used both revealed preference results and good judgment about which goods to aggregate. I feel that in estimating systems such as these both are needed. However, I feel I m ust point out th at Fisher and Fleissig have used a revealed preference test for a direct utility function to MARCH/APRIL 1994 ) 130 back up the specification of an indirect utility function. Direct utility function results may be suggestive of the structure of the indirect utility function, but they are not necessarily m ore than suggestive. Fisher and Fleissig's third and fourth problems with the expenditure system literature are imprecise estimations of partial derivatives and use of locally integrable models, for which the first- and second-order conditions do not obtain at some data points. The Fourier System was de veloped to handle these criticisms of other specifications, such as the translog, so Fisher and Fleissig have admirably handled these criti cisms as they set out to do. The last problem Fisher and Fleissig set before themselves to solve is misspecification, often nonspecification, of dynamics in expenditure systems. They model the dynamics w ith two very general specifications. One, the time series model, is statistical in nature. Another, the dy namic utility function model, is consistent with economic theory. My view is th at they are cor rect to model the dynamics in very general ways. Gerald W hitney and I (1994) have found that data in similar categories to those that Fish er and Fleissig have used can only be rational ized by a well-behaved direct utility function w ith some incomplete category adjustm ent wi thin some quarters. But since Fisher and Fleissig are unable to choose betw een their two dynam ic specifications, we cannot yet say that they have correctly specified the dynamics. They have, however, certainly done a better job modeling the dynamics than other researchers in this area. In a sense they have begun the de bate on how to correctly model the dynamics w ithin flexible consum er expenditure systems. In summary, with a couple of reservations, Fisher and Fleissig have done a good job in meeting the criteria they set forth for their model. Next, I tu rn to the question of how their models compare w ith someone else’s criteria for an ideal econometric specification. CARL CHRISTS CRITERIA FOR AN ECONOMETRIC MODEL At last year’s St. Louis Fed conference, Carl F. Christ suggested seven characteristics of an ideal econometric model. I will next examine Fisher and Fleissig's paper in light of this ideal. FEDERAL RESERVE BANK OF ST. LOUIS Christ’s first criterion is th at the estimated model should provide a good description of some interesting set of past data. Certainly, Fish er and Fleissig’s model has been used to inves tigate an interesting issue—money holdings. There are also a reasonable num ber of coeffi cients that are statistically different from zero, and they test and find the residuals of their model are w hite noise. The second criterion that Christ sets out (and one that he stressed) is that the model should be testable against data that w ere not used to estimate it and were not available w hen it was specified. Fisher and Fleissig have not done this. Since their sample ends in 1985, and Fisher and Fleissig have presum ably form ulated their model in recent years, this would be a tough challenge. A model estimated on data th at ends eight years ago could not be expected to predict today’s data very accurately. The new data set collected by the research staff of the St. Louis Fed could be used to estimate a dynamic flexible model, which then could be put to this test over the next few years. Christ’s third criterion, related to his second, is that the estimated model should describe events for at least a few quarters after it was form ulated and estimated. As w ith Christ's se cond criterion, Fisher and Fleissig’s specification cannot be reasonably put to this test. But a specification estimated w ith the St. Louis Fed’s updated data could be. The fourth criterion is that the model should make sense in the light of our knowledge of economics. Of course, the dynamic Fourier specification is flexible w ith respect to arbitrary elasticities, and it also does not generate negative shares. But Fisher and Fleissig’s specification does generate asset pairs th at switch from sub stitutes to complements over their sample. This is a puzzling result that they do not explain. Christ’s fifth criteria is that a simple model is superior to a complex model. Fisher and Fleis sig’s model is not simple, leaving open the possi bility that an otherwise equal b ut simpler model will be found. Of course, this could be said of any specification. This does suggest that some one might w ant to test Barnett’s Asymptotically Ideal Model w ith this type of data since it has similar characteristics to the Fourier model and may be simpler, depending on the form ulation used. 131 The sixth criteria for judging a model is that, other things being equal, a model that explains a wide variety of data is better. Fisher and Fleissig’s model does explain a wide variety of data, but some of it has been aggregated. An argu m ent could be made to estimate this model be fore aggregating the data. But Fisher and Fleissig have used the soundest aggregation techniques in the literature, the model they necessarily used is very parameter-intensive, and the disag gregate data series is of a relatively short du ration. Christ’s seventh and final criterion is that models that nest special cases are preferable. Fisher and Fleissig’s dynamic Fourier models nest the static Fourier and, in that respect, meet Christ’s ideal. Unfortunately, these models do not nest other Flexible Functional forms nor do the dynamic specifications nest each other. Of course, Fisher and Fleissig’s dynamic Fourier flexible functional form does not meet all of Christ’s ideals. Fisher and Fleissig did not, nor would they, claim that it does, and I do not mean to give the impression that they would make such a bold claim for their model. Their model seems to m eet the first and the fourth through the seventh criteria fairly well. Criteria two and three concern the ability of flexible ex penditure systems to predict future behavior, which seems a worthwhile area of investigation to pursue w ith such specifications. For the most part, Fisher and Fleissig’s specifi cations meet their own criteria that they set out to meet, and Christ’s criteria for an ideal specifi cation that they were probably only generally trying to meet. Their paper is an im portant con tribution to a growing literature on economic m onetary aggregates. I want to close w ith a few comm ents on this literature. THE ECONOMIC MONETARY AGGREGATES LITERATURE I feel that Fisher and Fleissig's paper is an im portant contribution to the question of w hat is money. Much of my w ork in this area has in volved nonstochastic revealed preference tests. Not much is known about the power of such tests, and there are doubts about the validity of these tests, so w ork such as Fisher and Fleissig’s showing that per capita behavior is consistent w ith stochastic models is very important. The literature on economic m onetary ag gregates suggests that the aggregates on which the central bank focuses may not be the ones that people use. If people are using one ag gregate and the central bank is controlling another, then stable "policy” may lead to an u n stable price level. Policy in such a situation might be destabilizing, because the public and the central bank are engaged in a two-sided game, w ith each side having a different objective—the m onetary aggregate each uses. This implies that it is im portant for central banks to attem pt to identify w hat the public in their country is using as money. Also, there may not be an economic m onetary aggregate in an area. W hen looking for an eco nomic m onetary aggregate, the question we are really asking is, "Is there a common currency for a particular area?” This area may or may not be a nation state. If there is no economic m onetary aggregate in an area, then, again, "m onetary” policy would not likely lead to predictable results. Finally, there may be multiple economic m one tary aggregates in use. Consumers may be using one aggregate and business another. Controlling both aggregates may be mutually exclusive. In such a case, optimal m onetary policy may re quire minimizing some loss function over the aggregates, w ith each one weighted by how closely related each aggregate is to the price level. REFERENCES Christ, Carl F. “Assessing Applied Econometric Results,” this Review (March/April 1993), pp. 71-94. Swofford, James L. “ Microeconomic Foundations of a Com mon Currency Area,” working paper, University of South Alabama, 1991. ________ , and Gerald A. Whitney. “A Revealed Preference Test for Weakly Separable Utility Maximization with Incom plete Adjustment,” Journal o f Econom etrics (January/ Febru ary 1994), pp. 235-49. MARCH/APRIL 1994 133 William A. B arnett and Ge Zhou William A. Barnett is a professor o f econom ics a t Washington University, St. Louis. Ge Zhou recently received a doctorate in econom ics from Washington University, St. Louis. Research on this project was p a rtia lly supported by NSF grant SES 9223557. We wish to thank William Brainard for his comments, which substantially influenced the final revision o f this paper. Financial Firm s’ P ro d u ctio n a n d Supply-Side M on etary A ggregation U nder D ynam ic U ncertainty . HIS PAPER IS FOCUSED ON the production theory of the financial firm and supply-side m onetary aggregation in the fram ew ork of dy namics and risk. On the dem and side, there has been m uch progress in applying consum er de m and theory to the generation of exact m one tary aggregates and integrating them into consum er demand system modeling.1 However, on the supply-side, m onetary services are produced by financial firms through financial intermediation, and, hence, exact supply-side m onetary aggregation m ust be based upon financial firm output aggregation. Most of the literature on exact aggregation theory is based upon perfect certainty, w hich often is a reasona ble assum ption regarding contem poraneous con sum er goods allocation decisions. Risk, however, is an im portant consideration in modeling the decisions of financial intermediaries. F urther more, that risk not only applies to future prices ’ See Barnett, Fisher and Serletis (1992). 2“ Demand-side” and “supply-side” imply respectively the demand for monetary services by consumers and manu facturing firms, and the production of monetary services by financial intermediaries. Barnett (1987) has shown that con sumer’s demand for money and manufacturing firm’s de mand for money result in the identical aggregation problem, at least in the perfect certainty case. However, supply-side aggregation of produced monetary services and interest rates, b u t also to contem poraneous interest rates and thereby to the contem porane ous user costs of produced m onetary services. In this paper we derive a model of financial firm behavior under dynamic risk, and we find the exact m onetary services output aggregate. We estimate the Euler equations that comprise the first-order conditions for optimal behavior by financial firms. Barnett (1978,1980) introduced economic aggregation and index num ber theory to demandside m onetary aggregation by applying Diewert’s (1976) results on superlative index num bers. The proposed Divisia index in B arnett’s w ork is an elem ent of Diewert’s superlative index num ber class. Analogous to demand-side m onetary aggre gation, Hancock (1985,1987), Barnett (1987), and Barnett, Hinich and W eber (1986) have provided results on supply-side m onetary aggregation.2 They use neoclassical economic theory to model creates uniquely different aggregation problems resulting from the existence of required reserves, which alter the user cost of produced monetary services. For further results regarding demand for monetary services by manufacturing firms, see Robles (1993) and Barnett and Yue (1991). MARCH/APRIL 1994 134 financial firm s’ production, so the existing eco nomic aggregation and index num ber theory are directly applicable. In fact, throughout the litera tu re on applying economic aggregation and in dex num ber theory to m onetary aggregation, researchers usually assume perfect certainty. Exceptions are Barnett and Yue (1991) and Poterba and Botemberg (1987), who generalize to demand-side exact m onetary aggregation under risk. Supply-side m onetary aggregation under risk has not previously been the subject of research. Introduction of dynamics and uncertainty into supply-side m onetary aggregation requires extensions of earlier research in this area. A financial firm ’s portfolio is generally diversified across different investment instrum ents, and the portfolio’s rate of re tu rn is unknow n at the time th at the investment decision is made. Hence, the assum ption of perfect-certainty and single-period modeling is not appropriate. Furtherm ore, super lative index num bers, such as the discrete time Divisia index, have known tracking ability only un d er the assum ption of perfect certainty. In this paper, we develop a dynamic approach to supplyside m onetary aggregation under uncertainty. Historically, the literature on financial inter mediation has produced many diverse models, often linked only weakly w ith neoclassical eco nomic theory and having various objectives. The early view of the creation of money by financial firms, prim arily viewed to be banks, was the deposit multiplier approach. By this theory in its original form, the process of creating money is simply determ ined by the reserve requirem ent ratio. A nother approach is based upon the Miller-Modigliani theorem , which asserts the irrelevance of financial firm s to the real econo my in a setting of a perfect capital m arket. In recent years, many economists have questioned the appropriateness of either of those two very different propositions and attem pts have been made to extend those theories by weakening the underlying assumptions. Another approach is based upon the capitalasset pricing model (CAPM). U nder the assum p tions of th at model, either the financial firm’s portfolio rate of re tu rn is normally distributed or investors have a quadratic utility function de fined over end-of-period wealth. Under either of 3The papers of Tobin (1961) and Brainard and Tobin (1963, 1968) were the first to argue forcefully for the use of micro economics and equilibrium theory in modeling the financial firm. FE D ER A L RESERVE BANK O F ST. LOUIS those assumptions, the financial firm ’s optimal portfolio behavior can be represented by max imizing utility over the portfolio’s expected rate of retu rn and variance. This approach has been useful in modeling the optimal portfolio allocation decision conditionally upon the real resource in puts, w hich are not explained endogenously. A nother im portant approach is represented by Diamond and Dybvig (1983). They apply tradi tional consum ption-production theory and use an intertem poral model subject to privately ob served preference shocks to examine the equi librium betw een banks and depositors. The studies in this tradition have been successful in explaining bank runs. However, banks, serving solely as a production technology to depositors, play only a passive role in that approach. Another approach is represented by Hancock (1985,1987), Barnett (1987), and Barnett, Hinich and Weber (1986). They treat the financial inter m ediary in the same m anner as a conventional production unit and use neoclassical firm theory to model a financial interm ediary’s production of output services and employment of inputs subject to the firm’s technological feasibility con straint.3 This approach fully models the role played by financial firms as producers of m one tary services. Moreover, it provides the needed tools to apply existing economic aggregation th e ory to aggregation over financial firm s’ output m onetary services, which comprise the econo my’s inside money. However, those studies have not developed a dynamic model of financial firm s’ production u n d er uncertainty. This paper provides that difficult extension of financial firm modeling and output aggregation u n d er neoclas sical assum ptions w ith dynamic risk. With the theoretical model of a financial firm's m onetary services production and the derived exact theoretical output aggregate, we estimate the model’s param eters and test for weak separability of output services from factor inputs. We then substitute the param eter esti mates into the weakly separable output aggrega tor function to generate the estim ated exact supply-side m onetary aggregate.4 Tb this end, we develop a procedure for testing weak separability and for estimating the param eters of a flexible functional form specification of bank technology. The estimation is accomplished “Diewert and Wales (1987) and Blackorby, Schworm and Fisher (1986) have illustrated the difficulty of maintaining flexibility, regularity and weak separability simultaneously. 135 through Hansen and Singleton's (1982) general ized m ethod of m om ents approach to estimating Euler equations. O ur empirical results are based upon com mercial banking data. Our evidence indicates th at banks’ outputs are weakly separable from factor inputs in the transform ation function. Moreover, even under uncertainty, the Divisia in dex provides a better approximation to the estimated theoretical aggregate than does the simple-sum or CE index.5 These findings support the existence of a supply-side m onetary ag gregate and the potential usefulness of the Divisia index to aggregate over the weakly separable m onetary assets on the supply side of money markets. The result is a m easure of in side money, in the sense of m onetary services produced by private financial firms. The paper proceeds as follows. In the next section, we construct our theoretical model of m onetary service production by financial firms u n d er dynamic uncertainty. The model reduces to a dynamic stochastic choice problem, for w hich we derive the Euler equations. In the third section, we present our approach to flexi ble param etric specification, weak separability testing and param eter estimation using Hansen and Singleton’s (1982) generalized m ethod of mo m ents estimation. The fourth section formulates the empirical application using banking industry data. The fifth section contains the empirical results, including param eter estimates, weak separability test results, the estimated theoretical aggregate, and the comparison among index num ber approximations to the estimated exact aggregate, w here the index num bers considered include the Divisia, simple-sum and CE indexes. Section G brings together the dem and side with the supply side to investigate the implications of our model in general equilibrium. Section 7 pro vides a graphical illustration of the errors-in-thevariables problem produced by the use of the simple-sum index as a m easure of the m onetary service flow. The final section presents a few concluding rem arks. 5The formula for computing the Divisia index is in Barnett (1980). Further details regarding the data sources used with the index are in Thornton and Yue (1992), who also provide instructions on downloading the data from the Fed eral Reserve Bank of St. Louis’ public electronic bulletin board, called FRED. The formula for computing the CE (“currency equivalent” ) index is in Rotemberg, Driscoll and Poterba (1991). 6See Barnett (1987). 7As used in this paper, portfolio is the sum of all investments. THEORETICAL MODEL In this section, we derive our theoretical model of m onetary services production by financial firms under dynamic uncertainty. Consider a financial firm which issues its own liabilities and reinvests the borrow ed funds in prim ary finan cial markets. In this process, real resources such as labor, m aterials and capital are used as fac tors of production in creating the services of the produced liabilities. Those produced liabili ties are deposit accounts providing m onetary service combinations that would not have exist ed in the economy w ithout the financial firm. The liabilities of the financial firms include, for example, dem and deposits and passbook ac counts, and are assets to the depositors. The value added through the creation of those assets by a financial interm ediary is that firm ’s contri bution to the economy’s inside money services. W ithout the existence of financial firms and the accounts that they create, investors in money m arkets would be limited to the use of prim ary money-market securities as the short m aturity assets in their portfolios. While the produced liabilities of financial firms may not appear to be "outputs” to an accountant looking at the firm’s balance sheet, the produced liabilities of financial firm s are the outputs of the firm s’ production technologies.6 The financial firm ’s profits are made from the interest rate spread betw een the financial firm ’s financial assets (loans) and the firm’s produced liabilities. That spread m ust exceed the real resource costs, in order for the firm to profit from its operation. Let Yt be the real balances of the financial firm ’s asset (loan) portfolio during period t.7 Let Rt be the portfolio rate of return, which is unknow n at the beginning of each period. Financial firms also hold excess reserves in the form of cash, which has a nominal retu rn of zero. The real balance of cash holding is Cr Let y Hbe real balances in the firm ’s z'th produced account type and hit be holding cost per dollar for that liability, w here i= l, ...,/.8 The am ount of the jth real resource used is zjt, and 8The holding cost h:, is defined as hit = rjt + R tk jt. In this for mula, rit is the account’s net interest rate, which is defined such that all the benefits (for example, service charges) and costs (for example, deposit insurance) generated by the borrowed funds have been factored into the interest rate, and R fa is the implicit tax rate on the financial firm from the existence of a reserve requirement on that ac count type. Required reserves are assumed to yield no interest and hence, produce an opportunity cost to the financial firm, since the firm otherwise could have invested the required reserves at a positive rate of return. MARCH/APRIL 1994 136 its price is w y where Let Pt be the general price index, which is used to deflate nominal to real units. All financial transactions are contracted at the beginning of each period, b u t interest is paid or received at the end of the period. The cost of employing resource zjt is paid at the start of the period. The firm ’s variable profit at the beginning of period t in accordance w ith Hancock’s (1991, equation 3.1) formula, is (1) tt, = (1+fl.j yt_p|_ - y I 1 1 pI+cf. 1 1- c ( pt. pt + iP r J i-i " S j-i W jtZr The first two term s in equation 1 represent the net cash flow generated from rolling over the loan portfolio during period t. The third and fourth term s represent the change in the nom i nal value of excess reserves. The fifth term is the net cash flow from issuing produced finan cial liabilities. The last term is total payments for real resource inputs. Portfolio Yt investment, however, is constrained by total available funds, under the assumption that all earnings are paid out as dividends. The relationship is (2) Y,P, = £ ( l -k„) y f . - C f i - t , i-l (4) Max £ ,[ f ] (— )'"M 7 rs)| - f 1+M s.t. Q(yl s y Is, Cs, z ls,..„ z js) = 0 V s > t, w here Et denotes expectation conditional on the information known at time t, /t is the subjective rate of time preference and is assum ed to be constant, U is the utility function, irs is the vari able profit at period s given by equation 3, and Q is the firm’s transform ation function, defining the firm ’s efficient production technology from (5) Q(yl s y ls, Cs, z j =0 V s> t. In accordance with the usual properties of a neoclassical transform ation function, Q is con vex in its argum ents. In addition, the inputs are distinguished from the outputs by the inequality constraints:9 (6) A Q <o, AQ. < o v ; = 1 ,..., J dC, dzj, and (7) 7-1 w here kit is the reserve requirem ent ratio for the i'th produced account type, with 0 < kjt < 1. Rearranging, equation 2 can be seen to state I that total deposits y itPit are allocated to 1= 1 required reserves, excess reserves, investment in loans, and payments for all real resource inputs. Substituting 2 into 1 to eliminate Yt, we obtain the firm’s profit function subject to its balance sheet constraint: (3) 7 , = 2 T subject to the firm’s technology. We fu rth e r as sume the financial firm ’s intertem poral utility function is additively separable. Then, the firm’s maximization problem can be expressed by the following dynamic choice problem: {[<!+*,_.) > o V (= 1 ,..., /. dy„ We also assum e that Q is continuous and second-order differentiable. Substituting equation 3 into 4, we have (8) Max £, [ | ] (— r 'u i j ^ H d + R ^ t t - k .'j = i 1 s =t +ky A } - ^ - iC .- i^ - i- S j-1 + s.t. Q(yls ,..., y Is, Cs, z u W j, r - l Z>, - l ' ) z j = 0 V s > f. 1-1 - < V i )]> V ipt- i+ W 1+ - n . - A - A - l - T , (l+ « ,-l) ,I} w j, t - 1 z j, (-1 • y-1 We assume the financial firm chooses the level of borrow ed funds, excess reserves, and real resource inputs to maximize its expected dis counted intertem poral utility of variable profits, 9See Barnett (1987), Hall (1973) and Diewert (1973). FE D ER A L RESERVE BANK O F ST. LOUIS We now proceed to derive the Euler equations, comprising the first-order conditions, for this stochastic optimal control problem. We use Bellman’s method. To do so, we m ust put the de cision into Bellman’s form, w hich requires iden tifying the state and control variables and determ ining that the decision, stated in term s of those variables, is in the form providing known Euler equation structure. 137 We assume that the financial firm behaves competitively, so that the prices his_t and w _1 are taken as given by the firm. In addition, hjs l and wjs_j are nonstochastic, since they are lagged one period. From the same perfect competition assumption, it follows that Rs, k.s, and P are random processes that are not controllable by the firm. We select as state variables during period s: y. , V i, z , V /. C ,, R „ R , k , h , V i. w. - 1 V j. P —1 and P . We choose yIS V i and , „ 7 zjs V j to be the control variables during period s. 1 J l,S - l J ,S 7 J ' J ,S - 1 J ' s s —1 s —1 s s' is ' 1,S —1 J Define w to be the vector of all of the state variables, and define us to be the vector of all control variables. Let A be the subset of state variables defined by A = (R . k . h , V i, w , V j,Ps). We assume that As follows a first-order Markov process, w ith transitions governed by the conditional distribution function F(As+1|As). Hence, the transition equation for state variables (R „ R , k , h , V i, w. , V P , P ) is implicitly defined by F(As+1|As). The transition equations for y.s l V / and zjs_1 V j are the trivial identities s s J S —1 s' is ' 1 ,8 - 1 S 7 J ,s - 1 sr j ' is S - l7 1 ,8 - 1 7 J ,S - 1 is implicitly determ ined by the transform ation function 5. The objective function in equation 8 is an in finite summ ation of discounted utilities of varia ble profits, starting at period t. Recalling the time shifts appearing in our definition of the state and control variables during period s, we see th at the discounted utility of variable profit at period s depends only on that period's state variables and control variables. By examining the transition equations, it is evident that each state variable is a function of only previous controls and not of previous values of the states. In p a r ticular, if we let g represent the vector of all transition functions, we can rew rite the dynamic decision problem as Ma*- E, [ £ S-t- UJ1 i 1+ M + = g<UJ ' S ^ L 1 S 0) y> = y,v v s This dynamic problem m eets all of the condi tions to be a recursive problem in the Bellman form. Using Bellman’s principle, we can derive the first-order conditions for solving the dynam ic problem 8. The Bellman recursive equation and (10) zjs = zy V s. The role played by these two equations in our application of Bellman’s m ethod follows from the fact that each of the variables in equations 9 and 10 are included both among the control and state variables, although w ith a time shift distinguishing them in each of their roles.10 Hence, with the appropriate time shift in the subscript, equations 9 and 10 can be viewed as connecting together some of the control and state variables. This connection accounts for the function of those equations as transition equa tions. In particular, the left-hand sides can be identified as next-period state variables, while the right-hand sides can be identified as currentperiod control variables. Hence, each of those equations can be interpreted as defining the evolution of a state variable conditionally on a control variable. The transition equation for Cs l v ( » r () = m a x E t [ U l i r t( w t, u , ) ] + v(w l+l) | wt, s.t. w l+ = g(u,)), 1 1+ M w here v(»r() is the optimized value of the objec tive function. The first-order conditions for the Bellman equation are 11 11 L r 37 ( ’ Iatt + -f - 1 H l +H out owt K J I ‘ = o. The functional form of v is unknown. However, since =Q we can use the Benveniste and 3 iv. 10The use of such trivial identities as transition equations (laws of motion) in optimal control and dynamic program ming is not unusual. For example, it is common in optimal growth models to define current capital stock to be a state variable, while next period’s capital stock is defined to be a control, with those state and control variables tied together by a trivial identity. The nontrivial dynamics is found in the objective function of such models. See, for example, Sar gent (1987, p. 24). MARCH/APRIL 1994 138 Scheinkman equations to eliminate ^2L.(wl+ 1)}1 dw t The general form of the Benveniste and [irt: Scheinkman equations is dv , . dU , > 3 tt a—, {w'] = Io-7 , (7r< ^ — {w'' u ‘] aw T ) 9 a§' («V M ,) 1+ M K j]- Since d g ' = 0, the above equation implies 3 »v. , , o1 3v , 4 3(7 , , fies the relevent theoretical regularity conditions w hen the domain of U(7r() is constrained to 37T, — irt+d > 0 j with h constrained to satisfy h > 0. W hen p > 1, absolute risk aversion (ArrowPratt) is decreasing, and w hen p > l , absolute risk aversion is increasing. The power utility function special case is very widely used. Since that functional form exhibits constant relative risk aversion (CRRA), the pow er utility function often is called the CRRA or isoelastic case.13 Differentiating (14) w ith 7r(, we get (12) — (»v,) = — (7 — JL (w,, Ut). r,) OH' 07T OIV (15) — = h dir; l-p Substituting 12 into 11, we get Using equations 13 and 15 along w ith the de fined state variables, control variables and tra n sition equations, we obtain 3 i ,) 9 < > (13) E, —U (7T\ —w>(»V„ U,) 07Tf 01^ ! 3g' 3(7 T— T — (“ ,) T - < 0 1 + fu 3u, 37T( 0 7, T in',f+l' U, J I " 'J = °- (16) E,{Ptkit { h l-p w here p, h and d are three param eters to be estimated. The following useful utility functions are fully nested special cases of the HARA class:1 2 ' \ +d)P~' + p ~ ~ w +fi,) (i-fc,)-(i+ /i„) l+M + A very general specification of utility to represent risk is the hyperbolic absolute risk aversion (HARA) class, defined by ____ T -d)P (14) U(ir) = 1 - P v h 7 , H , P 'l - p tt +d)P = 0 V y., and r 3 Q /3 z h o-i (17) EU’Ji. " ( - 2 - 7 +d f T 'I ' '3Q /3C , l - p ,+1 - (l + fl,)vv, = 0 ^ 7 I+ +d)P_1 T1 l-p V z , j =1 J. a. risk neutrality: p = 1, U(7r,) = /j7rt, b. quadratic: p = 2, U (,Trt) = -(1/2) (-fi7r, + d)2 , c. negative exponential: p = -oo and d= 1, U(irt)= - e - hr', d. power: d=0 and p < 1, (7<7r() = (7rf/p), e. logarithmic: d = p =0, U(tt) = log irr The general HARA specification for LKtt) satis " S e e Sargent (1987) for an excellent presentation of dynamic programming. 12See Ingersoll (1987, pp. 37-40). In case (d) below, imposing the restriction d = 0 alone on equation 14 will not produce the exact form provided for the power function. However, the form acquired subject to that sole restriction is a posi tive affine transformation of the power function. Hence both forms represent the same risk behavior. FEDERAL RESERVE BANK OF ST. LOUIS Equations 16 and 17 are a system of I+J nonlinear equations. Theoretically, from 16 and 17 plus the transform ation function 5, we could solve for {Yu, ■■ y„f C„ ■) z Jt). However, in practice the solution could be produced only numerically, since a closed form algebraic solution rarely exists for such Euler equations. 13See, for example, Barnett and Yue (1991). 139 In the following discussion, w e extend the dynamic decision 8 into the m ore general case incorporating learning by doing technological change. In the econom etric literature on es timating returns to scale in manufacturing, increasing returns to scale usually are found, despite the fact that increasing returns to scale violates the second-order conditions fo r profit maximization. W e believe that a likely source o f this paradox is the potential to confound techno logical change w ith returns to scale, w hen learning by doing technological change exists but is not incorporated within one’s model. Let y t be the vector o f y it fo r all i and z ( be the vector o f zJt fo r all j. We then w rite the maximization problem as (18) May. e \Y\ VjLj. 1+ u s .t. Q (y ^ Cy z ^ j V j ) = 0 V s > t. The appearance o f y s_1 in the transformation function represents learning by doing. Firm technology improves through experience. At the present stage o f this research, w e are not using the learning by doing extension o f our model in our empirical work, so w e only pro vide the Euler equations below, without supply ing the details o f the derivation. Those Euler equations under learning by doing are 3 7( T (7r() ^ (w , u ) (19) E t{ ay, 1 + rd U , Tqi ‘ a i ; 3 4 9lr, , 3 T 7 1 d C l/d y^ TTm d O / a c~ ", dU ,+i' dir, dir. K + * ut J 1 2 dC. + 3Q /dyt , ,3 [/ , , ---— ----- L (Wt, U,) ---- w ,) T ' dCl/dC. ' dir. 11 + dir, a c, - K + i' “ m W = 0 *y, 14While the risk-neutral case is acquired directly by making those substitutions in the original decision problem, the resulting Euler equations are not acquired simply by mak ing those substitutions in the risk-averse Euler equations, 16 and 17. The reason is that a cancellation within the Euler equations that is produced when the rate of discount is the constant, n, does not apply when the rate of dis count becomes the variable, Rt. In particular, after replac ing p with 1.0, and ^ with Rt, it also is necessary to multiply the two terms within equation 17 by 1/(1 +Rt) to get the risk neutral case Euler equations. No such adjust and (20) E, O TTt (ir,) ^ (w t, u ) OZ{ 1 fd U , , dir + - — h - (7 r-+.’ ^ — (yv'^> u< ) +i 1 + / oirt ^ " z ,_i 3 Cl/d C. dir, dir, ~ 3 C, u ,J l) = O Vz, Equations 19 and 20 are generalizations o f (16) and (17). If learning by doing is excluded by imposing dCl/dyt^ = 0, then (19) and (20) reduce to (16) and (17), respectively. In the rest o f the current paper, w e return to the special case o f no technological change. A further nested special case is also interest ing. W e acquire risk neutrality by setting p = 1. As is conventional under risk neutrality, dis counting is acquired objectively by replacing the subjective rate o f time discount, /*, by Rt}4 One reason fo r interest in that special case is that, in general equilibrium theory, the assumption o f complete contingent claims markets combined with perfect competition can be shown, under certain additional assumptions, to produce the conclusion that firm s w ill be risk neutral, even if their owners are risk-averse. The risk aversion o f the owners then is captured within the contingent claims prices, which are taken as given by the firm s’ managers under perfect competition.1 5 W hile this theoretical issue is interesting, w e do not consider it alone to be a convincing rea son to impose risk neutrality on the manage ment o f an industry that behaves in a manner exhibiting clear risk aversion. However, w e are interested in that fact that the Divisia index, along with virtually all o f the literature on index number theory, is produced under the assump tion o f perfect certainty. This fact would suggest that the tracking ability o f such index numbers may degrade as the level o f risk aversion inment is needed within equation 16, since no relevant fac tors cancelled out in the derivation of equation 16. This observation also is relevant to the risk-neutral Euler equa tions 80 and 81 below. 15See, for example, Debreu (1959, ch. 7) and Duffie (1991, section 6.3). Regarding the complications produced by in complete markets, see Magill and Shafer (1991, section 4). MARCH/APRIL 1994 140 creases. Hence, w e produce results both with and without risk neutrality imposed, as a means o f exploring the extent to which the tracking ability o f index numbers is degraded in the risk averse case relative to the risk-neutral case. Under risk neturality, our Euler equations reduce to1 6 R f X - k } - r.( (19') e [ p , ~ v ~ ----- -a + P , ------------------ 1 r ) E\P, ~ l + R. ' l + R, dCl/dC) V yit, and (20') Et[p, R, dCl/d z,, l+R ao/ac. - wj = 0 V zjt, j = l J . The assumption o f perfect competition is itself sufficient fo r the existence o f a representative firm . See Debreau 1959, p. 45, result 1. Hence, the theory acquired from our model can be applied w ith data aggregated over banks.1 7 SUPPLY-SIDE MONETARY AGGREGATION AND A WEAK SEPARABILITY TEST Having form ulated our dynamic model o f financial firm production under uncertainty and having derived the Euler equations, w e can pro ceed to investigate the exact supply-side m one tary aggregates that are generated, if the firm ’s output m onetary services are weakly separable from inputs. Supply-Side Aggregation Most m oney in m odern economies is inside money, which is simultaneously an asset and a liability o f the private sector. Inside m oney pro vides net positive services to the economy, as a result o f the value added that is created by the '^Observe that only one time subscript exists in the riskneutral Euler equations, so that the solution becomes stat ic. Once the nonlinear utility function has been removed from the objective function, the terms with common time subscripts can be grouped together. However, under risk aversion, even under our assumption of intertemporal strong separability, more than one time subscript exists wi thin the utility function for each time period, since both current and lagged t appear as subscripts in equation 3 for each value of profit, r t . Hence, the dynamics found wi thin the objective function of equation 4 cannot be re moved by regrouping terms. 17ln fact, Debreu’s theorem can be used to aggregate over all firms of all types in the economy to produce the ag gregated technology of the country. The representative firm maximizes profits subject to that aggregated technology. However, we use the theorem only to aggregate over the RESERVE BANK OF ST. LOUIS FEDERAL financial intermediation that produces the inside money. In our model, the borrow ed funds that are outputs produced by financial intermediaries are inside money. Inside money may take vari ous form s such as demand deposits, interestbearing checking accounts, small time deposits, and checkable money market deposit accounts. The sum o f the dollar value in such accounts does not measure the services o f inside money, any m ore than the sum o f subway trains and roller skates measures transportation services, since the components o f the aggregate are not perfect substitutes. The aggregation-theoretic exact quantity aggregate does, however, measure the service flow.1 8 Th e procedures involved in identifying and generating the exact quantity aggregates o f microeconomic theory are described in detail by Barnett (1980). The approach necessarily involves tw o steps: identification o f the com po nents over w hich exact aggregation is admissible and determination o f the aggregator function de fined over those components. The first step de termines w hether or not an exact aggregate exists, and the second step creates the exact ag gregate that is consistent w ith m icroeconom ic theory. The second step cannot be applied unless the first step succeeds in identifying a component cluster that satisfies the existence condition. That existence condition, which is the basis fo r the first stage clustering o f com po nents, is blockwise weak separability. In accor dance w ith the definition o f weak separability, a blocking o f components is admissible if and only if the goods in the block can be factored out o f the structure o f an econom y through a sub function. In other words, it must be possible to formulate the economic structure in the form o f a composite function, w ith the goods in the cluster being the sole variables entering into the inner function o f the structure. I f that condition firms in one industry. It should be observed that the ease of aggregation over firms under perfect competition is in marked contrast with the complexity of the theorems on aggregating over consumers. 18See, for example, Blackorby, Schworm and Fisher (1986) regarding the importance of using appropriately aggregated output data from firms. 141 is satisfied, an exact quantity aggregate exists over the goods in the cluster and the aggregator function that produces the exact aggregate over those goods is the inner function within the composite function. Let y = {y ll,...,y iy and * = (C(, z lt,...,z Jt)' w here y is the vector o f the firm ’s outputs and x is the vector o f the firm ’s inputs. The transformation function becomes Q (y,*)= 0 . An exact supply-side aggregator exists over all o f the elements o f y if and only if y is weakly separable from x within the structure o f Q. Mathematically, that statement is equivalent to the existence o f tw o functions H and y 0 such that Q(y,x)=H(y0 (y),x), w here y 0 (y) is a convex function o f y.ia In aggre gation th eory y 0(y) is called the output aggrega tor function. Furthermore, suppose that y0(y ) is linearly homogeneous in y. Under this assump tion, if each y. grows at the same common rate, the theoretical aggregate y n (y) w ill grow at that rate. Clearly, without that condition, y 0(y) could not serve as a reasonable aggregate.2 0 As shown by Leontief (1947a, 1947b), the weak separability condition is equivalent to (2 1 ) = 0 f o r a ll k. d?ck dQ(y,x)/dy.' If a subset o f the components o f y w ere weakly separable from all o f the other variables in Q, then an exact output aggregate would exist only over the services o f that subset o f components and not over the services o f all outputs. I f w e can test fo r the separability structure o f the transformation function and acquire the func tional form o f y 0 (y), when y is weakly separable from x, then w e could estimate the parameters o f y 0(y) to acquire an econom etric estimate o f the exact output aggregate. ,9See Barnett (1987). “ Without linear homogeneity of y0, the exact aggregate would become the distance function, rather than y0, and would reduce to y0, only under linear homogeneity of y0. We do not pursue that generalization in this study, but see Barnett (1980) for details. 21The Divisia monetary aggregate index was introduced by Barnett (1978, 1980). The simple-sum index is the tradition al monetary index acquired by simply adding up the com Although aggregation theory can provide us with the tools to estimate the exact aggregator function, the resulting aggregate is specification and estimator dependent. Alternatively, the literature on statistical index number theory provides nonparametric approximations to aggre gator functions w hen the existence o f the aggre gator can be demonstrated through a weak separability test. Statistical index numbers pro vide only approximations to the theoretical ag gregate, however, and w hen uncertainty exists, little is known about the tracking ability o f statistical index numbers as approximations to the exact aggregates o f microeconomic theory. In this paper w e consider the Divisia, simplesum and CE indexes to explore their abil ities to track the econom etrically estimated exact output aggregate.2 We produce our econometric 1 estimate o f the exact theoretical aggregate, fo r comparison w ith the index numbers, by using generalized method o f moments (GMM) estima tion o f the parameters o f the Euler equations under rational expectations. We do the GMM estimation both under risk aversion and under the imposition o f risk neutrality, to investigate sensitivity o f our conclusions to risk aversion. Flexibility, Regularity and Weak Separability In empirical applications, there are tw o w idely used approaches to testing fo r the weak separa bility condition that is necessary fo r economic aggregation: the nonparametric, nonstochastic approach based upon revealed preference and the statistical, parametric approach.2 Since w e 2 are working from within a parametric specifica tion, the conventional parametric approach to testing the hypothesis is to be preferred. In fact, w e shall see that weak separability w ill be a strictly nested null hypothesis within our para metric specification, and, hence, conventional statistical testing is available immediately. In ad dition, the nonparametric approach, at its cur rent state o f development, is nonstochastic and, hence, has unknown power. ponent quantities without weights. The CE index is the cur rency equivalence aggregate, originated by Rotemberg (1991) and Rotemberg, Driscoll and Poterba (1991). For an alternative interpretation of the CE index as an economic monetary stock index connected with the Divisia service flow, see Barnett (1991). 22See Swofford and Whitney (1987). MARCH/APRIL 1994 142 Restrictive parametric specifications can bias inferences. As a result, flexible functional form s have been developed and are w idely used in current studies. A flexible functional form, by definition, has enough free parameters to approximate locally to the second-order any arbitrary function.2 However, using flexible 3 functional form s creates a new problem. These models, unlike earlier, m ore restrictive models, may not globally satisfy the regularity conditions o f economic theory, including the monotonicity and curvature conditions. It would be desirable to be able to impose global theoretical regularity on these models, but most o f the models in the class o f flexible functional forms lose their flexi bility property, w hen regularity is imposed.2 We 4 use a model that permits imposition o f regulari ty, without compromise o f flexibility. W hile flexibility and regularity are desirable in any neoclassical empirical study, weak separabil ity in some blocking o f the goods is also needed to perm it aggregation over the goods in that block. W e again are presented with the risk o f losing flexibility by imposing a restriction, and in fact imposing weak separability on many flexible functional form s greatly damages the specifica tions’ flexibility. For example, imposing weak separability on the translog function does great damage to its flexibility.2 Because o f the difficul 5 ties in imposing regularity and separability simul taneously without damage to flexibility, parametric tests o f w eak separability have been slow to ap pear and have been applied only to the static, perfect certainty case in which duality theory is available. In our case o f dynamic uncertainty, very little duality theory is currently available. In this section, w e develop an approach that permits testing and imposing blockwise weak separability within a globally regular and locally flexible transformation function that is arising from a dynamic, stochastic choice problem. Our approach uses D iew ert and Wales’ (1991) sym 23The flexibility here is sometimes called Diewert-flexible or second-order flexible. See Diewert (1971). The flexibility ap plies only locally. However, Gallant (1981, 1982) introduced the Fourier semi-nonparametric functional form, which can provide global flexibility asymptotically. Barnett, Geweke and Wolfe (1991) have developed the alternative seminonparametric asymptotically ideal model (AIM), which is globally flexible asymptotically and has advantages in terms of regularity. 24See Gallant and Golub (1984), Lau (1978) and Diewert and Wales (1987). However, if we can choose a model whose regularity region contains the data, then the regularity will be satisfied without imposing additional restrictions. 25See Blackorby, Primont and Russell (1977). Denny and Fuss (1977) propose a partial solution to avoid destroying FEDERAL RESERVE BANK OF ST. LOUIS metric generalized McFadden functional form to specify the technology o f the firm .2 In the dis 6 cussion to follow, w e first specify the m odel’s form under the null hypothesis o f w eak separa bility in outputs. W e then provide the m ore general form o f the m odel that remains valid without the imposition o f weak separability. Using the notations defined previously, if y is weakly separable from x, then Q (y ,x)= H (y 0(y),x). W e further assume that the transformation function is linearly homogenous. Instead o f specifying the form o f the full transformation function Q directly and thereafter imposing weak separability in y, w e impose weak separa bility directly by specifying H(y0 ,x) and y 0(y) separately. W e acquire our weakly separable form fo r Q by substituting y 0 (y) into H(y0 ,x). Since our specifications o f ytt(y) and H(y0,x) are both flexible, it follows that our specification o f Q is flexible, subject to the separability restriction. W e specify H to be the symmetric generalized McFadden functional form (22) Hty^x) = a(y ()+ a'x + - I y^ x ' ] A [^'] / a'x, w ith a '*# 0 , w here a0, a '= (a i; ...,an and A ), consist o f parameters to be estimated. The matrix A is (n + l)x (n + l ) and symmetric. The vector a'=(aj,...,an) is a fixed vector o f nonnegatvie constants.27 Th e division by a 'x in 22 makes H linearly homogeneous in y0 and XTb conform w ith the partitioning o f the vector (y0 ,*'), w e partition the matrix A as w here A n is a scalar, A 1 is a l x n row vector, 2 flexibility. Their approach is to impose weak separability conditions at a point. However, local weak separability is not sufficient for the existence of a global aggregator function. 26Diewert and Wales (1987) alternatively also developed the generalized Barnett model. Although we have not used that model in this study, the generalized Barnett model has been applied to the analogous perfect-certainty case by Barnett and Hahm (1994). Regarding the merits of the generalized Barnett model in testing for weak separability, also see Blackorby, Schworm and Fisher (1986). 27We use the term “ fixed constants” to designate constants that the researchers can select a priori and treat as cons tants during estimation. 143 A 2 is an n x l column vector, and A an nxn 1 symmetric matrix. Since A is symmetric, it fol lows that A 12 =A'2V Let (y*,;«*)#0 be the point about which the functional form is locally flexible. That point is selected by the researcher in advance, in a man ner analogous to the selection o f the point about which a Taylor series is expanded. Since the transformation function is assumed to be linearly homogeneous, the specification in the above form is not parsimonious, and hence, w e further can restrict the model without losing the local flexibility property.2 We therefore impose 8 (23) a 'x * = 1, (24) A ny * + A l2 * = 0, x and (29) y0(y) = b y + ± y 'B y lp 'y , w ith the parameters satisfying (30) p 'y * = 1, (31) y * = b ’y*, and (32) B y * = Ow w here b ’ = ( b t , ... ,bm and the m x m symmetric ), matrix B consists o f parameters to be estimated, p '= (p v -,p m is a fixed vector o f nonnegative ) constants, and ^ * # 0 is the point at which local flexibility o f equation 29 is maintained. Substituting 29 into 28, w e get (25) A ^ y * + A x * = On, w here On is an n-dimensional vector o f zeros. Under 23, 24 and 25, it can be verified that the number o f free parameters in equation 22 equals the minimum number o f free parameters needed to maintain flexibility. Solving 24 and 25 fo r A n and A 12, w e have (26) A ;2 = - Ax*/y0 * (33) Q (y,*) = H{y0(y),x) = a0 ( b 'y + - (p'y) y 'B y ) + a 'x 2 + — (a ’x) x 'A x 2 ~ (y *a 'x ) 1 * 'A x (b 'y + - (P 'y f 'y 'B y ) x 2 and (y * 2 'x ) 1 * 'A x * (b 'y ) a x (27) A n = x * ’ x*/y*2. A Substituting 26 and 27 into 22 yields (28) (P 'y f 'y 'B y f , = afy n+ a 'x + ^ (a 'x) 'x 'A x - (a 'x)~'x*'A x(yJy*) + - (a',*:) 1 x*'Ax*(yJy*f. 2 D iew ert and Wales (1987) have proved that H(y0 ,x), defined by equation 28, is flexible at Oft**)In a similar w ay w e define y 0(y) to be which is a flexible functional form fo r Q ^ ,* ) and satisfies weak separability in outputs. Neoclassical curvature conditions require Q(y,^) and y 0(y) to be convex functions, and neo classical monotonicity requires d O J d y > 0 and dCl/dx^O . D iew ert and Wales (1987), theorem (10) have shown that H(y0 ,x), defined by 28, and v jy ), defined by 29, are globally convex if and only if A and B are positive semidefinite. 28A flexible functional form is parsimonious if it has the mini mum number of parameters needed to maintain flexibility. Diewert and Wales (1988) have acquired the minimum number of parameters needed to provide a second-order approximation to an arbitrary function. If a specification for an arbitrary function with n variables is flexible, it must have at least 1+n+n(n+1)/2 independent parameters. In our case, the linear homogeneity imposes 1+n extra con straints on the first and second derivatives of H, so the minimum number of parameters needed to acquire flexibili ty is reduced by 1+n. MARCH/APRIL 1994 144 For 0(y„*:) to be convex, w e further need If w e evaluate these derivatives at (y * * * ), w e have a 0 ,34) dYu (39) If 34 holds, then Q(y,;t) is globally convex in w hen H(yn,x) is convex in (y0 ,*) and y 0(y ) is convex in y .2 9 If the unconstrained estimates o f A and B are not positive semidefinite symmetric matrices, positive semidefiniteness can be imposed w ith out destroying flexibility by the substitution (35) A = q q ' and 30 W . “’b and (40) dn dx = a. Applying the method o f squaring technique, w e impose on 39 and 40 the monotonicity conditions3 1 (41) 90, * j y (y * , X * ) = a „ b > 0 (36) B = u u ' w h ere q is a low er triangular n xn matrix and u is a low er triangular m x m matrix.3 In estima 0 tion, w e replace A and B by low er triangular matrixes q q ' and uu', so that the function 33 is globally convex if 34 is true. Monotonicity restrictions are difficult to im pose globally. However, w e can impose local monotonicity w ith simple restrictions. D ifferen tiating 33 with respect to (y,*), w e get (37) ^ = a0 [b + i (2(fi'y) 'B y - ( f t ’y ) 2 y 'B y )] P dy 2 - (v *a 'x )~ 'X* A x lb + - (2(/3'y)~'Bv 2 and ^ Q { y * ,x * ) = a<0. Equation 41 assures that the monotonicity condi tions are satisfied locally at {y*,x *)• We have shown that the functional form de fined by equation 33 and restricted to satisfy equations 23, 30-32, 34-36 and 41 is flexible, locally monotone, and globally convex, provided that the assumed weakly separable structure is true. Although w e do not impose global m onoto nicity, w e do check and confirm that m onotonic ity is satisfied at each observation within our data. In the following discussion, w e w ill define a m ore general flexible functional form that does not require weak separability. - ( P ’y ) zp y 'B y )] + (y*za 'x f 'x * A x * [b + i (2(P’y )~ 'B y -(P 'y )~ 2[)y 'B y )] The number o f independent parameters in equation 33 is (b 'jr + I ( p 'y f 'y 'B y ) (42) and (38) an dx a + - [2 (a'x) A x -(a'x ) otX'Ax1 £ . n(n + l) m (m -l) l + n + -------- + m - l + --------- . 2 2 We know that the minimum number o f param eters required to maintain flexibility fo r a linear ly homogeneous function w ith n + m variables is j - [(y0 a 'x) * A x * - (y* a 'x ) \v*ax * 'Ax] (43) l + n+ m + (n + m ) (n + m + 1) (1 + n + m). (b 'y + — (p 'y V 'y 'B y ) 2 ~ ~ (yTi2® '* ) V o * a,X * A x * ( b ' y + ^ (p 'y )~ 1 'B y ): y 29See Diewert and Wales (1991) for the proof. 30See Lau (1978) and Diewert and Wales (1987). 3 See Lau (1978). 1 FEDERAL RESERVE BANK OF ST. LOUIS Subtracting 42 from 43, w e get n (m - l), which is the num ber o f additional parameters that must be introduced into equation 33 to acquire 145 a flexible functional form fo r a general transfor mation function. Let (44) Q (y,x) = H(ya(y),x) + c 'y + y 'C x / (y 'y +A'*), w here y and A are vectors o f nonnegative fixed constants, the vector c ’ = (c t,...,cm and the m xn ) matrix C are new parameters to be estimated, and the division by y 'y + A '* makes Q linearly homogeneous. Because o f the linear hom ogenei ty p rop erty w e have m ore free parameters than needed fo r flexibility and, hence, w e can im pose the following additional restrictions without losing local flexibility: c 'y * = 0, (47) y * ’C = 0 'n , C x *= O m , (54) v (q 'x * ) = Om , and (55) u 'y * = O m. (56) 3 Q 3 7 = a«b + c w here (y *,x*) is the point at which local flexibil ity is maintained. Under equations 45-48, the number o f new free parameters added into 44 is exactly equal to n (m -l). Diew ert and Wales (1991) have proved that the function 44 is a flexible functional form at (y *,**) fo r a general nonseparable transformation function. Global convexity is difficult to impose in this case. However, w e can derive the restrictions fo r local convexity at (y *,¥*). D eriving the Hes sian matrix o f 44 and evaluating at (y * x * ), w e have and (57) a o dx ~ a' As above, w e use the m ethod o f squaring to impose nonnegativity on 56 and nonpositivity on 57. The estimated results then satisfy local monotonicity. Comparing 33 w ith 44, w e see that weak separability o f outputs in 44 is equivalent to: (58) H0 ' c = Om and v m x n = Om x n (49) V 2 (y * * * ) = Q a0 + b b 'x * 'A x * /y*z C -b x * 'A / y * B C ' - A x * b'/y* (53) y * 'v = 0 'n, W e now turn to imposing local monotonicity. Differentiating 44 with respect to (y,x) and evaluating at (y*,jK*), w e have and (48) Using 50-52, w e rew rite 47, 48 and 32 as The function defined by 44 and satisfying 23, 30-31, 45-46 and 50-55 is a flexible function al form fo r a general transformation function at (y *jtr*). In addition, local convexity is satisfied. (45) y'y* + A '** = l, (46) w h ere q and u are low er triangular matrices in troduced fo r reasons described above, and v is an unrestricted m xn matrix. Then V 2 (y *„**) is Q a positive semidefinite symmetric m atrix.3 2 A J* If V 2 Q(y*„*r*) is positive semidefinite, then Q ^ * ^ .* ) js convex (y * x *). Let (50) A = q q ', (51) C = vq', and (52) B = a - ‘ [v ir' + u u '], Note that under the null hypothesis, H0, equa tion 44 reduces to 33. Hence, y is weakly separable from x if and only if H0 is true. We have derived tw o flexible functional forms with appropriate regularity properties. One structure holds in the general case and the other under the null hypothesis o f weak sep arability. W e now are prepared to test weak separability and to estimate the parameters o f the transformation function. The basic tool is Hansen and Singleton’s generalized m ethod o f moments (GMM) estimator. 32See Lau (1978) and Diewert and Wales (1991). MARCH/APRIL 1994 146 Substituting the functional form given by either 33 or 44 into the Euler equations 16 and 17, w e obtain our structural model, which con sists o f a system o f integral equations. A closed form solution to such Euler equations rarely ex ists. However, GMM permits estimating non linear rational expectations models defined in terms o f Euler equations. Hansen (1982) has proved that under very weak conditions, the GMM estimates are consistent and asymptotically normally distributed.3 3 In the GMM fram ework, there are tw o methods o f testing hypotheses.3 The first approach ap 4 plies Hansen’s asymptotic x2 statistic to test for no overidentifying restrictions. We impose the w eak separability restrictions 58 on the flexible functional form 44, estimate the restricted sys tem, and then run Hansen’s test fo r no overiden tifying restrictions. Since 44 reduces to 33 after imposing the weak separability restrictions, w e can substitute equation 33 itself directly into the Euler equations to impose the null fo r testing. If the test o f no overidentifying restrictions is re jected, then the restrictions imposed under the null hypothesis are rejected, w here in our case the null is the weakly separable structure im posed on the transformation function. The second approach to hypothesis testing w ith GMM is based on the asymptotically n or mal distribution o f the GMM parameter estima tors. Let 6 be the vector o f parameters to be estimated in equation 44. Then the GMM esti m ator 6 has an asymptotically normal distribu tion with mean 9 and covariance matrix 2. Let r be an [n ( m - l ) ] x l vector w hich contains all n ( m - 1) independent parameters in the vector c and the matrix v. The hypothesis o f weak separability can be rew ritten now as t = 0 or equivalently as a set o f linear restrictions o f the form (59) S6 = t= 0 , w h ere S is an [n (m -l)]x [(n + m + l)/2] matrix whose elements are all zeros and ones. 33Hansen (1982), Hansen and Singleton (1982), and Newey and West (1987) provide a detailed discussion of GMM estimation. 34See Mackinlay and Richardson (1991). 35Demand deposits consist of checking accounts, official checks, money orders, treasury tax accounts and loan ac counts. Time deposits consist of regular savings, money market deposit accounts, other time accounts, retirement accounts, and certificates of deposit under $100,000. Non FEDERAL RESERVE BANK OF ST. LOUIS From the known asymptotic distribution o f 6, w e have (60) \/f (Sd -SO ) 3 M O,SSS'), w h ere T is the num ber o f observations. Under the null hypothesis, H0: SQ = 0, w e have V f j 2 M O ,S Z S '.), w here t statistic = (61) ( V f t ) '[ S Z S '] 1 ( V r t ) <t> = Sd. W e obtain the follow ing x 2 Although 2 is unknown, w e can replace it by a consistent estimate without changing the asymp totic results. The test is one sided, with the null o f separability rejected if < is large. t> EMPIRICAL APPLICATION Barnett and Hahm (1994), and Hancock (1985, 1987, 1991) have analyzed m onetary service production by the banking industry in detail, under the assumptions o f perfect certainty and neoclassical joint production. The balance sheet o f a bank consists o f fund-providing functions and fund-using functions. The fund-providing functions include demand deposits, time deposits and nondeposit funds.3 The fund-using 5 functions include investment, real estate m ort gage loans, installment loans, credit card loans and industrial loans. In our theoretical model, the sources o f funds are the firm ’s borrow ed funds, and the uses o f funds are the firm ’s port folio. The total available funds on the balance sheet are total assets minus premises and other assets. On the average, demand deposits and time deposits account fo r over 85 percent o f total available funds. The equity capital included in the non-deposit funds can be treated as a fixed factor that does not enter the variable profit deposit funds consist of equity capital, federal funds pur chased, borrowed money, capital notes and debentures, time deposits of $100,000 and over, other money market instruments, and other liabilities. 147 function.3 For these reasons, w e only choose 6 demand deposits and time deposits as borrow ed funds in our model. Tlirning to inputs, excess reserves are total cash balances minus required reserves. Other real resource inputs are labor, materials and capital.3 Capital is treated as 7 fixed, and w e include only variable factors in the transformation function. An obvious direc tion fo r possible future extension o f this research would be the incorporation o f some or all capital as variable factors to produce in fer ences applicable to a longer run perspective than that implicit in our definition o f variable and fixed factors. Using equations 16 and 17, the Euler equa tions are (62) E, iP ,k J T = p » , + d )p-1 w here D t is demand deposits, T t is time deposits, L t is labor input, M ( is materials input, and wlt and w2 are the prices o f labor and materials t respectively. Using the notations in section three, w e can w rite y '= ( D ;,T,) and x '= (C t,L t,M t). If the weakly separable structure o f the trans form ation function is true, then equation 33 is the transformation function. As discussed in sec tion three, the weak separability hypothesis can be tested by applying Hansen’s %2 statistic. The derivatives o f O w ith respect to its argu ments are given by equations 37 and 38. The fixed constants and the center o f the local ap proximation need to be selected before estima tion. We choose /0 =1, y * '= ( 1,1), and **'=(1,1,1) + P ,U ^ [{1 + R, ) {1~ ku ] ~ (1 + /,1.) 0 30/3D,, , h + R, ' ,p-k n —=------ ----------7T . + d) 1= 0, 3 0 / 3 C , -p 1 ,+ 1 (63) E, [P ,k J 1Z p * t+ d )P~l +P . „ (64) 30/3T, . h ,,P-u „ f (— 7 r(+1+ d )p }= o , ' SO/3C, l - p ' ' 3 0/3 C, - ( 1 + R ,)w «] ^ * l+i +d )P ~'}= 0, /X - V i = l, 2, 3 and E, j [ pR 3 0 / 3 M, ' ' 3 0/3 C, 36See Barnett (1987). Equity capital includes preferred and common stocks, surplus, undivided profits and reserves, and valuation reserves. 2, w here t* represents the midpoint observation.3 8 We correspondingly rescale each price by mul tiplication by the midpoint observation. That rescaling o f prices keeps dollar expenditures on each good unaffected by the rescaling o f its quantity. We select the fixed nonnegative constants a. and / . such that ? (67) a = |^|/£] |^| y-i and and (65) (66) y,=y\/y- v /=i, 1 [p R 3 0/3 L t " as the center o f approximation. To locate that center within the interior o f the observations, w e rescale the data about the midpoint obser vation V i = l , 2, 3 (68) pi = l ^ l / J |^| V 1 1, 2, = j-1 w here £ and y are the sample means o f £ and y respectively. Note that a. and satisfy equations 23 and 30, as is required. W ith our data sam- 38The data point at which all quantities are set to unity can be arbitrary, 37Labor includes managerial labor and nonmanagerial labor. Materials include stationery, printing and supplies, tele phone, telegraph, postage, freight and delivery. MARCH/APRIL 1994 148 pie, w e find a^=0.33, a, = 0.35, a^ = 0.32, ^ = 0 .5 8 , and p., = 0.42. Before estimating the independent parameters, w e need only impose the inequality restrictions. Equation 31 implies b2= l - 6 , , and the monoto nicity condition (41) requires b > 0. Hence, it also follows that b <2. Combining these condi tions, w e can replace b1 and b2 by (69) bj = sin'(^) and b., = cos'C^) and estimate ize a0= l. Since = 0, w e also normal The monotonicity condition 41 requires ai < 0, which w e impose by replacing a, by - a f V i= l,2 ,3 , w here V i= l,2 ,3 , are the new parameters to be estimated. The convexity con ditions are imposed by replacing A and B by the low er triangular matrices q q ' and u u ' respectively, w here q and u are EMPIRICAL RESULTS 0 ^21 ^22 0 .^ 1 ^32 *? 3 3 3 <„ 0 7 and n = U„ 0 Equation 32 implies (70) u„ 0 0 11, Solving 70, w e get u2 = - u n and u = 0. Sub ] .lz stituting them into equation 36, w e have (71) B = u~ The prim ary data source is the Federal Reserve’s Functional Cost Analysis (FCA).3 W e got our 9 data from the Federal Reserve Bank o f St. Louis. The data used are the National Average FCA Report, which contains annual data from 1966 to 1990. Hence, there are a total o f 25 observa tions in our annual data. Monthly data is not available from the FCA. From the FCA, w e ac quired banks’ portfolio rate o f return, the net interest rates on demand deposits and time deposits, and the nominal quantity o f demand deposits, time deposits and cash balances.4 The 0 prices and quantities o f labor and materials are aggregate producer prices and quantity indexes from the data in the FCA Report and the Survey o f Current Business.4 The required reserve ratio 1 is from the Federal Reserve Bulletin. The implicit price deflator is the implicit GNP deflator from the Citibank data base. W e deflate the nominal dollar balances o f all financial goods to convert them into real balances. 1 -1 -1 U The above discussion identifies all the in dependent parameters to be estimated in the specification o f the transformation function. They are un, the low er triangular matrix q, and the vector a '= (a 1 ,a3 ,a2 ). 39The Functional Cost Analysis program is a cooperative venture between the Federal Reserve Banks and the par ticipating banks. This program is designed to assist a par ticipating bank in increasing overall bank earnings as well as to improve the operational efficiency of each bank function. 40The net interest rate equals the interest paid minus service charges earned plus FDIC insurance premiums paid. 41See Barnett and Hahm (1994) for a detailed discussion about the aggregation of labor and material. FEDERAL RESERVE BANK OF ST. LOUIS We use the GMM estimator in the TSP main fram e version (version 4.2) to estimate our model. In the disturbances w e allow fo r condi tional heteroskedasticity and second-order moving average serial correlation. Using the spectral density kernels in TSP, our estimated results are robust to heteroskedasticity, auto correlation and positive semidefinite weighting matrix. To use the GMM method, instrumental variables must be selected. We choose as instru ments the constant, the federal funds rate, the discount w indow rate, the composite bond rate (maturities over 10 years), the holding cost o f demand deposits and time deposits, the lagged banks’ portfolio rate o f return, excess cash reserves, and capital. In estimation, w e replace h by h2 to impose nonnegativity o f the resulting h2. That nonnegativity is needed fo r regularity in the definition o f the H ARA class. The GMM parameter estimates, subject to imposition o f weak separability o f outputs from inputs, are reported in Table 1. A ll three para meters in the utility function are statistically 149 Table 1 GMM Estimates Using the HARA Utility Function with Weak Separability in Outputs Imposed_________________________ Param eter h2 p- 1 d M+ 1 k U11 9n 921 931 922 932 933 !1 a2 ®3 Estimate Standard error 0.003 2.330 0.001 1.090 58.382 0.232 0.186 0.418 0.105 0.477 0.120 0.116 0.323 0.436 0.280 0.122 25.625 0.0 44 0.1 6 5 0.201 0.4 18 0.0 78 0.1 0 6 0.0 48 0.101 0 .1 6 2 0 .5 0 5 0 .0 3 5 0.0 58 0.0 38 insignificant at the 5 percent level. As a result o f the very low precision o f those three para m eter estimates, it is clear that w e have in troduced risk aversion in a manner incorporating too many parameters fo r the available sample size. Hence, w e need to restrict HARA to one o f its less deeply param eterized special cases. As observed in the second section, the HARA class reduces to the popular pow er (CRRA isoelastic) utility function. We now test w hether that popu lar special case is accepted. Equation (61) in the third section provides a statistic to test that a set o f parameters is jointly equal to zeros. W hen the set o f parameters in cludes only one element, the x2 test statistic 0, given by equation (61), equals the number o f ob servations multiplied by the square o f the t-statistic o f that parameter. We calculated that 0=0.0033, w hile the critical value is 6.635 at the one percent significance level. Hence, w e cannot reject d=0, and the pow er utility func tion is accepted. W e reestimate the model using that specification. 42Actually only the upper bound imposed on p is required by theory. Hence, if we had found that the lower bound im plied by our substitution was binding, we would have switched to the more sophisticated substitution of 2 -co sh (p) in place of p. But in practice our estimate of p was strictly positive, so we did not have to resort to the in troduction of hyperbolic functions. Furthermore, our imposi tion of nonnegativity on ^ was equally as harmless, since no corner solutions were acquired on that inequality res triction either. In fact, in the HARA case, we did not im pose nonnegativity on n at all, since we got nonnegativity t-Statistic 0.024 0.091 0.012 6.602 290.459 0.555 2.372 3.931 2.178 4.725 0.743 0.230 9.117 7.523 7.448 To impose the inequality restriction 0 < p < 1, which is sufficient fo r regularity o f the power utility function special case, w e replace p by sin2 (pi and estimate p. In addition, to prevent the implausible possibility o f a negative subjective rate o f time discount, w e replace n by Ji2 and estimate jj..*2 The estimated results, subject to imposition o f weak separability o f outputs from inputs, are reported in 'Fable 2.4 All parameters 3 are significantly different from zero at the 5 percent level except fo r J un, and q3 . M onoto i, 3 nicity is necessarily satisfied at since local monotonicity was imposed at that point. W e use the estimated parameters to determine w hether monotonicity is satisfied elsewhere in the sample. Substituting the estimated param eters into equations 37 and 38, w e find that d Q / d y > 0 and 3 Q / 3 * < 0 everyw h ere in the sample. Hence, no violations o f monotonicity oc curred within the sample. Regarding curvature, w e have imposed global convexity on H(y0 and ,x) y 0(y). To verify global convexity o f Q (y,*), w e must check equation 34 at each data point. from our estimates without the need to impose it, and in retrospect it is evident that we could have done the same in the power utility case. 43The instrumental variables are the constant, the federal funds rate, the discount window rate, the composite bond rate (over 10 years), the three-month T-bill rate, the yields on demand deposits and time deposits, the lagged bank’s portfolio rate of return, and capital. MARCH/APRIL 1994 150 Table 2 GMM Estimates Using the Power Utility Function with Weak Separability in Outputs Imposed Parameter Estimate Standard error t-S tatistic P -524.629 0.351 60.692 0.171 0.240 0.461 0.103 0.418 0.093 -0.025 0.330 0.482 0.217 9.410 0.187 0.019 0.283 0.050 0.077 0.018 0.047 0.029 0.412 0.031 0.045 -55.754 1.877 3122.720 0.605 4.821 5.980 5.908 8.958 3.147 -0.062 10.762 10.607 10.836 \ U1 1 P11 Q21 Q31 Q22 Q32 P33 3 1 a2 S3 Differentiating H(y0,x) w ith respect to y0 w e get , (72) dH^ 0.020 mated parameters and fixed constants into equa tion 29 to get X) = a 0-(a 'x V 'x *' Ax/y* (74) y (D ,T ) = 0 .7 6 D + 0 .2 4 T + — [ 1 7 + (a 'x f'x *'A x *y i/y *2 , w h ere y0 is given by equation 29. Substituting the estimated parameters into equation 72, w e find that d H{y0 ,x)l dy0> 0 at every data point. Convexity o f Q is satisfied throughout the sample. The weak separability hypothesis is tested by using Hansen’s yj test fo r no overidentifying restrictions. His test statistic is (73) < = T Q rx e_f , D w here T is the number o f observations, Q is the value o f the objective function, e is the number o f orthogonal conditions, and / is the number o f parameters estimated.4 The calculated statistic 4 is 27.6, w hile the critical value is 41.64 at the 1 percent significance level. We cannot reject the w eak separability hypothesis. Hence, the exis tence o f a theoretical monetary aggregate over the outputs produced by banks is accepted. Substituting the parameter estimate o f £ from , Thble 2 into equation 69, w e obtain b 1 = 0.76 and b2 = 0.24. The estimated theoretical ag gregate then is acquired by substituting the esti 44The value of the objective function is defined as 0 = gN(9)'WNgH(d), where gN(d) is the sample mean of the moment conditions and l/VN is the_ weighting matrix that defines the metric in making gN (8) close to zero in the GMM estimation procedure. RESERVE BANK OF ST. LOUIS FEDERAL 0 ' ' ' '2 ----- 1 L.58D( + 0.42T(J It is important to recognize that this aggregator function should not be used fo r forecasting or simulation outside the region o f the data, and hence its usefulness is limited to research within the sample. W hile w e have confirm ed m onoto nicity within the region o f the data, this aggre gator function is not globally regular fo r all possible nonnegative values o f the variables out side that region. Having our econom etrically estimated theoreti cal supply-side m onetary aggregate, w e now pro ceed to investigate w hether any o f the w ell known nonparametric statistical index numbers can track the estimated exact aggregate ade quately. By converting from p back to p and then computing the degree o f relative risk aver sion, 1 - p , w e find that the degree o f relative risk aversion is l - . 0 7 = .93. Since risk neutrality occurs only fo r zero values o f relative risk aver sion, w e do not have risk neutrality. But there is no currently available theory regarding the tracking ability o f nonparametric statistical in dex numbers w hen risk aversion exists. Hence, our only m ethod o f investigating the tracking 151 ability o f the m ore easily computed nonparametric statistical indexes is to estimate the exact in dex econometrically, as w e just have done, and compare its behavior w ith that o f the statistical index numbers. In this paper, w e compare the estimated theo retical aggregate w ith the Divisia, simple-sum and CE indexes. Rotemberg, Driscoll and Poterba (1991) have found that the growth rate o f the CE index is very volatile w ith monthly data. Hence, they have proposed (see their footnote 11) a m ethod o f smoothing that index’s growth rates by replacing the index’s weights by 13month, centered moving averages. Since w e are using annual data, there already is a form o f smoothing implicit in the data construction. Nevertheless, in addition to computing the annu al contemporaneous CE index, w e compute the smoothed index in accordance w ith the method selected by Rotemberg, Driscoll and Poterba. To parallel the 13-month centered movingaverage smoothing as closely as possible with annual data, w e use a three-year centered m ov ing average. In a sense, our results w ith uns moothed annual data slightly undersmooths relative to Rotemberg, Driscoll and Poterba’s method, w hile the three-year centered moving average oversmooths relative to Rotemberg, Dris coll and Poterba’s method. Nevertheless, as w e shall see, the CE index’s grow th rate remains too volatile. A centered moving average is not de fined at the start and end o f a sample. Hence, a special m ethod is needed to phase in the cen tered moving average at the start o f the sample and phase it out at the end o f the period. For that purpose, w e use the procedure advocated by Rotemberg, Driscoll and Poterba. Figure 1 contains plots o f the levels o f all those ag gregates. Figure 2 contains plots o f their growth rates. W e also separately plot the grow th rate o f each o f the fou r statistical index numbers (sim ple sum, Divisia, unsmoothed CE and smoothed CE), w ith the grow th rate o f the estimated theo retical path superimposed. These plots are given in Figures 3, 4, 5 and 6. W hile no econom etric estimation is needed to compute the Divisia index, it is important on the supply side to incorporate the required reserves implicit tax into the user cost formula, w hen computing the Divisia output index. The usercost form ula is needed to compute the prices o f m onetary services, since the Divisia quantity in dex is a function o f prices as well as quantities. On that subject, also see Barnett and Hahm (1994), Barnett, Hinich and W eber (1986), Han cock (1985, 1987, 1991) and Barnett (1987), w ho derive and supply the user cost o f supplied m onetary services, w hen required reserves yield no interest. The resulting real user-cost price fo r account type i is (l-/c,) R ,-r „ (75) = 1 + fi. M l 1 + R. (76) w here r., is the ow n rate o f return defined in it footnote 8, and w here (77) = R , - r it 1+ R. Th e nominal user cost is P t ^ tjr The second term on the right-hand side o f equation 76 is the discounted implicit tax on banks resulting from the nonpayment o f interest on required reserves. Equation 77 is the same form as the user-cost price paid on the demand side by depositors, w h ere Rt is the benchmark yield on a pure investment asset producing no services other than its ow n yield, so that equation 77 is the discounted foregone interest given up by the depositor in return fo r the services provided by asset type i. Clearly the Divisia index tracks the theoretical aggregate m ore accurately than any o f the other tw o indexes. The smoothed and unsmoothed CE index’s level paths are almost identical to each other, as shown in Figure 1, despite the im provem ent in the perform ance o f the CE index’s growth rate plot after smoothing. Before 1972, the Divisia and estimated theoretical index are almost identical. A fter 1972, a small gap opens betw een them. The CE index almost always underestimates the theoretical aggregate throughtout the sample period, w ith the gap grow ing to be larger after 1980. The simple-sum index always overesti mates the theoretical aggregate, w ith the gap grow ing to be large and remaining large after only a few years. In terms o f levels, the tracking error o f the CE index is smaller than that o f the simple-sum index, especially early in the same period. However, the CE index is much more volatile than the theoretical aggregate, especially from 1979 to 1983. Comparing Figures 5 and 6, w e see that the CE index w ith smoothed weights is less volatile than the unsmoothed in- MARCH/APRIL 1994 152 Figure 1 Levels of Five Monetary Aggregates (parameters of theoretical monetary aggregate estimated with risk aversion permitted) 1.1 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 1966 68 70 72 74 76 78 80 82 84 86 88 1990 Figure 2 Growth Rates of Five Monetary Aggregates (parameters of theoretical monetary aggregate estimated with risk aversion permitted) 0 .1 5 -i------------------------------------------------------------------------------------------------------------------- Estimated theoretical aggregate ----- Divisia index Simple-sum index CE index Smoothed CE index, in which the weights are three-year, centered moving averages 0.1 0.05 0 -0.05 - 0.1 -0.15 - 0.2 -0.25 T— I I— I— I— i— i— i— r — 1966 68 70 FEDERAL RESERVE BANK OF ST. LOUIS 72 74 76 78 80 82 84 86 88 1990 153 Figure 3 Growth Rates of Theoretical Monetary Aggregate and Divisia Index (with risk aversion permitted) Figure 4 Growth Rates of Theoretical Monetary Aggregate and Simple-Sum Index (with risk aversion permitted) MARCH/APRIL 1994 154 Figure 5 Growth Rates of Theoretical Monetary Aggregate and CE Index (with risk aversion permitted) Figure 6 Growth Rates of Theoretical Monetary Aggregate and Smoothed CE Index (with risk aversion permitted) RESERVE BANK OF ST. LOUIS FEDERAL 155 Table 3 GMM Estimates with Weak Separability in Outputs and Risk Neutrality Imposed Parameter 5 un Q11 P21 Q31 cl22 ^32 933 ® 1 a2 33 Estimate Standard error t-Statistic 61.82 0.27 0.18 0.38 0.07 0.44 0.11 0.16 0.33 0.50 0.23 0.005 0.019 0.005 0.022 0.007 0.023 0.063 0.132 0.002 0.003 0.006 11968.80 14.31 32.78 17.01 10.14 19.09 1.68 1.25 162.74 164.39 36.50 , dex, but the volatility still remains larger than that o f the estimated theoretical index. W e could experim ent w ith even m ore smoothing o f the CE index than is advocated by Rotemberg, Dris coll, and Poterba, but w e feel that fu rth er ex perimentation in that direction w ould produce an index having dynamics determined m ore by the ad hoc method o f smoothing than by the theory that produces the index. Furthermore, w e suspect that smoothing adequate to fix the index between 1979 and 1983 would oversmooth elsewhere. Hence, it seems that there is no way that the CE index can track the growth rates adequately throughout the sample. In short, as a measure o f the level o f the money stock, the simple-sum index perform s most poorly, w hile in terms o f grow th rates, the CE index perform s most poorly. In both cases, the Divisia index perform s best. These results are in the accordance with index number the ory, although most o f that theory is available in rigorous form only under the assumption o f perfect certainty. Our weak separability test sup ports the existence o f an inside-money output aggregate in banking, and our plots support the use o f the Divisia index as the best currently available statistical index fo r tracking that out put aggregate. For comparison purposes, w e repeat the above estimation and testing in the special case o f risk neutrality. The Euler equations, 62-65, under risk neutrality becom e4 5 45ln producing equations 80 and 81 as special cases of the corresponding risk-averse Euler equations, recall foot note 14. v . B(l-fcJ-r R dCl/dD, --------------1 (78) E.[P.— ----- ----- ~ + P,---- 5 l + fi( , , , R il- k )- r 1 + Rt 3Q /3C , R, 3 Q /3 T , (79) £,(P,—------ - ---- - + P , ------------------ 1 + /J, , , . R, = o, 1 + Rt 3 Q /3 Ct = 0, dCl/dL, 80 E [ P ---- 1 ----- -------- w j = 0, ' l + R. dCl/dC. and (81) E\P; R. dO/dM , ----------- —w l + R. dd/dc, 2 1 = o. The parameter estimates acquired from GMM estimation under risk neutrality, w ith weak separability in outputs imposed, are in table 3.4 Substituting the parameter estimate o f t, in 6 the risk-neutrality case into equations 69, w e obtain b ] =0.777 and bz = 0.223. The estimated theoretical aggregate then is acquired by sub stituting the estimated parameters and fixed constants into equation 29 to get (82) y jD 'jT j = 0.777D, + 0.223T( 1 r.2752 (D[-T ,)2 2 >-.58D( + 0.427; The value o f the weak separability test statistic, equation 73, is 9.25, w hile the critical value is 21.666 at the 1 percent significance level. We cannot reject the weak separability hypothesis and, hence, the existence o f a theoretical mone46The instrumental variables are the constant, the discount rate, the lagged banks’ portfolio rate of return, excess cash reserves and capital. MARCH/APRIL 1994 156 tary aggregate over the outputs produced by banks again is accepted. Furthermore, m onoto nicity and convexity again are accepted through out the region o f the data. requirements exist. M oney market equilibrium at a fixed contingent state, w hen one or both o f the monetary assets is subject to reserve re quirements, is illustrated in Figure 14. Figures 7-12 provide the risk-neutral plots analogous to those in Figures 1-6 under risk aversion. Imposing risk neutrality produced negligible gain in tracking ability fo r any o f the indexes. Hence, at least w ith this data, risk aver sion does not seriously compromise index num ber theory. In Figure 13, equilibrium is produced by the familiar separating hyperplane. The separating hyperplane simultaneously supports an indiffer ence curve from below and a production possi bility curve from above. Th e axes represent quantities o f each o f the tw o m onetary assets demanded and supplied. Equilibrium in the two markets exists at the mutual tangency o f the separating hyperplane, the indifference curve, and the production possibility curve at a given optimal level o f factor use. In equilibrium, the quantities demanded o f each asset are equal to the quantities supplied at the equilibrium point y e= (ye J- In addition, the gradient vector ^ye to the separating hyperplane produces the equilibrium user-cost prices. Th e vector o f usercost prices paid by the consumer, 0, are equal, in equilibrium, to the vector o f user-cost prices received by the financial intermediary, <r The t > user cost price o f asset type i is defined by equation 77 above. THE REGULATORY WEDGE Although the imposition o f risk neutrality did not improve the tracking ability o f any o f our indexes, the risk-neutral special case does permit especially simple graphical illustration o f equilibrium phenomena through the use o f separating hyperplanes. In particular, w ith risk neutrality and complete contingent claims m ar kets, each consumer maximizes utility and each firm maximizes profits conditionally upon any fixed, realized contingency (i.e., state). Hence, perfect certainty methods o f graphical illustra tion are available in the risk neutral case, with the understanding that the illustration is condi tional upon the realization o f all contingencies. If no regulatory w edge exists betw een the de mand and supply side, a hyperplane separates tastes from technology. But in the case o f com mercial banks, a regulatory w edge does indeed exist. This conclusion follows from the observa tion in footnote 8 that an implicit tax is imposed upon banks through the existence o f non interest bearing required reserves. Hence, the user cost price received by banks fo r the production o f m onetary services differs from the user cost price paid by depositors fo r the consumption o f those services. The difference is the implicit tax. The formulas fo r the user cost prices on each side o f the market fo r produced m onetary serv ices was derived by Barnett (1978, 1980, 1987) and computed by Barnett, Hinich and W eber (1986). The result is most easily illustrated in the case o f an econom y w ith one consumer, w ho consumes all o f the econom y’s m onetary serv ices, one financial intermediary, which produces all o f the econom y’s m onetary services, and two m onetary assets. Equilibrium in the monetary sector o f the economy at a fixed contingent state is illustrated in Figure 13, w hen no reserve RESERVE BANK OF ST. LOUIS FEDERAL W ith factor employment assumed to be set in advance at its optimum, p , the optimum level o f aggregate m onetary service production, y * is defined to be the solution fo r y * to the equation H(y*,x) = 0, w h ere y * = y (y ) and w h ere Q (y ,*) = H(y*,x) = H(y0(y),x), as explained in the sub section above. Hence, Figure 13 is drawn condi tionally upon that fixed setting o f y*, so that the production possibility surface is the set [(y1 2 (y ): y0 y ^ = y * l ( However, the situation is very different, w hen required reserves exist. In that case, tw o d iffe r ent supporting hyperplanes exist in equilibrium. One supporting hyperplane exists fo r the finan cial intermediary, and another exists fo r the consumer. In Figure 14, the line w ith gradient equal to the consumer’s monetary-asset user-cost prices, ♦ , is the consumer’s supporting hyper plane and it is his budget constraint in equilibri um. That line is tangent to the displayed indifference curve in equilibrium. The financial interm ediary’s supporting hyperplane has gra dient equal to the financial interm ediary’s usercost prices, <f. That hyperplane is the financial t > interm ediary’s iso-revenue line, which is tangent to the firm ’s production possibility curve at the equilibrium point. W hile the user-cost price paid by the consumer fo r the services o f asset type z 157 Figure 7 Levels of Five Monetary Aggregates (parameters of theoretical monetary aggregate estimated with imposed risk neutrality) 1966 68 70 72 74 76 78 80 82 84 86 88 1990 Figure 8 Growth Rates of Five Monetary Aggregates (parameters of theoretical monetary aggregate estimated subject to imposed risk neutrality) 0.15 — _ __ _ —— 0.1 0.05 Estimated theoretical aggregate Divisia index Simple-sum index CE index Smoothed CE index, in which the weights are three-year, centered moving averages 0 -0.05 - 0.1 -0.15 - 0.2 -0.25 t— 1966 i— i— i— r 68 70 72 74 76 78 80 82 84 86 88 1990 MARCH/APRIL 1994 158 Figure 9 Growth Rates of Theoretical Monetary Aggregate and Divisia Index (with imposed risk neutrality) Figure 10 Growth Rates of Theoretical Monetary Aggregate and Simple-Sum Index (with imposed risk neutrality) FEDERAL RESERVE BANK OF ST. LOUIS 159 Figure 11 Growth Rates of Theoretical Monetary Aggregate and CE Index (with imposed risk neutrality) Figure 12 Growth Rates of Theoretical Monetary Aggregate and Smoothed CE Index (with imposed risk neutrality) MARCH/APRIL 1994 160 Figure 13 Equilibrium with No Required Reserves Figure 14 Equilibrium with Required Reserves FEDERAL RESERVE BANK OF ST. LOUIS 161 is still defined by equation 77, the user-cost price received by the bank fo r producing those services now is defined by equation 75, which does not equal equation 77 unless no required reserves exist. The equilibrium point is the point y at which the tw o supporting hyperplanes intersect, and the angle between them is the regulatory w edge produced by the implicit reserve requirement tax paid by the financial interm ediary in the form o f foregone interest on required reserves. A t the equilibrium point both markets are cleared, and the consumer is maximizing utility subject to the displayed budget constraint, w hile the finan cial interm ediary is maximizing revenue subject to the displayed production possibility curve. THE ERRORS-IN-THE-VARIABLES PROBLEM This same figure also can be used to illustrate the magnitude o f the errors-in-the-variables problem produced by the use o f the simplesum index as a measure o f the flow o f m one tary services. Figure 15 illustrates the range o f the error on the demand side, w hile Figure 16 does the same on the supply side. The same il lustration could be produced on the supply side by replacing the tw o indifference curves that are convex to the origin with tw o production possibility curves, that are concave to the origin. The conclusion would be the same. In both figures, the hyperplane represents the set a = {(y„y2 y ,+ y 2 = a#J, ): w h ere M s is the measured level o f the simples sum index, w hile A is the set o f possible values o f the m onetary asset component quantities Ovy2 ) are consistent with the measured level o f the simple-sum index. For any such measurement on the simple-sum index, the value o f the demand-side m onetary service flow received by asset holders could be anywhere within the set (83) (u(y„y2 (yt,y2 A }. ): )€ The range o f that set is the gap betw een the utility levels at which the tw o indifference curves are drawn in Figure 15. C learly the upper indifference curve is the one w hich inter sects the hyperplane A at the highest possible utility level, w hile the low er indifference curve is the one which intersects the hyperplane A at the lowest possible utility level. We see that magnitude o f the errors-in-the-variables problem in that illustration, w hen measured by the range o f the set (83), is M m M m . The same conclu aiijn sion is produced on the supply side from Figure 16, but with set 83 replaced by4 7 [y,0vy>>: (y,/y2> e^)The simple-sum m onetary aggregates produce a disturbingly large and entirely unnecessary errors-in-the-variables problem. Figures 15 and 16 illustrate the reason. Figures 1-12 illustrate the effect, under circumstances that are most favorable to the simple-sum aggregates: a low level o f aggregation over assets having similar yields. W ith broader aggregation over assets having very different ow n rates o f return, in cluding currency w ith a zero rate o f return, the continued use o f simple-sum m onetary ag gregates by central banks becomes even more difficult to comprehend. The days w hen all m onetary components had zero-own rates o f return are long gone. CONCLUSIONS In this paper, w e develop a theoretical model o f m onetary service production by financial firms. Earlier models either have perm itted risk, but w ith minimal connection with neoclassical economic th eory or have made full use o f neo classical production theory, but under the as sumption o f perfect certainty. The latter case has been developed extensively by Barnett (1987), Barnett and Hahm (1994), and Hancock (1985, 1987, 1991). We extend that latter fully neoclassical production approach to the case o f risk aversion, subject to D iew ert and Wales’s symmetric generalized McFadden technology Our approach permits risk aversion without compromising second-order flexibility or neo- 47The magnitude of the gap, Mm — Mm may differ, when a ax [n, regulatory wedge is produced by required reserves, but the difference between the conclusions on the demand and supply side is not likely to be large. If the errors-in-thevariables problem is large on one side of the market it is likely to be approximately as large on the other side of the market. See Barnett, Hinich and Weber (1986) for relevant empirical evidence. MARCH/APRIL 1994 162 Figure 15 Demand Side Errors-in-Variables Figure 16 Supply Side Errors-in-Variables FEDERAL RESERVE BANK OF ST. LOUIS 163 classical regularity o f the specification. This is true with or without the imposition o f global blockwise weak separability, which w e therefore are able to test and to impose, w hen accepted. Using the resulting Euler equa tions, w e explore exact output aggregation in this paper. Although applicable to all financial interm edi aries, w e apply our approach only to the bank ing industry. W hile it is possible to impose regularity in curvature conditions upon the generalized McFadden specification, monotonicity can be imposed only locally without damaging the models flexibility. Diew ert and Wales’s alter native specification, called the generalized Bar nett model, is globally regular both in terms o f curvature and monotonicity, and hence, that model was used by Barnett and Hahm (1994) in the perfect certainty case. However, in the cur rent paper, using the generalized McFadden model, the estimated parameters satisfy the neo classical monotonicity and convexity conditions fo r all observations, even though only convexity was imposed globally. Hence, w e doubt that our conclusions would have been much different if w e had used the generalized Barnett model in producing our estimated Euler equations. The hypothesis that bank’s outputs are weakly separable from inputs is accepted. Hence, the existence o f an exact supply-side theoretical m onetary aggregate is accepted fo r banks. The resulting output aggregate is the banking indus try ’s contribution to the economy's inside money services. W hile our theory provides a means o f econometrically estimating the exact supply-side m onetary aggregate, no theory currently is avail able to support the use o f a nonparametric statistical index number as an approximation to the parametrically estimated exact aggregate. Considering the complexities o f the GMM esti mation involved in producing the estimated exact aggregate, a nonparametric statistical index would, in practice, be much easier to compute and use. W e compute the currently most popular o f those indexes and find that at least fo r our sample, the Divisia index tracks the estimated theoretical index m ore accurately than the others. This conclusion holds regardless o f w hether or not w e impose risk neutral ity during estimation o f the exact theoretical aggregate. Risk aversion does not appear to produce appreciable degradation o f the tracking ability o f the Divisia index w ith our data. W e believe that the approach developed in this paper could be used to investigate technological change in banking, economies o f scale and scope in banking, value added in banking and its connection w ith inside m oney creation, and the transmission mechanism o f m onetary policy. We have in fact taken a first step in the direc tion o f producing one o f those extensions: We have derived and supplied the Euler equations with learning by doing technological change in cluded in technology. A longer-run fram ew ork fo r the theory also could be productive. In par ticular, some o f the factors excluded from the variable cost function as fixed factors could be incorporated among the variable factors. Bank capital is one such example. Incorporating capi tal among the variable factors could perm it in tegration o f the m odel w ith economic growth theory, in which capital evolves endogenously in accordance w ith a law o f motion. In short, this is just a start in a direction that w e expect w ill be very productive fo r researchers interested in the role o f financial institutions in the economy. REFERENCES Barnett, William A. “ A Reply to Julio Rotemberg,” in Michael T. Belongia, ed., Monetary Policy on the 75th Anniversary of the Federal Reserve System. Kluwer Academic Publishers, 1991, pp. 232-43. _______ . “ The Microeconomic Theory of Monetary Aggrega tion,” in William A. Barnett, and Kenneth J. Singleton, eds., New Approaches to Monetary Economics. Cambridge University Press, 1987, pp. 115-68. ________‘‘New Indices of Money Supply and the Flexible Laurent Demand System,” Journal of Business and Eco nomic Statistics (January 1983), pp. 7-23. _______ . “ The Optimal Level of Monetary Aggregation,” Journal of Money, Credit, and Banking (November 1982), pp. 687-710. ________“ Economic Monetary Aggregates: An Application of Index Number and Aggregation Theory,” Journal of Econo metrics (vol. 14, 1980), pp. 11-48. ________“ The User Cost of Money,” Economic Letters (vol. 1, no. 2, 1978), pp. 145-49. _______ , and Jeong Ho Hahm. “ Financial-Firm Production of Monetary Services: A Generalized Symmetric Barnett Variable-Profit-Function Approach,” Journal of Business and Economic Statistics (January 1994), pp. 36-46. _______ , Douglas Fisher, and Apostolos Serletis. “ Consumer Theory and the Demand for Money,” Journal of Economic Literature (December 1992), pp. 2086-119. _______ , and Piyu Yue. “ Exact Monetary Aggregation Under Risk,” Washington University Working Paper No. 163 (De cember 1991). _______ , John Geweke, and Michael Wolfe. “ Seminonparametric Bayesian Estimation of the Asymptotically Ideal Production Model,” Journal of Econometrics (July/August 1991), pp. 5-50. MARCH/APRIL 1994 164 _______ , Melvin J. Hinich and Warren E. Weber. “ The Regulatory Wedge Between the Demand-Side and SupplySide Aggregation-Theoretic Monetary Aggregates,” Journal of Econometrics (October/November 1986), pp. 165-85. Duffie, Darrell. “ The Theory of Value in Security Markets,” in Werner Hildenbrand, and Hugo Sonnenschein, eds., Hand book of Mathematical Economics, vol. IV. North-Holland, 1991. _______ , and Yul W. Lee. “ The Global Properties of the Minflex Laurent, Generalized Leontief, and Translog Flexible Functional Forms,” Econometrics (November 1985), pp. 1421-37. Gallant, A. Ronald. “ Unbiased Determination of Production Technologies,” Journal of Econometrics (vol. 20, 1982), pp. 285-323. _______ , Edward K. Offenbacher, and Paul A. Spindt. “ The New Divisia Monetary Aggregates,” Journal of Political Economy (December 1984), pp. 1049-85. 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. Bernanke, Ben, and Mark Gertler. “ Banking and Macroeco nomic Equilibrium,” in William A. Barnett, and Kenneth J. Singleton, eds., New Approaches to Monetary Economics. Cambridge University Press, 1987, pp. 89-111. Bhattacharya, Sudipto, and Douglas Gale. “ Preference Shocks, Liquidity, and Central Bank Policy,” in William A. Barnett, and Kenneth J. Singleton, eds., New Approaches to Monetary Economics. Cambridge University Press, 1987, pp. 69-88. Blackorby, Charles, William Schworm, and Timothy Fisher. “ Testing for the Existence of Input Aggregates in an Econ omy Production Function,” University of British Columbia Discussion Paper No. 86-26 (1986). _______ , Daniel Primont, and R. Robert Russell. “ On Test ing Separability Restrictions with Flexible Functional Forms,” Journal of Econometrics (March 1977), pp. 195-209. Brainard, William, and James Tobin. “ Pitfalls in Financial Model Building,” The American Economic Review (May 1968, Papers and Proceedings 1967), pp. 99-122. ________“ Financial Intermediaries and the Effectiveness of Monetary Controls,” The American Economic Review (May 1963, Papers and Proceedings 1962), pp. 383-400. Brown, Randall S., Douglas W. Caves, and Laurits R. Christensen. "M odeling the Structure of Cost and Produc tion for Multiproduct Firms,” Southern Economic Journal (July 1979), pp. 256-73. Debreu, Gerard. Theory of Value, An Axiomatic Analysis of Economic Equilibrium, Cowles Foundation Monograph 17, Yale University Press, 1959. Denny, Michael, and Melvyn Fuss. “ The Use of Approxima tion Analysis to Test for Separability and Existence of Con sistent Aggregates,” The American Economic Review (June 1977), pp. 404-18. Diamond, Douglas W., and Phillip H. Dybvig. “ Bank Runs, Deposit Insurance, and Liquidity,” Journal of Political Econ omy (June 1983), pp. 401-19. Diewert, W. E. “ Exact and Superlative Index Numbers,” Jour nal of Econometrics (May 1976), pp. 115-45. ________“ Functional Forms for Profit and Transformation Functions,” Journal of Economic Theory (June 1973), pp. 284-316. ________“An Application of the Shephard Duality Theorem: A Generalized Leontief Production Function,” Journal of Po litical Economy (May/June 1971), pp. 481-507. _______ , and T.J. Wales. “ Flexible Functional Forms and Tests of Homogeneous Separability,” University of British Columbia Discussion Paper No. 91-12 (1991). ________“ A Normalized Quadratic Semiflexible Functional Form,” Journal of Econometrics (March 1988), pp. 327-42. ________“ Flexible Functional Forms and Global Curvature Conditions,” Econometrica (January 1987), pp. 43-68. FEDERAL RESERVE BANK OF ST. LOUIS _______ . “ On the Bias in Flexible Functional Forms and an Essentially Unbiased Form: The Fourier Flexible Form,” Journal of Econometrics (vol. 15, 1981), pp. 211-45. _______ , and Gene H. Golub. “ Imposing Curvature Restric tions on Flexible Functional Forms,” Journal of Economet rics (December 1984), pp. 295-321. Hall, Robert E. “ The Specification of Technology with Several Kinds of Output,” Journal of Political Economy (July/August 1973), pp. 878-92. Hancock, Diana. A Theory of Production for the Financial Firm. Kluwer Academic Publishers, 1991. ________“Aggregation of Monetary Goods: A Production Model,” in William A. Barnett, and Kenneth J. Singleton, eds., New Approaches to Monetary Economics. Cambridge University Press, 1987, pp. 200-18. ________“ The Financial Firm: Production with Monetary and Nonmonetary Goods,” Journal of Political Economy (Oc tober 1985), pp. 859-80. Hansen, Lars Peter. “ Large Sample Properties of Generalized Method of Moments Estimators,” Econometrica (July 1982), pp. 1029-54. _______ , and Kenneth J. Singleton. “ Generalized Instrumen tal Variable Estimation of Nonlinear Rational Expectations Models,” Econometrica (September 1982), pp. 1269-86. Ingersoll, Jonathan E. Theory of Financial Decision Making. Rowman and Littlefield Publishers, Inc., 1987. Jacklin, Charles J., and Sudipto Bhattacharya. “ Distinguish ing Panics and Information-Based Bank Runs: Welfare and Policy Implications,” Journal of Political Economy (June 1988), pp. 568-92. Lau, L. J. “ Testing and Imposing Monotonicity, Convexity and Quasi-Convexity Constraints,” in Melvyn Fuss, and Daniel McFadden, eds., Production Economics: A Dual Approach to Theory and Application. North-Holland, 1978, pp. 409-53. Leontief, Wassily. “ Introduction to a Theory of the Internal Structure of Functional Relationships,” Econometrica (Oc tober 1947a), pp. 361-73. _______ . “ A Note On The Interrelation Of Subsets Of In dependent Variables Of A Continuous Function With Con tinuous First Derivatives,” Bulletin of the American Mathematical Society (April 1947b), pp. 343-50. Mackinlay, A. Craig, and Matthew P Richardson. “ Using . Generalized Method of Moments to Test Mean-Variance Ef ficiency,” Journal of Finance (June 1991), pp. 511-27. Magill, Michael, and Wayne Shafer. “ Incomplete Markets,” in Werner Hildenbrand and Hugo Sonnenschein, eds., Hand book of Mathematical Economics, vol. IV. North-Holland, 1991. Newey, Whitney K., and Kenneth D. West. “ A Simple, Positive Semi-Definite, Heteroskedasticity and Autocorrelation Con sistent Covariance Matrix,” Econometrica (May 1987), pp. 703-08. Poterba, James M., and Julio J. Rotemberg. “ Money in the Utility Function: An Empirical Implementation,” in William A. Barnett, and Kenneth J. Singleton, eds., New Approaches to Monetary Economics. Cambridge University Press, 1987, pp. 219-40. 165 Robles, Barbara J. “ The Optimal Demand for Money in U.S. Manufacturing: A Dynamic Micro Theoretic Approach,” University of Colorado at Boulder Discussion Paper No. 93-15 (June 1993). Swofford, James L., and Gerald A. Whitney. “ Nonparametric Tests of Utility Maximization and Weak Separability for Consumption, Leisure, and Money,” Review of Economics and Statistics (August 1987), pp. 458-64. Rotemberg, Julio J. “ Monetary Aggregates and Their Uses,” in Michael T. Belongia, ed., Monetary Policy on the 75th Anniversary of the Federal Reserve System. Kluwer Aca demic Publishers, 1991, pp. 223-31. Thornton, Daniel L., and Piyu Yue. “ An Extended Series of Divisia Monetary Aggregates,” this Review (November/December 1992), pp. 35-52. _______ , John C. Driscoll, and James Poterba. “ Money, Out put, and Prices: Evidence from a New Monetary Ag gregate,” NBER Working Paper No. 3824 (August 1991). Tobin, James. “ Money, Capital, and Other Stores of Value,” The American Economic Review (May 1961), pp. 26-46. Sargent, Thomas J. Dynamic Macroeconomic Theory. Harvard University Press, 1987. Woodland, Alan D. “ On Testing Weak Separability,” Journal of Econometrics (December 1978), pp. 383-98. MARCH/APRIL 1994 166 William C. Brainard William C. Brainard is Arthur M. Okun professor of economics at Yale University. Commentary I t IS A PLEASURE TO TAKE PART in a scholarly conference focusing on empirical and theo retical issues relevant to the conduct o f m onetary policy and the behavior o f financial markets. Most o f the discussion at this conference, and indeed most o f the w ork on m onetary aggre gates, including a great deal by Bill Barnett, has been about the demand side. Demand elasticities and the degree o f substitutability o f various m onetary assets and liabilities are not just o f academic interest. They have direct and obvious relevance to the conduct o f m onetary policy. W hile there has been a great deal o f w ork on demand, there has been relatively little w ork on the supply side o f the market. The scarcity o f supply studies is perhaps inherited from simpler days when commercial banks, limited in their ability to compete fo r deposits and constrained in their portfolio choices, dominated the supply o f m onetary assets. In a w orld w here currency and demand deposits are the prim ary monetary assets, with no close substitutes, m onetary con trol was relatively simple; control o f bank reserves provided a tight control on the supply o f demand deposits and shifts betw een currency and demand deposits w ere relatively easily m onitored and offset. In today’s economy, w ith a rich menu o f monies and near monies, the task is not so simple. The instruments o f control— the supply o f reserves, reserve requirements and the discount rate—have remained essentially the same w hile the menu o f m onetary assets has proliferated. Financial firm s and markets can alter significantly the suppliers o f their assets and liabilities without policy accommodation. In this environm ent not only is there a question o f Digitized for FEDERAL RESERVE BANK OF ST. LOUIS FRASER what to control, but control itself is less direct and the timing and magnitude o f the response to policy less certain. In these circumstances, the Barnett-Zhou examination o f the competitive supply o f m oney and near monies by financial firm s is a w elcom e and important enterprise. The authors focus on the supply behavior o f the most important o f the financial interm edi aries, commercial banks. They m odel the bank ing industry as a competitive profit maximizing firm, stressing the dynamic nature o f bank’s op timization problem and the presence o f risk. The banking firm maximizes the present dis counted value o f expected utility, w h ere the util ity in each period is a function o f that period’s " c a s h flo w ” and displays risk aversion. The bank decides on its supply o f liabilities, taken to be demand and time deposits in the empirical anal ysis, and its demand fo r excess reserves and "loans.” Both assets and liabilities mature in one period, w ith the returns on loans being uncer tain. A production function determines the real resource costs associated w ith these portfolio decisions; in the estimation this function is as sumed to be weakly separable, so that the rela tive costs o f demand and time deposits do not depend on excess reserves nor on inputs o f labor and materials. The authors develop general methods fo r es timating dynamic and stochastic models o f bank behavior and demonstrate the feasibility o f us ing these techniques in the context o f a specific bank model. W hile the techniques could be ap plied in a w ide range o f settings, the model focused on in the paper incorporates tw o assumptions that severely limit the role fo r 167 dynamics. In particular, both bank assets and liabilities are assumed to mature in "one” period and the net proceeds o f the borrow ing and lending decisions made by a bank in one period are entirely paid out w hen the assets and liabili ties mature one period later. This is implied by equation 2: Liabilities issued in a period exactly cover required reserves, excess reserves, port folio investment and the payments fo r real resources. Hence, equity is zero (as the author’s indicate there would be no essential difference if it was non zero but constant); there is no room fo r retaining earnings. As a consequence, the only effect o f a decision in period (t) on net cash flow occurs at the beginning o f period (t+1). The net portfolio returns (the net cash flo w consequence o f decisions made in the previous period) are all paid out when they ar rive. Hence, if dividends—the net cash generated and distributed to the owners o f the firm — w ere entered in the utility function, the firm ’s optimization problem would be time separable and decisions could be made separately, period by period. What, then, makes the firm ’s decision dynamic in Barnett-Zhou? The reason is that the "cash flo w ” entered in the utility function is not the cash actually generated and distributed, but a measure o f profits developed by Diana Hancock (equation 1). This measure differs from actual cash flow by an amount reflecting changes in required reserves. In Hancock’s profit function required reserves are not recorded as an asset requiring the use o f funds. Yet, from the budget constraint in equation 2, liabilities exceed the sum o f excess reserves, loans and resource cost by exactly the amount o f required reserves. Hence, the Hancock profit function records as a positive cash flow an amount equal to required reserves in the period liabilities are incurred, and records a negative cash flow in the period they are repaid. These components o f Hancock’s profit function simply reflect the need to place a portion o f the deposits in reserve. I have difficulty understanding how to motivate their inclusion in the utility function; they do not cor respond to payments to the owners o f the firm, nor do they constitute an increase in the net w orth o f the bank. Although counting these flows as profits has essentially no effect on the present value o f the bank, it does serve to cre ate a link betw een time periods, making the problem dynamic. Several features could be added to the author’s model o f banks which would greatly increase the role fo r dynamics. Illiquidity and maturity mismatch may be less important today than earlier in the postwar period, but they re main significant reasons fo r treating the bank as a multiperiod firm . One extension w ould be to incorporate the fact that some o f banks' invest ments are in assets w ith maturities substantially greater than the maturity o f their liabilities. If held to maturity, investments made today have to be financed by future borrow ing. Second, since some bank assets are relatively illiquid, it would be interesting to build into the specifica tion some costs o f rapid asset disposal. Similarly, as w ith physical investment, there are costs o f adjustment on the rate o f acquisition o f assets. Another important extension would be to treat explicitly the dynamics o f equity growth. As with any firm , grow th in equity, either by new issue or by retention o f earnings, plays an im portant role in the grow th o f the industry. Ex plicit treatment o f capital accumulation seems particularly desirable given the capital require ments placed on bank portfolios, requirements which many thought w ere an important con straint on bank lending in recent years. Includ ing these elements w ould not only substantially increase the importance o f dynamics in the model, it would also add to the menu o f risks by, fo r example, allowing fo r the risks reflecting the interaction o f illiquidity and deposit uncer tainty. Not only can the author’s m odel be ex tended to analyze m ore complicated models o f banks, but it w ill undoubtably be useful in the study o f other financial intermediaries, institu tions that share many o f the features o f banks and, like banks, should be analyzed within a dynamic and stochastic fram ework. A num ber o f the author’s results are quite in teresting. A fter restricting the utility function to the CRRA class, they find the degree o f risk aversion significant and on the order o f one. They test and find they cannot reject weak separability, hence their estimates are consistent w ith the existence o f a theoretical m onetary ag gregate. The estimated aggregator function it self, evaluated at a point w here demand and time deposits are o f equal magnitudes, gives a marginal rate o f transformation implying that one dollar o f demand deposits is equivalent to approximately three dollars o f time deposits. This sounds like a plausible magnitude in the current regulatory environment; it would be interesting to know how different estimates would be fo r an early subset o f the data w hen MARCH/APRIL 1994 168 reserve requirements, portfolio restrictions and capital requirements w ere so different. In the author’s specification, excess reserves are an important input, w hile required reserves are seen as a sterile asset entering the firms technology neither as an input nor as an output. Th ere was a time w hen reserve requirements w ere quite high relative to estimates o f bank’s ow n transactions needs and this assumption would seem quite plausible. As reserve require RESERVE BANK OF ST. LOUIS FEDERAL ments have fallen, however, the distinction b e tween required and excess reserves has becom e less sharp. Another interesting extension o f the model, therefore, would be to include required reserves as an input and to test w hether their importance has fallen over time. As these comments suggest, I have found this an innovative and stimulating paper which opens up several new avenues fo r future research, and I look forw ard to watching the progress in this important enterprise. 169 William A. Barnett and Ge Z h o u 1 Response to Brainard’ s Commentary W 'E HAVE GREATLY BENEFITED from , Brainard’s stimulating comments on our paper. W e agree w ith his suggestions fo r extensions to this research, and in fact expect to have ex tended versions available in the near future. For example, Barnett, Kirova and Pasupathy (1994) are including an extended model in their paper being prepared fo r the Federal Reserve Bank o f Cleveland conference in September 1994. That model contains dynamic capital grow th through Tobin’s q, both fo r financial intermediaries producing m onetary services as outputs and fo r manufacturing firms demanding m onetary serv ices as inputs. In addition, that model contains an endogenous dividend payout decision, produced by entering loans into technology as an output, along w ith the deposits which cur rently are the sole outputs. Introducing loans into the technology as an output eliminates the need fo r Barnett and Zhou’s (1994) equation 2, which determines loans as a function o f deposits under the assumption that all earnings are paid out as dividends. Some readers may find Brainard’s comments difficult to interpret, however, since they reflect unpublished background material developed during correspondence. The following few the orems are relevant to understanding the nature o f the m odel’s dynamics and the merits o f ex tending the model further to exhibit deeper dy namics. Equation 2 in our paper is imposed to require the firm to pay out all earnings as dividends, since introduction o f an endogenous dividend payout decision is beyond the scope o f our paper. Equation 1 is in the general form o f the profit function used by Hancock (1985, equation 3.1) in her book and in her papers on financial intermediation under certainty. Equation 1 holds, regardless o f how the dividend payout de cision is made. Our equation 3 is acquired by imposing equation 2 on equation 1 through direct substitution. Hence, equation 3 is Han cock's variable profit function under the restric tion that all earnings are paid out as dividends. Under exactly that same payout restriction, Barnett (1987, equation 3.7) derived a different variable profit function fo r the same financial intermediary, and Barnett and Hahm (1994) recendy have used Barnett's form ulation o f the variable profit function in estimating the tech nology o f commercial banks. In correspondence, w e found that Brainard had a very strong preference fo r use o f Barnett’s, rather than Hancock’s, variable profit function under the payout restriction. The discussion about dynam ics in Brainard’s comment, including his discus sion o f the accounting fo r required reserves, 'B arnett’s research on this paper was partially supported by NSF grant number SES 9223557. We have benefited from many discussions with Richard Anderson on this subject, and a lengthy and highly informative exchange of faxes with William Brainard. MARCH/APRIL 1994 170 explains his reasons fo r preferring Barnett's function to Hancock's. Further changing to the notation in Barnett and Obviously Barnett would not dispute the merits o f Barnett’s variable profit function, and he is not at all displeased that Brainard so strongly prefers his variable profit function to Hancock’s. Nevertheless, it may seem paradoxical that tw o different variable profit function fo r mulas exist fo r the same firm under the same assumptions, and that w e chose to use Han cock's rather than Barnett's formula in our paper in this volume. As w e shall observe below, the distinction betw een the tw o variable profit functions is actually much more subtle than may appear to be the case, and the choice b e tween them in our paper is little more than an econom etric trick—barring empirical evidence to the contrary. are the nonfinancial variable factors, and w,, jt’ 7 = 1,...,./, are their prices. We then can rew rite the variable profit function as: Zhou (1994), let B, = £ w .z ., w h ere z „ J=1,...,J, jm 1 W e begin by verifying that equation 1 in our paper is indeed exactly Hancock’s variable profit function, w ith only the notation changed. * H = - J j wj,zj, , + £ J P r o o f: Hancock's (1991, equation 3.1) variable profit function, using her notation, is = ~B, - it" p, . - w here Bt = expenditure on variable factors; Pt = the general price level; yit = a financial asset quantity if i=\,...,N y or a financial liability quan tity if i= N 1+ l,...N 1+ N 2 hit = the holding period ; yield on yit if yu is an asset, or the holding cost on y.t if y.t is a liability; and w e define the indi cator function bi such that b = 1 if y.( is a liabil ity and b = -1 if y. is an asset. A m ore convenient notation would be to use the symbol A to denote assets instead o f the no tation N 1 and L to denote liabilities instead o f the notation N 2. Making that change in notation and using the indicator function b as defined above, w e acquire: = ~ B, - s - yM p , M + K -M t-A -i - y fX t= L + 1 As in Hancock (1991), the assets consist o f loan investments and excess reserves, which in our notation are Yt and Ct respectively. Furthermore, let Rt be the single period holding yield on Yt and let the yield on C( be zero, as in Hancock (1991). The variable profit function now becomes: i= L + 1 + C .- A -i ~ C,P, + < + 1 - yitpX 2Also, see equation (3) in Hancock (1985) for another state ment of Hancock’s formula. FEDERAL RESERVE BANK OF ST. LOUIS - y.,p,] - YtP,, which is exactly equation 1 in Barnett and Zhou (1994). Q.E.D. In the next theorem, w e prove that the d iffer ence betw een the discounted present value o f the profit flo w produced from Hancock’s form u la (that is, Barnett and Zhou’s 1994 equation 1) and the discounted present value o f the profit flow produced by Barnett’s (1987, equation 3.7) formula is a function only o f initial conditions. The proof is produced under imposition o f Bar nett and Zhou’s equation 2, which requires all earnings to be paid out as dividends. Let yit be deposits in account type i and let r , be the single period holding yield on that ac count. Let K it be the required reserve ratio on that account. Before proving the equivalency theorem, w e define the tw o formulations o f the variable profit function as follows. D e fin itio n 1: Barnett’s (1987, equation 3.7) variable profit function is L A +L + s J ui + K - M t - ip.-i - 1 w],z], - s T h e o re m 1: The variable profit function de fined by equation 1 in Barnett and Zhou (1994) is identical to Hancock’s (1991, equation 3.1) vari able profit function.2 J J 171 w here the user cost o f account yit is n„ = P, w hile the capitalized value o f Barnett’s profit stream 7 is r„ l + R, and the user cost o f excess reserves ct is % = p, Substituting the formulas fo r the profit streams into the tw o capitalized values and manipulating algebraically, w e find l + R, D e fin itio n 2: Barnett and Zhou’s (1994, equa tion 3) variable profit function is 7TflZ, = S ( [( ! + /!,_,) + C b = 5 ^ ( i + R ,J w j l _izjt_l, T h e o re m 2: The discounted present value o f the firm , CB produced from Barnett’s profit , function flow, 7rB, differs from the discounted present value o f the firm, Cir produced from Barnett and Zhou’s (in other words, Hancock’s w ith no retained earnings) profit function flow, wB by a function, K(I), containing only initial ZI, conditions, /. In other words, there exists K(I), depending only upon initial conditions, such that C„ = CB+ K(/). P r o o f: Define the discount factor 8s such that f 1 when s = t = n v (1+fl.) w hen s > f + l, a=t a w here Ra a = t,...,s -l, are current and expected , future values o f the rate o f return, Rt, defined above to be the single period holding yield on Yt. The discounted capitalized value o f the profit stream 7rBZ( at time f is 5 *BZs > % B s [ E rj v. 5 l-l ‘^ ‘S E w z. - ri c ] 7 -1 JS JS C S S and 1 w here hIS = rIS + kISRs . Note that by Theorem 1, Definition 2 also de fines Hancock’s (1991, equation 3.1) variable profit function under the restriction that all earnings are paid out as dividends (Barnett and Zhou’s (1994), equation 2). A fter multiplying Bar nett's variable profit function (defined by Defini tion 1 above) through by 1 + Rt, it is easily seen that the profit function preferred by Brainard in his comments (as further clarified by our pri vate correspondence) is Barnett’s profit function. The equivalency theorem, producing a connec tion between Definitions 1 and 2, follows. s = S s ., +Ky,,pJ - K A - A - i - 8 ~ 5 = K + CB f w here K = S [[(l+ B ,.,) i l - k uJ - n + h itl_1 ))yiil_,PtJ Observe that K depends only upon initial condi tions, since the intertemporal decision is made at time t over periods t, t + 1, t + 2,.... Q.E.D. Theorem 2 proves that under the restriction that all earnings are paid out as dividends and except fo r a function o f initial conditions, Bar nett’s variable profit function and Hancock's variable profit function are simply different ways o f spreading the capitalized value o f the firm over time. Any flo w o f funds or transaction that appears in one formula necessarily also ap pears in the other, but potentially w ith a time shift between them. Those time shifts are all properly discounted, however, as demonstrated by the fact that the tw o profit streams produce the same capitalized value up to K(I). In his com ment, Brainard observes correctly that the choice between the tw o profit flow formulas “ has essentially no effect on the present value o f the bank.” Theorem 2 above makes that point clear. The discussion that follows w ill extract from Theorem 2 its precise implications fo r the model estimated by Barnett and Zhou in this volume. Our discussion w ill compare the solutions o f the tw o decisions defined below. D e c isio n 1: For some utility function, U , the firm determines its factor demands and output MARCH/APRIL 1994 172 supplies by maximizing, EU(CB which is the ex ), pected utility o f the capitalized value CB . D e c is io n 2: For some utility function, V, the firm determines its factor demands and output supplies by maximizing, EV(CH w hich is the ex ), pected utility o f the capitalized value CH . Observe that all terms in each capitalized value are inside the respective utility function, which is not assumed to be intertemporally separable in either case. The marginal utility o f anything varied within either capitalized value depends upon everything else in that capitalized value. In short, neither utility function is inter tem porally separable and the solution o f either decision is deeply dynamic. In fact, each solu tion is intertem porally simultaneous w ith all time subscripts appearing in all Euler equations. To determine w hether there are any substan tial differences betw een Decisions 1 and 2, w e now define the follow ing concept. D e fin itio n 3: TWo decision problems are observationally equivalent if the solution functions (factor demand and output supply) functions produced by solving one problem are identical to the solution functions produced by solving the other at any fixed setting o f the initial con ditions. Th e following theorem and corollary are now easily proved. T h e o re m 3: For any given fixed value o f the initial conditions function, K(I), and any given utility function, U, there exists a utility function, V, such that V(CH = U(CB fo r all possible settings ) ) o f the firm's decision variables (the controls). P r o o f: For given K (I) and U, define V such that V(x+K(/)) = t/W fo r all nonnegative values o f the scalar ,x. N ow let ,x= CB and let CH=}c + K(I). By , substitution, the result is immediate. Q.E.D. C o r o lla r y 1 to T h e o re m 3: Decisions 1 and 2 are observationally equivalent. P r o o f: The corollary follows immediately from Definition 3 and Theorem 3. Q.E.D. The implications o f the above results at this point are the following. To justify the introduc tion o f risk aversion into the decision o f the 3lf contingent claims markets are complete, then the owner will instruct the manager to maximize profits conditionally upon the prices in contingent claims markets. Those prices contain the information about the risk aversion of the own er and, hence, the managers will be instructed to behave in a risk-neutral manner relative to those prices. See Duffie (1991) and Magill and Shafer (1991). FEDERAL RESERVE BANK OF ST. LOUIS firm , w e implicitly assume the existence o f in complete markets.3 How to model the decisions o f firm s w ith incomplete contingent claims m ar kets is controversial. One approach that has been proposed is to apply principle agent theory in a form that produces incentive compatibility, w hen the decision is delegated by the owners to a professional manager. Th e source o f the risk averse, concave utility function is the utility function o f the principle agent. Having introduced expected utility maximiza tion into the firm ’s decision in that controversial manner, w e then see from the above corollary that it makes no difference w hether w e use Hancock’s variable profit function or Barnett’s in producing the Euler equations to be estimated. The Euler equations are identical and the deci sion is deeply dynamic, w ith all time subscripts appearing in each Euler equation. The choice betw een the tw o profit formulas is a choice b e tw een tw o different methods o f spreading the same capitalized value over time. But since it is the capitalized value itself that enters as the sole argument o f the utility function, the m ethod o f spreading over time is irrelevant. Corollary 1 is the result. The problem at this point is that estimating a system o f simultaneous Euler equations is b e yond the state o f art. We need a means o f decreasing the depth o f the m odel’s dynamics. An obvious m ethod would be to impose a separability restriction on the utility o f capital ized value. W e could use complete separability, blockwise separability, weak separability, or strong separability. Separability restrictions are testable structural restrictions, and behavior is not invariant to such structural restrictions.4 In addition, nothing in principle agent theory helps us to choose betw een such restrictions, which in fact all may be wrong. The utility function may indeed be nonseparable, and the decision may be unavoidably deeply dynamic. Further more, w e are aware o f no empirical results that would help us to choose betw een the many sim plifying separability restrictions, and the few results in that area in Barnett (1981) indicate that separability restrictions are strong restric tions that often are rejected in empirical tests. 4This issue does not exist in the perfect-certainty or riskneutral case, since in those cases there is no utility func tion to be structurally separable. The invariance theorem, then, is the end of the story. 173 Under such circumstances, applied researchers regularly choose simplifying assumptions on the basis o f their usefulness in estimation. One pos sibility is intertemporal strong separability in Hancock’s profit stream. Another possibility is intertemporal strong separability in Barnett's profit stream. M ore formally, those tw o possibili ties are Assumptions 1 and 2 below, respectively. A s s u m p t io n 1: The utility function, V, is intertemporally strongly separable in {C H(:f = l,...,°°}. A s s u m p tio n 2: The utility function, U, is intertemporally strongly separable in {C B(:f= l,...,°°}. Th ere are many other such possibilities pro duced by grouping together terms in the capi talized value in different manners. Behavior is not invariant to choices between those possible separability restrictions. In terms o f the degree o f simplification o f the Euler equations, com plete intertemporal separability in Barnett’s profit stream (Assumption 2), as assumed by Barnett and Hahm (1994), produces the most ex trem e simplification. The decision becomes com pletely static. Complete intertemporal separability in Hancock’s profit stream (Assumption 1) produces a more modest decrease in the depth o f the dynamics: The solution becomes recur sive, w ith tw o time subscripts appearing in the Euler equations. Barnett and Zhou (1994) selected and imposed the latter restriction, since the resulting recur sive form o f the solution assists in GMM estima tion. Brainard (1994), in his commentary, argues forcefully fo r intertemporal strong separability in Barnett's profit stream. We have no reason to dispute his strong prior on this subject. His views are reasonable, and obviously Barnett (1987) and Barnett and Hahm (1994) must have had somewhat similar priors in mind w hen they published their work. Nevertheless, it is also possible that the opposite extreme may be true. The utility function may be completely nonseparable, so that both Assumptions 1 and 2, along with all other possible separability restric tions, may be w rong.5 The Euler equations would thereby be intertemporally simultaneous, so that w e cannot readily estimate the model w ith current methods because o f the depth o f the dynamics. Even worse, it may be the case that the use o f a risk averse principle agent as a means o f introducing risk aversion into the deci sion o f the firm may be a defective approach. That question at present is unresolved in eco nomic theory.6 Under these circumstances, w e feel justified in choosing our separability restriction based upon the resulting estimation convenience. Producing interesting dynamics w ith long-run economic grow th was not an objective o f Barnett and Zhou (1994), which was an exploration in aggre gation theory fo r firm s under uncertainty. We agree w ith Brainard that far m ore interesting dynamics would be produced by introducing a law o f motion fo r capital, which indeed w ill be included in Barnett, Kirova and Pasupathy (1994). We wish to acknowledge that the above clarifying proofs resulted from our correspon dence w ith Brainard, and w e are indebted to him fo r motivating this exploration o f the con nection betw een Hancock’s and Barnett’s form u lations. Many o f his other suggestions will be used in future extensions o f our research such as the estimation o f the model w ith learning-bydoing technological change. Although w e have not yet estimated that model, the Euler equa tions fo r that extended model are provided in Barnett and Zhou (1994) and the dynamics in that model are indeed dynamic in an interesting manner. REFERENCES Barnett, William A., Milia Kirova and Munich Pasupathy. “ Es timating Policy Invariant Technology Parameters in the Financial Sector, When Risk and Growth Matter,” Paper to be presented at the Federal Reserve Bank of Cleveland, September 1994 macroeconomics conference on Liquidity, Monetary Policy, and Financial Intermediation. _______ , and Ge Zhou. “ Financial Firm’s Production and Supply-Side Monetary Aggregation Under Dynamic Uncertainty,” this Review (March/April 1994). _______ , and Jeong Ho Hahm. “ Financial-Firm Production of Monetary Services: A Generalized Symmetric Barnett Variable-Profit-Function Approach,” Journal of Business and Economic Statistics (January 1994). _______ . “ The Microeconomic Theory of Monetary Aggrega tion,” in William A. Barnett and Kenneth J. Singleton, eds., New Approaches to Monetary Economics. Cambridge University Press, 1987, pp. 115-68. ________Consumer Demand and Labor Supply. NorthHolland, 1981. 5ln fact, the assumption of intertemporal separability of preferences has become controversial in the real business cycle literature. See, for example, Kydland and Prescott (1982). 6See for example Magill and Shafer (1991). MARCH/APRIL 1994 174 Brainard, William C. “ Commentary,” this Review (March/April 1994). Duffie, Darrell. “ The Theory of Value in Security Markets,” in Werner Hildenbrand and Hugo Sonnenschein, eds., Handbook of Mathematical Economics, volume IV. NorthHolland, 1991. Hancock, Diana. A Theory of Production for the Financial Firm. Kluwer Academic Publishers, 1991. Digitized forFEDERAL RESERVE BANK OF ST. LOUIS FRASER ________“ The Financial Firm: Production with Monetary and Non Monetary Goods,” Journal of Political Economy (October 1985) pp. 859-80. Kydland, Finn E., and Edward C. Prescott. “ Time to Build and Aggregate Fluctuations,” Econometrica (November 1982), pp. 1345-70. Magill, Michael, and Wayne Shafer. “ Incomplete Markets,” in Werner Hildenbrand and Hugo Sonnenschein, eds., Hand book of Mathematical Economics, volume IV. North-Holland, 1991. 175 Jerome L. Stein Jerome L. Stein is professor of economics and Eastman professor of political economy at Brown University. Can the Central Bank Achieve Price Stability? i i l L FOMC’S STATED POLICY objectives are to "foster price stability and promote sustainable grow th in output.” Can these objectives be achieved with the tools available? We know that there is a long-run relationship between the ratio M/y=Money/real GDP and the P =G D P deflator o f the form (a) P = V(M2/y), w h ere V is the velocity function, shown in Figure 1. The Federal Reserve would like to select ranges fo r monetary growth over the coming year consistent w ith price stability.1 This is the policy o f monetary targeting. The ration ale fo r the policy o f m onetary targeting is the existence o f a stable and reliable relationship b e tween the rate o f grow th o f monetary aggregate Mi [denoted and the rate o f inflation (denoted tt) either during year t or possibly t + 1 o f the form (b) ir(t) = c + c'Hift) or (c) ir(t) = c + c 'n f t -1 ). 1By price stability, we mean a desired rate of change of prices, which need not be zero. 2See Belongia and Batten (1992), Thornton (1992), Garfinkel and Thornton (1989), and Ritter (1993). Equation (a) is a long-run relation between the price level and the stock o f money per unit o f real GDP, and equations (b) and (c) are shorterrun relations betw een the rate o f grow th o f prices and the rate o f growth o f money. They are quite different. It has been amply demonstrated by monetarists that neither the grow th o f M l nor o f M2 produces a stable and reliable relationship o f the form (b) or (c).2 The targeting o f M l was abandoned w hen the velocity function changed drastically after 1980, and M2 targeting was then used. There was subsequent disappoint ment with targeting M2. Figure 2a-d shows w hy m onetary targeting equations (b) and (c), either fo r M l or M2, are not reliable. The source o f the problem is the instability and un reliability o f the velocity function (V I fo r M l, and V2 fo r M2 in Figure 3a). This led Alan Greenspan (1993) to question the usefulness o f M2 targeting [equation (b) or (c)]:3 "...the relationship betw een money [M2] and the econom y may be undergoing a significant trans formation....This is not to argue that money grow th can be ignored in form ulating m onetary policy....Selecting ranges fo r monetary growth over the coming year consistent with desired 3The article by Ritter (1993), “ The FOMC in 1992: A Mone tary Conundrum,” conveys the serious problems that arose when the FOMC tried to implement the policy of monetary targeting. MARCH/APRIL 1994 176 Figure 1 GDP Deflator and the Ratio of M2/Real GDP GDP deflator M2/Real GDP Figure 2a Inflation and the Growth of M2 Inflation (change in GDP deflator) FEDERAL RESERVE BANK OF ST. LOUIS Growth of M2 177 Figure 2b Figure 2c Inflation and the Growth of M1 Inflation and the Lagged Growth of M2 Inflation Inflation Lagged growth of M2 Growth of M1 Figure 2d Figure 2e Inflation and the Lagged Growth of M1 Inflation and the Growth of Divisia M2 Inflation Inflation Lagged growth of M1 Growth of Divisia M2 MARCH/APRIL 1994 178 Figure 3a Velocity of M1 and M2 Figure 3b Velocity of Divisia M2 Digitized for FEDERAL RESERVE BANK OF ST. LOUIS FRASER 179 economic perform ance, however, is especially difficult when the relationship [velocity] b e tw een money and income has becom e uncer tain. Recent experience suggests that...measuring money against such ranges may lead to errone ous conclusions regarding the stance o f m one tary policy.” Greenspan’s disappointment w ith the use o f m onetary targeting (M2) has led him to revive the concept o f interest rate targeting. The ultimate question is how the central bank should try to produce price stability and sus tainable growth. Our paper addresses several important questions: 1. Is there an economically significant, structur ally stable, policy-rule-invariant relationship between the rate o f growth o f a monetary ag gregate and the rate o f grow th o f the price level? If so, that m onetary aggregate is referred to as an indicator. W hat monetary aggregates, if any, qualify as indicators? 2. W hich monetary aggregate is an intermediate target? An intermediate target is defined as a variable Z which is an indicator and is also controllable over a range o f policy regimes. 3. Under what conditions can Federal Reserve policy be used to speed the recovery and what w ill be the consequences fo r the rate o f inflation? 4. Does the controllable Treasury bill rate quali fy as an indicator or intermediate target? Our major conclusions are: A. The relation between the growth o f the m onetary aggregate and inflation is indirect. The change in the rate o f inflation depends upon the unemployment rate and the grow th o f real balances which changes real ag gregate demand. Neither the grow th o f M2 nor the growth o f adjusted reserves per se conveys very much useful information about the course o f inflation in the near future, b e cause the inflation and unemployment rates 4An intermediate target is an indicator, but not necessarily the reverse. 6For economy of notation throughout the paper, the operator D represents either the discrete first difference operator D x = x(t)-x(t-1 ) or the continuous time derivative Dx=dx/dt as appropriate in the context. interact in a dynamic manner. W ithin the context o f the dynamic model, the grow th o f M2 is a good indicator o f the rates o f infla tion and unemployment. B. The grow th o f M2 has both an endogenous component and a directly controllable part. The link between the grow th o f M2 and reserve grow th was tight from 1958-1975 and then became very weak from 1975-1992. Hence, the grow th o f M2 is prim arily an indi cator. The grow th o f adjusted bank reserves is an intermediate target fo r the rate o f infla tion, but less so fo r the unemployment rate, within the context o f the dynamical system. C. Weighted m onetary aggregates are inferior to M2 as an indicator. D. The nominal or “ real” Treasury bill rate fails completely as an indicator, so it cannot be an intermediate target.4 The flo w chart below describes the relation betw een the research design and the conclu sions stated above. The Federal Reserve has been seeking a direct relation between the grow th o f m onetary aggregate Mi, w here Dflog M i) is denoted /i., in a given year and the rate o f inflation Dflog P) = tt in the subsequent year.5 We have seen that there is no direct rela tion between n J t - l ) and inflation ir(t). The rea son is that the relation between inflation and money grow th is indirect and works through a dynamic model. We first derive the structural equations o f a dynamical system involving the state variables X, which are the inflation (7r), the anticipated inflation (7r*) and unemployment rates (u).e The input is the rate o f grow th o f a m onetary aggregate fi = DM/M. The resulting reduced form system (the SM dynamical system) is o f the form D X = A X + Bfj. + e', described in Table 1 or the flo w chart below.7 SM dynamical m odel money growth X *- D X=AX+Bn+e' indicator * - / « - [i = CX + bz + e" * - z x control 6The unemployment rate u(t)=U (t)-U e is the deviation be tween the measured unemployment rate U(t) and the equilibrium level Ue. 7SM denotes the Stein Monetarist dynamical model as developed in Stein (1982). MARCH/APRIL 1994 180 Table 1 The Reduced-Form Equations for the Dynamics of Inflation and Unemployment, from the SM Model_________________ (7) Du = a„u + a 727r + a13tt* + e'; a 7 < 0, a 12 < 0, a13 > 0, a 12 + a 13 = 0 7 (8) Dtt = a21u + a 22 * + &23 ^ * a 22 ^23 ^24 = ^ 24m ^ &21 ^ ®22 ^ Q *^23 ^ ^24 ^ Q ^ (9) Dir* = -C7r* + c n; 1 > c > 0 (9.1) Tr-ffj = (1-cy> r'(t-n ) + c(1-c) r f t- 1 ) + c £ ^ tff-s ; = C(L)fi 2 Note: u = U-Ue = unemployment rate U less the equilibrium rate Ue. n = inflation rate. p. = rate of monetary growth. We estimate a surrogate o f the SM dynamical model, in which the dependent variables X are the observable unemployment and inflation rates. The equations o f the surrogate model are: w ( t ) - i r ( t - l ) = b ’0 - b’J J b - l ) + b ’2 [f if t - D - fift-D I + et; U(t) - U ( t - l ) = a’0 - a\ U (t -1 ) - a’J i i ( t - l) - ir ( t - l) I + e2; E(et) = 0. The effect o f the growth o f the m onetary in put / upon the rate o f inflation is indirect: It x operates through the dynamical system, which also involves the unemployment rate. The change in the rate o f inflation depends upon the unemployment rate and the rate o f change o f real balances (/ —7r). The change in the unem x ployment rate depends upon its level and the change in real balances. W e have already seen that there is no direct relation betw een the rate o f m onetary expansion n (t -1 ) and the subse quent rate o f inflation ir(t). However, w hen we consider how the rate o f M2 monetary expan sion n ( t - l ) operates upon the dynamical system, implied by the structural equations, the growth o f M2 is a very good indicator o f the subse quent rates o f inflation and unemployment. The matrices A and B are structurally stable and policy-rule invariant; and the surrogate system is a good predictor. This is conclusion A above, that the grow th o f M2 is a good indicator. We show that the grow th o f M2 is better than alter native m onetary aggregates (conclusion C). Digitized for FEDERAL RESERVE BANK OF ST. LOUIS FRASER We then consider the intermediate target issue: To what extent is the grow th o f M2 controllable? This is the next link in the flow chart: / = CX + z bz + e". The rate o f monetary expansion / has x tw o components. One component is the grow th o f reserves z which is controllable. The other component is CX, the induced part o f the grow th o f M2, which responds to the state o f the economy. W e estimate this relationship. From 1958-75, the grow th o f M2 was determined by the controllable grow th in reserves. A fter 1975, and especially after 1984, the grow th o f reserves did not have that effect and the grow th o f M2 was endogenous. The reason is that the growth o f the non-M l component o f M2 was not con trollable by the grow th o f reserves (conclusion B). The grow th o f M2 was an intermediate tar get prior to 1975, and much less so afterwards. Combining the tw o links, w e ask w hether the controllable growth o f reserves z operates through the dynamical system, o f the following form: X < ---- DX = (A + BC)X + (Bb)z + (Be" + &) < ----- z intermediate target control variable The answer is that this system is acceptable for the inflation rate and less so fo r the unem ploy ment rate in recent years. This is in conclusion B. Finally, w e ask w hether the controllable It'easury bill rate operating through the dynamical sys tem can be considered to be an intermediate target. Conclusion D is that there is no inform a tional content to the controllable Treasury bill rate. It is neither an indicator nor an interm edi ate target. 181 THE SM DYNAMIC MODEL8 The Structural Equations There are five structural equations and one identity to the SM dynamic model. First: The rate o f inflation 7r = Dp/p depends upon the ex cess demand fo r goods, J(t) = aggregate real demand less current real GDP, and the rate o f grow th o f unit labor costs DW/W, w here IV is unit labor costs. This is equation 1, w here D = dJdt, y is a parameter: (1) ir = DW/W + yj. produces portfolio balance.9 In terms o f the usual Keynesian 45-degree diagram, J is the ve r tical distance between aggregate demand and current real GDP (the ordinate on the 45-degree line). The basic parameter o f the aggregate de mand curve is real balances per unit o f capacity output. Hence, the excess demand fo r goods de pends upon the unemployment rate (which is negatively related to the ratio o f actual to capac ity output), real balances m (t) = M/PY* per unit o f capacity output Y* and disturbances r\(t). (4) yJ(t) = J(U, m; r\) = J t U + J, In (m ) + Second: The growth o f unit labor costs de pends upon the state o f the labor market, reflected by the deviation between the unem ployment rate U(t) and its equilibrium rate Ue, and by the anticipated rate o f inflation ir*. This is equation 2. That is, the anticipated rise in the real unit labor costs depends negatively upon the excess supply in the labor market, w here the excess supply is reflected in the unemploy ment rate. (2) DW/W = tt* - h [U (t)-U e l Equation 3 simply states that the observed un employment rate is positively related to the un observed excess supply o f labor. The demand fo r labor depends negatively upon real unit labor costs, and the supply o f labor has an al gebraically greater relationship w ith the real unit labor costs than does the demand. Hence, the observable unemployment rate, which is positively related to the unobservable excess supply o f labor, depends positively upon real unit labor costs W /P. (3) U(t) = b0 + b, In (W/P) The real excess demand fo r goods J(t) = real aggregate demand less real GDP is equation 4, when w e have solved fo r the equation, which 8This is explicitly developed in Stein (1982), and Infante and Stein (1980). Here, we attempt to simplify and focus exclu sively upon the basic characteristics. The SM refers to my version of a monetarist system. The techniques of analysis are different from conventional monetarists since the veloc ity function is not used and the SM model involves an in teraction of unemployment and inflation. The conclusions, however, are quite close to those of Friedman, hence the term monetarist. In a sense, the SM dynamic model lies between the thinking of Friedman and Tobin. 9This is discussed in equation 19 in connection with the in termediate target. Y ], J > 0. Substitute equations 2 and 4 into equation 1 to obtain 1.1. It is clear that the inflation equa tion is not the usual expectations augmented Phillips curve, since it contains the real balances as variables as well as the unemployment rate and rate o f anticipated inflation. (1.1) ir = tt" - h[U-Uel + y J j U + J2 ln (m ) + rj The anticipated rate o f inflation slowly con verges to the trend rate o f m onetary growth per unit o f output, equation 5. Variable fi(t) is the rate o f m onetary grow th and n is the trend rate o f grow th o f output. Th ere are two established facts: (a) There is a long-run, positive relation between the price level and some m onetary aggregate (Figure 1), and (b) On a year-to-year basis, there is no reliable relationship 7t = c + c ' /jl between money grow th and the subsequent rate o f inflation (Figure 2). That is, there is very little inform a tional content in the current rate o f monetary expansion concerning the rate o f inflation in the near future.1 0 In our Bayesian fram ework, there is a prior anticipated rate o f inflation 7r*tt).1 Then, there is 1 10We have no need to use the subjective concept of antici pated or unanticipated money growth. 1 We use the concept of Asymptotically Rational Expecta 1 tions as developed in Stein (1992a,b). Our results are not sensitive to the specific form of the anticipated inflation equation. Any anticipations function that satisfies the following conditions will suffice. First, in the steady state, a change in the rate of monetary expansion changes actual and anticipated inflation by the same amounts. Second, a change in the rate of monetary expansion at time t does not change the current rate of anticipated inflation by as much. MARCH/APRIL 1994 182 Table 2 The Surrogate System: Estimated Inflation and Unemployment Equations Growth M2 = n ir Unemp U Growth reserve = z Inflation ir Unemp U 1.96 [0.00] 0.76 [0.00] 0.29 [0.00] -0 .2 3 [0.00] -----0.76 0.72 1.6 [0.06] -0.34 [0.016] 0.92 [0.00] -----0.16 [0.02] 0.78 0.18 Variable Inflation constant U(t— 1) 7r(t— 1) „ ( t-1 ) z(t-1) ADJ.R-SQ LM prob (F) 1.4 [0.1] -0.39 [0.01] 0.86 [0.00] 0.21 [0.03] -----0.77 0.07 1.6 [0.01] 0.69 [0.00] 0.23 [0.00] ------0.13 [0.01] 0.71 0.72 Notes: Sample period 1958-92, annual, N=35. Columns one and two refer to equations 10 and 11 for growth of M2; columns three and four refer to equations 12 and 13 for growth of reserves. The two-tail significance level is shown in brackets. current information, which is the current rate o f m onetary expansion [p(t) - nj. Combining the two, the posterior anticipated inflation Tr*(t+1) = ( I - c ) t t *(t) + clfiftj-n l, is a linear combination o f the prior and the current information. The coefficient c is the w eight given to the current sample o f information. Subtract the prior from both sides and derive: (5) D ir* = ir*(t + l;t ) - ir*(t) = d n (t) - n - ir*(t)]. The "credibility" argument is contained in the value o f coefficient c. If the public believes that the central bank is committed to an inflation target [the p rior ir*(t)], then variations in the current rate o f m onetary expansion l / i f t ) - n ] w ill be given a low weight and coefficient c will be small. Coefficient c reflects the predictability that the current rate o f m onetary grow th w ill continue fo r a long time and the tightness o f the relation between money grow th and infla tion over the relevant horizon. The rate o f grow th o f real balances relative to the trend rate o f grow th o f output n is equation (6), which closes the system. (6) Dm/m = p - ir — n. These dynamic interactions betw een the infla tion rate, unemployment rate and m onetary poli cy must be explicitly considered if w e are to answer the questions posed at the beginning o f this paper: Specifically, what is an indicator and what is an intermediate target? Equations 1-6 are solved in the dynamic form described by FEDERAL RESERVE BANK OF ST. LOUIS Table 1. These differential equations im ply the steady-state relations as well as the medium-run dynamics. The steady-state solution is that: The unemployment rate converges to the equilibrium rate. The latter is independent o f m onetary fac tors. The actual and anticipated rates o f inflation converge to the grow th o f the money supply (or grow th o f the m oney supply less the long-term grow th rate o f the economy). Equation 5 or 9 may be solved to yield equation 9.1 in Table 1. The anticipated rate o f inflation at any date t is a w eighted sum o f past rates o f monetary ex pansion, w ith declining weights. AN EMPIRICAL SURROGATE SYSTEM USING M2 AS INPUT The system described in Ihble 1 involves the measured unemployment and inflation rates and the nonobservable anticipated rate o f inflation. For empirical analysis, w e convert the SM dy namic model in 'Iable 1 into a surrogate system, involving measurable quantities only. These, in the form o f equations 10 and 11 below, are used fo r empirical estimation in Table 2. The sur rogate system mimics the dynamical system. First w e explicitly derive, from equations 1-6 o f the SM model, the reduced form equations in 'Table 1. Then w e show how the surrogate sys tem is derived from the SM model. Differentiate equation 3 w ith respect to time and use 2 to obtain 7: (7) Du = b(ir * - hu - it) + e' = anu + a1 7 + a1 tr* + e! 2 r 3 183 Differentiate 1 w ith respect to time, using 4 -7 to obtain equation 8. The constraints on the coefficients follow from definitions o f atj and b..: (8) Dir = - hb(Jj - h)U - I(J, - h)b + J jir + [(Jt - h ) b - c l it * + (J, + c) (/ i-n ) + e" = a2IU + a,,7r + a2 ir* + b2 ( - n ) + e" 3 4 Equation 9 is equation 5 above: (9) D n* = -C7T* + c/x The continuous time dynamical system 7-9 in T^ble 1 may be w ritten as DX = AX + B/x + e, w here X = (u,ir,ir*). W e use e as a generic representation o f a random variable w ith a zero expectation. In this paper, w e use annual rather than quarterly data because w e obtained clear-cut, significant results w ith annual data (Tible 2), whereas nothing o f economic significance em erged w hen w e used the noisy quarterly data, as shown in the appendix. W hen the data are annual and one just uses the observable U, 7r and n the surrogate empirical system is equa tions 10 and 11. (10) ir(t) = b0 + b j U ( t - l ) + b2i r ( t - l ) + b3 (t - 1 ) + e'; n Hg: b, + b3 = 1; fa, < 0 (11) U(t) = a0 + atU ( t - l ) + a2n ( t - l ) a3 (t - 1 ) + e"; n H0 a, + a3 = 0; a3 < 0 : Th ere are tw o important theoretical con straints concerning m onetary neutrality. Equal rises in m oney grow th and inflation do not 12This can be seen as follows. The estimates (from Table 2) of the surrogate system 10 and 11 are 10.1 and 11.1. The SM model (Table 1) can be written as (A.1)-(A.3) when the following values are used. The half-life of the deviation of: (i) the inflation rate from its equilibrium value is two years, (ii) the unemployment rate from its equilbrium value is 3.5 years and (iii) anticipated inflation from its equilibrium is five years. This gives us the coefficients in the principal di agonal of matrix A. (ii) The effects of inflation and anticipat ed inflation upon the change in unemployment and the change in inflation are equal and opposite (see equations 7, 8). (iii) All variables are measured as deviations from their steady-state values. Then the SM dynamic system is: (A.1) (A.2) (A.3) Dir = -.197 7r Du = -.7tt Dir* = - .1 U - ,347u + .197t ' + ,7tt* - ,138-k * change real balances and, hence, have no effect upon the unemployment rate. Similarly, in the steady state, the actual and anticipated rates o f inflation w ill change by as much as the rate o f m onetary expansion. One is not free to con struct any m onetary aggregate as either an indi cator or an intermediate target simply on the grounds that it seems w ork over the period considered. Instead, the m onetary aggregate must be closely linked to the theory, such that the variable satisfies certain neutrality con straints. The neutrality constraints in the indica tor system are as follows. In a comparative steady state, money and prices change by the same proportion, there is no effect upon the un employment rate. The constraint in inflation equation 10 is that in the steady state a change in the rate o f m onetary expansion w ill change the actual and anticipated rates o f inflation by the same amount: b2 + b3 = 1. The constraint in unemployment equation (11) is that, w hen money and prices change by the same amount, there is no effect upon real unit labor costs and no change in the unemployment rate: W ith these constraints, the surrogate system 10 and 11 mimics the SM dynamic system, Table l.1 2 Regarding equations 7-9 or 10 and 11, a rise in the rate o f m onetary expansion relative to the initial rate o f inflation has several effects. First, it raises real balances w hich raises aggregate de mand. The rise in aggregate demand raises the rate o f inflation. Second, the rise in the rate o f monetary expansion raises the anticipated rate o f inflation (by coefficient c in equation 5 or 9 above). The rate o f growth o f the nominal wage w ill rise, by the anticipations effect in equation 2 above. This effect w ill not be great because a Surrogate system (estimates from Table 2, rounded) (10.1) D*- = - . 2 T - .4 u T (11.1) Du = ,25ir - .3 u Let the initial conditions, corresponding to points B and C in phase-diagram Figure 8 be as follows for the two systems. SM B t (0 ) u (0 ) * '( o ) -2 2 -2 C 0 -2 Surrogate system B C -2 2 0 -2 0 The trajectories of the inflation and unemployment varia bles are very similar. MARCH/APRIL 1994 184 rise in the current rate o f m onetary expansion w ill convey little information about the rate o f inflation, as is seen in Figure 2. The net effect w ill be that the rate o f inflation w ill rise, as a result o f both the rise in aggregate demand due to the rise in real balances, and the rise in the grow th o f nominal unit labor costs. However, real unit labor costs w ill decline and unemployment w ill decline. These are the short-run effects. As time proceeds, the decline in unemployment and a rise in the rate o f anticipated inflation w ill raise real unit labor costs and the unemploy ment rate w ill converge to its equilibrium rate. Later, w e shall consider the intermediate tar get system, equations 12 and 13, w here the input is the growth o f reserves z. (12) tr(t) = b’0 + b \ U ( t - l ) + b 'M t - 1 ) + b'3 z ( t - l ) + e'; H 0 b ’z + b'3 = 1; b\ < 0 : (13) U(t) = a’0 + a \ U (t - l) + a'2 i r ( t - l ) + a'3 z ( t - l ) + e"; H0 a', + a’3 = 0; a'3 < 0 : We ask in the next section whether, within the context o f the dynamical system, there are eco nomically significant (the neutrality constraints are satisfied), structurally stable, policy-invariant relations equations 10 and 11. W hen the input n ( t - l ) is the grow th o f M2, the answer to all o f these questions is yes, and there is no change in the values o f the coefficients even w hen policy changed drastically. E m pirical Estimates o f the Surrogate System: The Input is the G row th o f M 2 13 Table 2 summarizes the empirical results fo r both equations 10 and 11, w here the input is fj. the grow th o f M2. Column one refers to inflation equation 10, column tw o refers to unemployment equation l l . 1 In each cell is the value o f the 4 regression coefficient and, in brackets, the two-tail significance level. Summary and diagnostic statis tics are at the end o f the table and in the text. 13AII of our data are from the data bank of the Federal Reserve Bank of St Louis, and our software package is MicroTSP® 7.0. 14The last two columns refer to the intermediate target sys tem ^discussed later) where the input is the growth of FEDERAL RESERVE BANK OF ST. LOUIS The Inflation Equation Table 2, column one, describing SM inflation equation 10 indicates that the grow th o f M2 is a good indicator, within the context o f the secondorder dynamical system. The coefficients have the hypothesized and statistically significant signs, satisfy the theoretical constraints, have remarkable structural stability despite changes in policy rules, and this equation has considera ble predictive accuracy. First, each coefficient in column one has the hypothesized sign and is significantly different from zero. The coefficient o f the lagged unem ployment rate b 1 = -0 .39 , w ith a two-tail sig nificance level o f 0.01; the coefficient o f the lagged M2 grow th b3 = 0.21 w ith a significance level o f 0.03. The coefficient o f the lagged infla tion b2 - 0.86 with a significance level o f 0.00. Second, the neutrality requirement is satisfied. The Wald test concerns the neutrality hypothe sis that b2 + b3 = 1: In the steady state a rise in the rate o f m onetary expansion raises the rate o f inflation by the same amount. The sum o f these coefficients is not significantly different from unity: the probability lb2 + b3 = 1] = probl.86 + .21 = 1] = 0.52. Third, there are some mixed results concerning equation evaluation tests. Th ere is no strong evi dence o f serial correlation o f the residuals. The LM/Breusch-Godfrey statistic tests w hether the lagged residuals add to the explanatory pow er o f the equation. The hypothesis that the co effi cients o f all o f the lagged residuals are zero has a probability o f 0.07. Th e Ramsey RESET test in dicated that there seems to be no specification error in the formulation o f the inflation equa tion. The ADF statistic fo r the stationarity o f the residuals was -2.4, which is a bit low to main tain the stationarity hypothesis. The ARCH test statistic allows us to reject the hypothesis o f heteroskedasticity. Fourth, is the issue o f structural stability and predictability, during a period w hen there w ere changes in the policy rule. There is no single, commonly accepted break point fo r the policy rule change. Structural stability is examined in tw o ways, displayed in Figures 4 and 5. W e exa mine w hether the coefficient b3 o f lagged money grow th in inflation equation 10 (Table 2, reserves z. Column three refers to inflation equation 12, and column four refers to unemployment equation 13. 185 Figure 4 Recursive Estimate of the Coefficient of Lagged M2 Growth in Equation 10 Figure 5 Dynamic Ex Ante Forecast of Inflation, Using Lagged M2 Growth as the Input MARCH/APRIL 1994 186 column one) is stationary or w hether it evolves over time and responds to changes in the policy rule. Figure 4 is a recursive estimate o f co effi cient b 3(t) using data through time t. If b 3(t) dis plays significant variation as more data are added (as time increases), it is strong evidence o f instability. I f policy rule changes significantly affect the structure, the coefficient estimates w ill undergo dramatic changes. Figure 4 shows remarkable stability fo r coefficient b 3(t), whereas the velocity series (Figure 3) show significant variation. The other coefficients in equation 10 (T&ble 2, column one) also are quite stable. If the inflation equation using M2 is structur ally stable, it should be useful fo r prediction: Otherwise, M2 is not an indicator. Figure 5 dis plays an AT-period-ahead dynamic forecast. There! is never any correction fo r previous forecast e r rors. The graph INFM2 uses previously predict ed values o f the rate o f inflation as the lagged dependent variable in the next prediction, but uses actual values o f the lagged unemployment rate and rate o f monetary expansion.1 It is 5 necessary to know the state o f the economy measured by U (r - l) as w ell as the rate o f m one tary expansion ^ (t - 1 ) to predict the subsequent rate o f inflation ir(t). A comparison o f the actual rate o f inflation w ith the dynamic ex ante fo re cast using the grow th o f M2 as the input indi cates that the actual rate converges to the predicted rate. Hence, equation 10 is structurally stable, policy-invariant and useful fo r prediction. Compare Figure 5 w ith Figure 2 to see the im portance o f knowing the state o f the economy to predict inflation. A unit root test on the grow th o f real balances (/ - ir) indicated that it is stationary x at a level o f 2.8 percent per annum. That is £(/x - tt) = 2.8 per annum. Since the steady state rate o f inflation ir = i i - n , w h ere n is the long term grow th rate, the estimates are sensible. From Table 2 column one, and the above, the half-life o f the convergence o f inflation to its steady state value ft - 2.8 is 3.47 years.1 6 15This is the FORCST command in MicroTSP®. 16Let the growth of real balances f i- ir be denoted by x. The UROOT equation was Dx=2.1-0.75 x +0.4 Dx(-1). The coefficient 0.75 is significant, UROOT(C,1) = -4 .3 (MacKin non 1 percent = -3.6). Hence, x is stationary and will con verge to the steady-state value 2.1/0.75=2.8, used above. From Table 2, if the unemployment rate is at its equilibrium value, let p be the deviation between the inflation rate and its steady state value: D p = -.2 p (rounding). This implies that the half life is T=log 0.5 / log 0.2 = 3.47 years. FEDERAL http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis For all o f these reasons, w e therefore con clude that, within the context o f difference equation 10: (1) The growth o f M2 is a good in dicator o f inflation, and (2) there is no evidence that policy rule changes had any effects upon the relation betw een money (M2) grow th and in flation in equation 10. The U n em p loym en t Rate Equation and the E ffect o f M 2 G row th We have seen that, within the context o f the SM model, the grow th o f M2 is a good indicator o f inflation. In that equation, the change in the inflation rate depends positively upon the lagged grow th o f real balances which raises the excess demand fo r goods (aggregate demand less cur rent GDP) and negatively upon the state o f the labor market measured by the lagged unemploy ment rate, which reflects the cost-push effects. Even if one knew the path o f the growth o f M2, it would be insufficient to predict the course o f inflation, unless one could also predict the path o f the unemployment rate. The omission o f the unemployment rate is the main reason fo r the poor relation betw een the rate o f inflation and the grow th o f M2 in Figure 2. To understand how the FOMC can achieve price stability and "sustainable grow th in output,” and how M2 grow th affects both inflation and unemploy ment, w e must examine the interactions be tween M2 growth, inflation and unemployment. Table 2, column two, examines the unemploy ment rate equation 11 during the same sample period used fo r the inflation rate. It shows how the rate o f grow th o f M2 affects the unem ploy ment rate and is perfectly consistent with the theory described above. The coefficients are subject to several constraints. The coefficient a5 o f the lagged unemployment rate must be less than unity fo r convergence to the equilibrium rate U e=a0/ ( l - a 1 7 The coefficient o f the lagged ).1 growth o f real balances should be negative, since it produces the rise in aggregate demand fo r goods. This means that the coefficient a, o f lagged inflation should be positive (raise unem17The mean unemployment rate 1957-92 is 6 percent. The estimate of a0 = 1.9 with a standard error of 0.55. The esti mate of 37=0.76 with a standard error of 0.09. If a, = 0.7 and a0=1.8, then Ue is 6 percent. 187 ployment) and coefficient a3 o f lagged m onetary expansion should be negative (lower unemploy ment) and equal to - a 2. The neutrality con straint is fa, + a3 = 0): A rise in the steady state rate o f m onetary expansion w ill produce an equal rise in the rate o f inflation, and no change in the equilibrium unemployment rate. Each coefficient has the correct sign and is significant at the 1 percent level. The neutrality hypothesis is satisfied. The prob[H 0 a2 + a3 = 0] : = probl.29 - .23 = 0] = 0.46 means that m one tary factors cannot affect the steady-state unem ployment rate. However, changes in the lagged rate o f m onetary expansion produce short-run changes in the unemployment rate.1 8 The equation (column two) passes the diagnos tic tests.1 This equation is structurally stable over 9 various policy regimes, and the equation has considerable predictive accuracy. Figures 6 and 7 indicate the predictive value and stability o f the coefficients o f the unemployment equation, despite the many changes in the policy regime. Figure 6 compares the actual unemployment rate with the rate forecasted from a dynamic ex ante simulation, w here previously predicted values o f the unemployment rate are used as the lagged dependent variable, but actual values are used fo r lagged inflation and grow th o f M2. The fo re cast refers to the equation in column tw o in which the input is the grow th o f M2. The actual rate o f unemployment converges to the predic tion. Figure 7 is a recursive estimate o f the coefficient a3 o f the effect o f the lagged rate o f M2 growth. Despite the many changes in the policy rule used by the m onetary authorities, this coefficient is remarkably stable. All o f this evidence suggests that, if the policy variable is the rate o f grow th o f M2, the policy ineffective ness hypothesis is not in evidence. The struc ture o f the model and values o f parameters have been very stable despite changes in the policy rule used by the Federal Reserve, the deregula tion o f financial markets and the high mobility o f international capital. 18These results are inconsistent with the New Classical Eco nomics, but are consistent with basic monetarist (Fried man) views. Notice that we only work with measurable variables and do not use arbitrary and subjective estimates of anticipated or nonanticipated money growth. Belongia points out that the measure of unanticipated money growth is very sensitive to the monetary aggregate considered (as well as to what are the regressors in the equation for antic ipated money growth). W H Y THERE IS NO DIRECT RELATION BETWEEN MONEY GROWTH AND THE RATE OF INFLATION On the basis o f the theoretical and empirical analysis, w e may explain w hy Figure 2 shows no relation betw een the current rate o f inflation and the current or lagged rate o f money growth. From equations 10 and 11, w e derive a phase diagram, Figure 8. From these equations and the coefficient estimates in Table 2 (rounded) columns one and two, derive equations 10.1 and 11.1. The curve die = d(inflation) = 0, which corresponds to equation (10.1), is the set o f un employment rates u(t) = U(t) - Ue and inflation rates ir(t), such that inflation is not changing. The curve d u=d (unem p)=0 is the set o f unem ployment and inflation rates, such that the un employment rate is not changing; and it corresponds to equation 11.1. (10.1) d(inflation) = i r ( t ) - i r ( t - l ) = - 0 . 2 [i r ( t - l ) -H (t -l)l - 0 .4 u (t -l) = 0 (11.1) d(unemp) = u ( t ) - u ( t - l ) = 0 .2 5 [w (t-l) - n ( t - l ) ] - 0 .3 u (t -l) = 0 Let the rate o f money grow th (relative to ca pacity output) be m. Point (m,0) in Figure 8 is the steady state: w here the unemployment rate u = U - U e is zero, and w here inflation is equal to m oney grow th (relative to capacity growth). The curve d(inflation) = 0 is downward sloping fo r the following reason. W hen inflation is be low m, there is a rise in real balances, which raises excess aggregate demand and hence the rate o f inflation, l b keep inflation from chang ing, there must be a rise in u which reduces the cost-push element. The d(inflation) = 0 is nega tively sloped, and the directions o f horizontal motion are towards the curve d(inflation) = 0. The curve d(unemp) = 0 is positively sloped fo r the following reason. Suppose that the prob=0.16 indicates that there is no problem with heter oskedasticity and using the Ramsey RESET test, we do not find any evidence of misspecification. 19There is no evidence of serial correlation. The probability of the F-statistic that all of the coefficients are zero is 0.00, the adjusted R-square=0.76; DW=2.0; ARCH (2 lags) MARCH/APRIL 1994 188 Figure 6 Dynamic Ex Ante Forecast of the Unemployment Rate, Using Lagged M2 Growth as the Input (Equation 11) Figure 7 Recursive Estimate of the Coefficient of Lagged M2 Growth in Equation 11 FEDERAL RESERVE BANK OF ST. LOUIS 189 Figure 8 Phase Diagram u=U-Ue Note: Steady state is point (m,0), where m is growth of M2 less long-term growth of the economy. economy w ere at point m and then the unem ployment rate rose (u > 0). The rise in unem ployment reduces the growth o f nominal labor costs and real unit labor costs tend to decline. This w ill cause unemployment to decline. Tb keep u from changing, aggregate demand must decline. A rise in inflation above m w ill reduce real bal ances which reduces aggregate demand. Th ere fore, the d(unemployment) = 0 curve is positively sloped. The vertical movement w ill be towards this curve, because above (below) it wages are grow ing at a smaller (greater) rate than prices. W ith the phase diagram, w e may answer two questions: (1) W hy do w e find, as in Figure 2, no relation between current or lagged m oney grow th and current inflation? (2) W ill a rise in the rate o f m onetary expan sion, designed to stimulate the economy, lead to higher inflation in the near future? The answer to these questions depends upon w here the economy is situated in Figure 8. There are tw o variables: (1) W hat is the deviation between the rate o f inflation and the rate o f m onetary expansion? W h ere is the economy along the abscissa? (2) W hat is the deviation between the unem ployment rate and its equilibrium value? W here is the economy along the ordinate? From any point, the system w ill converge to point m, w here the unemployment rate is at its equilibrium value, and the rate o f inflation is equal to the rate o f money growth (relative to the trend rate o f grow th o f the economy). The trajectories vary w ith the initial conditions. Given the estimates o f the coefficients in 10.1 and 11.1, the system w ill be damped cyclical.2 0 Consider tw o cases w h ere money grow th is m, but the initial conditions vary. We can explain w hy there is no relation betw een money growth and inflation in Figure 2. Suppose that, when the unemployment rate is above the equilibrium, an expansionary monetary policy is undertaken to accelerate the return to "full employment.” The rate o f m onetary grow th is raised above the inflation rate. The economy starts at point B. 20The characteristic equation implied by 10.1 and 11.1 is A + .5A + .16 = 0 The roots are complex, but the 2 . system is stable. MARCH/APRIL 1994 190 Th e trajectory w ill be BAm. Initially along BA, both the inflation rate and unemployment rate decline. The weakness in the labor market more than offsets the effect o f a rise in real balances upon aggregate demand, and the inflation rate declines. Wages decline relative to prices, and unemployment declines. Along BA, a rise in the rate o f monetary expansion does not lead to more inflation. W hen the economy reaches point A, the low er unemployment rate implies that the weakness in the labor market is insufficient to offset the effect o f a rise in real balances upon aggregate demand, and the inflation rate rises. Prices continue to rise relative to wages, and un employment continues to decline. Along Am, the inflation rate rises though the unemployment rate is above its equilibrium level.2 Along trajec 1 tory BAm, the inflation rate declines and then rises fo r the same rate o f money growth. Similarly, suppose that the economy started at point C, w here inflation is equal to money growth, but unemployment is below the equi librium rate. Nominal wages w ill rise which w ill raise the rate o f inflation. Wages w ill rise faster than prices, and the rise in real unit labor costs w ill increase unemployment. The economy moves along CD. At point D, the rate o f decline o f real balances lowers aggregate demand and offsets the wage-push effect. The rate o f inflation declines, wages continue to grow faster than prices and the unemployment rate continues to rise. Along trajectory CDm, the inflation rate rises and then declines fo r the same rate o f money growth. W e have explained w hy the rate o f money growth is a good indicator o f the rate o f infla tion only within the context o f the dynamic system, equations 10 and 11, w here inflation and unemployment interact. No useful inform a tion about the rate o f inflation is conveyed just by looking at the rate o f m onetary expansion per se as in Figure 2. I f the rate o f m onetary expansion is raised to speed a recovery, this need not im ply m ore inflation in the near 21This differs from the Keynesian NIRU view. See Modigliani and Papademos (1975, 1976). For a critique, see Carlson (1978) and Stein (1982, ch. 4). The analysis differs funda mentally from the New Classical propositions. Neither view is consistent with the results in Table 2. 22The importance of Divisia indices has been developed by Barnett. I am drawing upon Belongia (1993a,b) in the dis cussion of weighted monetary aggregates (WMA), who sup plied me with the data to use as WMA in the SM dynamic model. 23A WMA is constructed as follows (See Belongia). Let U j(t)=[R (t)-rj(t)] / [1 + R(t)], where R(t) is the return on a FEDERAL RESERVE BANK OF ST. LOUIS future. The exact trajectories fo r inflation and unemployment implied by equations 10 and 11, in T&ble 2, columns one and two, are easily calculated. THE USE OF WEIGHTED MONETARY AGGREGATES2 2 Several economists have argued that w e know that the standard measures o f m onetary aggregates violate the basic principles o f the economic theory o f index numbers, because simple-sum measures incorrectly assume that the components are perfect substitutes and, hence, cannot internalize pure substitution effects. Belongia stated that "T h e potential fo r this sort o f [substitution] shift in measured money, o f course, is exactly the type o f thing that may be behind the break in velocity and instability o f m oney demand functions." The contention o f Belongia, Chrystal and MacDonald (this Review) is that ostensible changes in the relationships betw een money grow th and inflation observed in the 1980s, w hich have been subjectively attributed to “ financial innova tions” are simply due to im proper measure ments o f the m onetary aggregate. Instead o f using ad-hoc, arbitrary measures o f the “ true” monetary aggregate, W M A have been con structed to internalize shifts among m onetary aggregates based upon substitution effects. These are basically Divisia indices, by which the components o f the W M A are w eighed by their share o f total expenditure on monetary services.2 3 Their contention is not obvious. Figure 2e, graphs (along w ith the regression line) the rate o f inflation against the grow th o f Divisia M2. There is no apparent relation betw een the two variables. Figure 3b plots the velocity o f Divisia M2 (nominal GDP divided by Divisia M2). The relation does not demonstrate any m ore stability than the velocities o f M l or M2 (Figure 3a). long term grade B corporate bond, rtft) is the asset’s own rate of return. Denote the vector of the u's by u = (u1 ,...,un), and the vector of the value of balances in the i-th asset category by q = (q p ■■ ■,qrJ. The weight S ;(t) of the i-th asset is (b) si (t) = ui(t)qj(t)/u(t).q(t), where the denominator is an inner product. The weighted monetary aggregate WMA is (c) WMA(t)=s(t).q(t). The period denotes the inner product operation. 191 We examine the hypothesis that the W M A are the correct empirical counterparts o f what is meant by money in the theory in the second section24: (1) The money should have the neutrality properties, noted alongside equations (10) and (11) above. A rise in the rate o f m onetary expan sion should produce the same rise in the steady state rate o f inflation. Equal changes in money grow th and inflation should have no effect upon the unemployment rate. (2) Th e W M A should satisfy the requirements fo r an indicator fo r both inflation and unem ployment. It should be able to explain variations in the rate o f inflation and how monetary policy exerts short-run changes upon the unemploy ment rate. Specifically: Given information in year (t-1 ), to what extent can the W M A be used to predict inflation and unemployment in year t? The W M A have the desirable property that they are not arbitrary measures o f “ m oney ness.” They have the limitation that their weights, which are interest rate differentials, are en dogenous variables. W hen a monetary com po nent is changed, the interest rate differentials change. Since the weights in the index change with the endogenous interest rates, the W M A is not a control variable and cannot be considered as an intermediate target. W e already analyzed M2 as an indicator in Table 2 fo r the sample period 1958-92. Table 3 compares three w eighted m onetary aggregates w ith M2 during the same sample period 1961-92, in terms o f equations 10 and 11. The three W M A are used: DM2 = Divisia M2; CE = Rotemberg's currency equivalent; DCE = Divisia currency equivalent. In each case fj.lt) is the rate o f growth (percent per annum) o f the aggregate. Our object is to see how each responds to points 1 and 2 above. Our conclusions, to be discussed, are: (1) The M2 aggregate is the best o f the poten tial indicators. (2) The Divisia currency equivalent DCE is ac ceptable. (3) The Divisia M2 (DM2) and the Currency Equivalent (CE) are unsatisfactory. The upper part o f Table 3 is inflation equation 10, and the low er part is unemployment rate equation 11. The entries are the regression coefficients and the two-tail significance levels in brackets. We also note the adjusted R-square and the probability implied by the LM statistic that there is no serial correlation. Consider the successes. First is M2 in column one. In the inflation equation, the sum o f the coefficients o f lagged inflation and lagged M2 growth (0.87 + 0.18) is not significiantly d iffe r ent from unity. Each coefficient is significant. In the unemployment equation, each coefficient is significant. The sum o f the coefficients o f lagged inflation and lagged M2 grow th (0.28 - 0.22) is not significantly different from zero. Second is the Divisia Currency Equivalent (DCE), which also passes these tests. However, the coefficients in the M2 equation are closer to their theoreti cal values than those in the DCE. Th e co effi cients o f lagged inflation and m oney grow th should be equal and opposite in sign. Next are the failures. The Divisia M2 (DM2) fails in the inflation equation. The coefficient o f its growth jx is not significant. The currency equivalent (CE) fails in the unemployment rate equation. The coefficient o f its grow th / is not z significant. M y conclusion is that M2 is the best o f the indicators w hen it is used in the dynamic SM model, in which both unemployment and inflation interact. A cogent analysis o f the deficiency o f Divisia indices o f money has been given by Otmar Issing o f the Deutsche Bundesbank (1992, p. 296). He wrote: "In phases with an interest rate pattern in which the yield on time deposits is almost that on the yield on public bonds outstanding, time deposits to all intents and purposes disappear from the defini tion of the money stock (CE aggregates) or hardly contribute at all to money stock growth (Divisia Aggregates). Should time deposit rates exceed the yield on bonds outstanding, then this leads to either negative growth of these aggregates or the changed maximum interest rate is taken into con sideration so that monetary capital components possibly contribute to growth in the money stock. The reason here is that — based on a utility max imization approach — liquidity is measured in 24lt is essential that one have a macroeconomic theory to evaluate whether an empirical measure of money cor responds to a theoretical concept. Barnett, Belongia and others correctly object to the ad hoc measures of “ money ness” that have been offered to replace M2. Many of these measures even fail to satisfy the neutrality requirement. MARCH/APRIL 1994 192 Table 3 Inflation and Unemployment Equations Using Alternative Measures of Money Inflation equation 13 Constant U(t— 1) 7r(t-1) Mi(t~1) ADJ-RSQ LM-prob i=lM2 1.88 -0.44 0.87 0.18 0.79 0.14 Unemployment equation 14 Constant 1.60 0.81 U(t— 1) 7r(t-1) 0.28 -0.22 w fl-1 ) ADJ-RSQ 0.82 LM-prob 0.80 G rowth rate o f the m onetary aggregate i i = DM2 i = DCE [0.03] [0.00] [0.00] [0.045] 2.15 -0.39 0.90 0.13 0.76 0.07 [0.02] [0.01] [0.00] [0.12] 2.44 -0.46 0.97 0.10 0.78 0.17 [0.00] [0.00] [0.00] [0.07] [0.00] [0.00] [0.00] [0.00] 1.19 0.75 0.26 -0.15 0.81 0.76 [0.04] [0.00] [0.00] [0.007] 0.92 0.80 0.18 -0.10 0.78 0.75 [0.10] [0.00] [0.00] [0.008] L U O n Variable 1.78 -0.19 0.85 0.027 0.78 0.12 [0.05] [0.21] [0.00] [0.06] 0.84 0.68 0.24 0.001 0.71 0.65 [0.23] [0.00] [0.00] [0.92] Notes: The sample covers 1961-92. N = 32. The two-tail significance level is shown in brackets. The data are from the Federal Reserve Bank of St. Louis. DM2 = Divisia M2; DCE = Divisia currency equivalent; CE = currency equivalent. terms of forfeited yields, while the dimension of risk — contrary to the portfolio optimization ap proach — is not taken into account. The interest rate for a particular form of investment not only contains a premium for foregoing liquidity but also a risk premium owing to yield volatility. As empirical studies show, in particular the CE-M3 ag gregate has in the past been subject to extreme fluctuations and the correlation with growth rates of GNP was in fact negative. Furthermore, the ve locity of circulation of this aggregate was substan tially more instable (sic) than that of M3.” sues: lb what extent is money grow th endo genous? lb what extent is money grow th controllable? In equation 14, part CX is endogenous, X is a vector o f the state o f the economy and z is the grow th o f reserves.2 5 Unless variations in fi are controllable, they are not responsible fo r variations in inflation and unemployment; and the central bank does not have the w herew ithal to control inflation in the medium run.2 6 (14) The Divisia M2 is too much dependent upon en dogenous weights, which are interest rate differentials, to be useful as an indicator o f a theoretical concept o f money. It misses the unique aspect o f m oney that it is the safe asset used as the medium o f exchange. The Controllability o f M o n e y G row th ij 2 l - CX + bz + e We can relate total reserves R to M2. There is a close relationship between reserves R and M l, through a system o f reserve requirements. Call the reserve requirement ratio R/Ml=a. W e can then write: R/M2 = (R/Ml) (M1/M2) = a (M1/M2) and therefore, We have shown that grow th o f M2, denoted ^ 2, is a good indicator. Th ere are tw o distinct is 25For notational simplicity, let e generically represent the ran dom variable with a zero expectation. 26ln the long run, as Figure 1 indicates, the price level is still closely tied to M2/real GDP. FEDERAL RESERVE BANK OF ST. LOUIS log M2 = log (M2/M1) + log R - log a. 193 Figure 9 Ratios of M2 to the Adjusted Monetary Base and Adjusted Reserves - 45 - 40 - 30 ■ i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i “ 15 195658 60 62 64 66 68 70 72 74 76 78 80 82 84 86 88 90 1992 The rate o f change o f M i (1 = 1, 2) is denoted fi: and the grow th o f reserves is denoted z. Thus, w e have equation 15, which w e relate to equa tion 14: (15) n2 = f/x, - fx j + z + e (14) = CX + bz + e The growth o f M2 in equation 15 has three components: the growth in adjusted reserves z, which is controllable; the grow th in the non-M l component o f M2, which is (n2 - n j ] and the e term, which reflects nonsystematic factors.2 7 Thornton noted several points. First, the Fed has a tight control on M l = fl/a via reserves. Se cond, the ratio o f M l to M2 declined from 0.5 in 1959 to 0.25 in 1977, and has then fluctuated around this level. Third, the policy variable, which is the grow th o f reserves, does not have a significant effect upon (/z, - /xj. The Fed can control only the M l component o f M2 but can not control the other component (fi, - fxj directly. For example, suppose that a rise in fis cal policy or private demand tends to raise the grow th o f nominal GDP, which induces a growth in the demand fo r money. The given grow th o f reserves controls the grow th o f M l. Th ere w ill be a grow th in M2 relative to M l to accommo date the induced rise in the demand fo r money. This means that CX = fyt, - n j is endogenous; and it may well be the major source o f variation o f the rate o f m oney grow th in equation 15. Figure 9 suggests that there has been a structural break in the controllability o f the grow th o f M2. The graph is the ratio o f M2 to adjusted reserves. It has a relatively constant positive trend until 1975. The trend rises drasti cally to about 1984. Then it falls to zero or be comes negative. A similar situation exists with the ratio o f M2 to the adjusted monetary base. We shall now be m ore precise. We consider tw o components o f money growth in equation 16 which correspond to 14 and 15. (16) nz = c' D N G D P (-l) + bz + e The control part is the controllable growth o f reserves. The induced part is related to the lagged grow th o f nominal GDP, denoted 27The controllability of M2 is the subject of the important paper by Thornton (1992), upon which we draw. MARCH/APRIL 1994 194 Table 4 The Rate of Growth of M2 as a Function of the Lagged Growth of Nominal GDP and the Growth of Adjusted Reserves (Equation 16)________________ Variable Constant DNGDP(-1) z DW ADJ R-SQ 1958-92 1958-75 1975-92 3.60 [0.02] 0.31 [0.05] 0.27 [0.06] 1.10 0.12 3.40 [0.01] -0.04 [0.71] 0.90 [0.00] 1.69 0.64 2.10 [0.45] 0.56 [0.04] 0.17 [0.39] 1.10 0.14 Note: The two-tail significance level is shown in brackets. DNGDP ( —l).2 The induced part corresponds to 8 CX in equation 14. That means that if the omit ted fiscal variables and shocks to private de mand induce a rise in the demand fo r money, the grow th o f M2 w ill respond, although the grow th o f M l is tied to the grow th o f reserves. Th ere are several implications from Table 4, which are consistent with Thornton’s findings.2 9 First, consider column two, which concerns the early period 1958-75. The growth o f reserves is the significant determinant o f the growth o f M2, w ith a coefficient 0.9, which is not significantly different from unity. The grow th o f lagged nomi nal GDP is not significant.3 A regression o f 0 on z and a constant gives almost the identical results, during the first period. Hence one could confidently claim that /z, = c + z + e, w here c is a trend which corresponds to the grow th o f M2/M1. The m oney supply was both controllable and the controllable part was the dominant component. Hence from 1958 to 1975, the grow th o f M2 was an intermediate target for both the inflation and unemployment rates. This is "Monetarism Triumphant.” Second, consider column one containing the entire period 1958-92. It would seem that both 28The lag is used to avoid a simultaneous equation problem. We also used for the induced part the Treasury bill and Treasury bond rates, which could reflect changes in the structure of interest rates, which would ultimately induce substitutions between M1 and M2. However, they were not significant additions to the growth of reserves. 29Similar results were obtained when the regressors were the lagged unemployment and inflation rates. 30The equation for this period passes all of the equation evaluation tests: There is no serial correlation (LM test), heteroskedasticity (ARCH test). RESERVE BANK OF ST. LOUIS FEDERAL the growth o f reserves and the grow th o f nom i nal GDP are significant. However, the equation evaluation tests tell a different story. The recur sive residuals (not shown) keep m oving outside the plus-or-minus-2 standard error bands, which implies that the structure is progressively chang ing. Figure 10 is a recursive estimate o f the coefficient o f the growth o f reserves. This co efficient has a clear downward trend from unity towards zero, indicating that the control part is becoming less significant since the mid-1970s. The reason is shown in column three containing the period 1975-92. This column is a direct con trast to column two, the 1958-75 period. The growth o f reserves is not significant. Th e lagged growth o f nominal GDP is significant. However, the regressors only explain 14 percent o f the variation in the growth o f M2. During the peri od 1975-92, it is not apparent that the grow th o f M2 was an intermediate target. THE INTERMEDIATE TARGET SYSTEM It is quite possible that w e have omitted sig nificant variables from the induced part CX o f money growth, so that it seems that money grow th is no longer controllable by the growth 195 Figure 10 Recursive Estimate of the Coefficient of the Reserves Growth in Equation 16 Figure 11 Dynamic Ex Ante Forecast of Inflation, Using Lagged Resources Growth as Input (Equation 12) ■ MARCH/APRIL 1994 196 Figure 12 Recursive Estimate of the Coefficient of Lagged Reserves Growth in Equation 12 o f reserves. Th e control equations involving the grow th o f reserves w ere described in the se cond flo w chart, shown here again: X < ----- DX = (A+BC)X + (Bb)z + (Be" + e ') < -------------------z intermediate target control variable A direct test o f controllability is equations 12 and 13, in the surrogate system, estimated in Table 2, columns three and four: Call this the control system. The next two sections show that the controllable growth o f reserves may be a good intermediate target for the rate o f inflation. The subsequent section shows that the short-term Treasury bill rate, which may be controllable, has no informational content; it is neither an indicator nor an inter mediate target. Interest rate targeting, which has had disastrous results both in the Great Depression and the pre-1979 periods, is to be avoided at all costs. (12) ir(t) = b ’ + b 'J U (t-l) + b 'j j f t - l ) 0 + b'3 z ( t - l ) + e ; H K + K = 2; b: < (> „: (13) U(t) = a' + a [ U ( t - l ) + a 'M t-1 ) + a'3 z ( t - l ) + e ; H0 a' + a' = 0; a'3 < 0 : 31We did not use the growth of the adjusted monetary base as the control variable for two reasons. First, it failed to satisfy the neutrality requirement. Second, it is not a relia ble control over the growth of M1 due to the significant var iations in the currency ratio. See Garfinkel and Thornton. FEDERAL RESERVE BANK OF ST. LOUIS The Grow th o f Adjusted R eserves Is A n Interm ediate Target3 1 In the control system, the control input is z the growth o f adjusted reserves. This variable is 197 clearly controllable.3 W e evaluate w hether the 2 control system is structurally stable and policyrule-invariant, w hen there have been changes in Federal Reserve operating procedures and policy, and financial market deregulation. Table 2, columns three and four, and the subsequent analysis show that the control system is quite significant fo r the inflation rate but less so for the unemployment rate. The Inflation Equation in the S M M o d e l with the Grow th o f Adjusted R eserves The inflation equation is 12. The rate o f infla tion rises when (1) real reserves rise, the growth o f reserves exceeds the current rate o f inflation, or (2) when the labor market is tight, the unem ployment rate is below its equilibrium rate. In the steady state, the rate o f inflation w ill rise by the same amount as the rise in the growth o f reserves. Real reserves converge to a constant. This is the neutrality hypothesis b'2*b'3 = 1 in 12. The second factor states that the coefficient o f the lagged unemployment rate is negative. from Table 2, column three, denoted INFRES. The large deviations fo r the 1977-80 period are corrected by 1986; and the model is back on track. Second, Figure 12 displays the structural stability in a clear and dramatic way. It is a recursive estimate o f coefficient b'3 which relates the effect o f a change in z ( t - l ) the grow th o f reserves in year t - 1 upon ir(t) the rate o f infla tion in year t, given the initial values o f unem ployment and inflation. This coefficient is fairly stable, despite the changes in policy regim e over the period. The conclusion is that the grow th o f adjusted reserves is an intermediate target for achieving price stability, within the context o f the dynamic equation. The U n em p loym en t Rate Equation with the Grow th o f Adjusted R eserves Table 2, column three, is consistent w ith these hypotheses. Each coefficient is significant and has the hypothesized sign.3 The Adj. R-SQ=0.78. 3 The neutrality hypothesis is confirmed. It is seen w ith a Wald test that the sum o f the co effi cients o f the inflation and growth o f reserves is not significantly different from unity: prob lb ' + b'3 = 1] = prob [0.92 + 0.16 = 1] = 0.44. The inflation equation is not sufficient to an swer the question: H ow can the central bank achieve price stability and "prom ote sustainable growth?” The reason is that the inflation rate is affected by the state o f the unemployment rate as well as by its past history and the grow th o f reserves. Attempts to reduce the rate o f infla tion by varying the grow th o f reserves w ill af fect, in the medium run, the unemployment rate. In turn, the unemployment rate w ill affect the inflation rate. Another dimension to this problem concerns w hether m onetary policy can also affect, in the medium run, the unemploy ment rate, and w hat w ill be the consequences fo r the rate o f inflation? W e show in several ways that this equation is structurally stable and policy-invariant. First, Figure 11 compares the actual rate o f inflation w ith a dynamic ex ante forecast derived We turn to equation 13 in Table 2, column four, to see to what extent the growth o f reserves affects the unemployment rate. Table 3, column four, is consistent with several hypotheses. First, 32We believe that the growth of reserves is controllable and has not been an endogenous variable, even in the 1979-82 period when there was fairly explicit interest rate targeting. If the growth of reserves were endogenous, then it should be responding to the growth of nominal GDP and the value of the Treasury bill rate. A rise in the growth of nominal GDP, given the Treasury bill rate, should increase the de mand for reserves and induce a greater supply. Similarly, given the growth of nominal GDP, a decline in the Treasury bill rate should induce a decline in the growth of reserves to force the treasury bill rate up to a desired level. We exa mined the issue of whether the growth of reserves (DRES=z) has been an endogenous variable by regressing it upon the lagged growth of nominal GDP [DNGDP(-1)] and the lagged Treasury bill rate [TB3(-1)], to avoid a simultaneous equation problem. The sample period is 1959-92. The dummy variable (DUM) was set at DUM=1 during the 1979-82 period, and DUM=0 otherwise. We constructed two variables, DUM*DNGDP and DUM*TB3, to highlight the short period of interest rate targeting. The regression equation was (14) DRES = 5.13 -0.25 7 DNGDP(-1) + 0.40TB3(-1) (t-stat) (2.7) (-1-09) (1.3) -0.48'DUM'DNGDP(-1) + 0.118DUM'TB3(-1) (-0.77) (0.22) ADJ R-squared = 0.00. No coefficient is significant and there is no evidence that the growth of reserves has been an endogenous variable in any significant way during the period 1959-92. 33There is no evidence of either serial correlation (LM test prob=0.18) or heteroskedasticity (ARCH test prob=0.49). The Ramsey RESET test of whether there are omitted vari ables, incorrect functional form or correlation between the regressors and the error term indicates that the probability that there is no specification error is 0.38. MARCH/APRIL 1994 198 Figure 13 Dynamic Ex Ante Forecast of the Unemployment Rate, Using Lagged Resources Growth as the Input (Equation 13) Figure 14 Recursive Estimate of the Coefficient of Lagged Reserves Growth in Equation 13 FEDERAL http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis 199 each coefficient has the correct sign and is sig nificant at the 1 percent level; and the R2= 0.71.3 4 Second, the neutrality hypothesis is confirmed. W hen reserves rise at the same rate as the rate o f inflation, there is no effect upon the unem ployment rate, which would then converge to its equilibrium rate. Hence, a' + a' = 0, that is, the coefficients o f inflation and the grow th o f reserves sum to zero. Using a Wald test, the Prob /a'+a' = 07 = Probl0.23 - 0.13 = 0] = 0.23, thereby confirm ing the neutrality hypothesis. makes any economic sense, the interest rate should be a significant input into the dynamic inflation and unemployment rate equations, either by itself [6 = 0] or as additional inform a tion [6 = 1] to the grow th o f M2, in equations (12.1) and (13.1). (12.1) -reft) = Cj + c2U ( t - l ) + c3i r ( t - l ) + 6c4 ( t - l ) + c j j f t - l ) + e n (13.1) U(t) = c; + c'2U ( t - l ) + c'3i r ( t - l ) Third, the explanatory pow er o f this equation is much less satisfactory than the inflation equa tion w here the input is the growth o f M2. In Figure 13, the actual unemployment rate is com pared w ith a dynamic ex ante simulation o f the value implied by the coefficients in lable 2, column four, w here the lagged dependent varia ble is the previously predicted value. The fo re cast predicts basic trends but gives misleading predictions o f the level o f the unemployment rate. Fourth, Figure 14 plots the recursive estimate o f the coefficient a'3 < 0 o f the lagged growth o f reserves. The absolute value o f this co effi cient has been diminishing over the sample period.3 A possible reason fo r the decline in im 5 portance o f the grow th o f reserves in the unem ployment equation may be that the growth o f reserves has becom e a less important deter minant o f money grow th (Figure 13) in a period w hen the non-M l component o f M2 has become more important. The Treasury Bill Rate is N o t an Interm ediate Target: It Adds N o Useful In form a tion The Federal Reserve has revived the issue o f interest rate targeting, w here the Treasury bill rate is an intermediate target. Is there evidence to support interest rate targeting? Th e Treasury bill rate, denoted is controllable. Hence, it should be used to evaluate interest rate target ing. The surrogate dynamic SM model implied equations 12 and 13. I f interest rate targeting 34There is no evidence of serial correlation (LM test prob=0.72) nor of heteroskedasticity (ARCH test prob=0.64). According to the RESET test, there is no evi dence of misspecification (RESET test prob=0.12). 35l do not have an explanation why the coefficient a3< 0 in Figure 9 is stable, but a'3<0 is not in Figure 14. + 6c'4 i . ( t - l ) + c U J t-1 ) + e' i Since the rate o f inflation is a regressor, a rise in the nominal interest rate in the regression corresponds to a rise in the observed real rate.3 6 Table 5 describes the results o f such a test. Column one is the inflation equation, which just uses the Treasury bill rate as a control [6 = 0]. It is seen that the coefficient o f the Treasury bill rate is not significant. It contains no additional information about what w ill happen to inflation. Column tw o adds the growth o f M2 as an input [6 = 1], The grow th o f M2 is highly significant (as it was in Tkble 2), and the Treasury bill rate remains insignificant. The conclusion here is that adding the Treasury bill rate adds no in for mation about what w ill happen to inflation. Columns three and fou r concern the unem ployment rate. In column 3 [6 = 0], the results are bizarre. The coefficient o f the nominal in terest rate is not significant at the 5 percent level, and the coefficient o f inflation is not sig nificant. Given the nominal interest rate, a rise in the rate o f inflation corresponds to a decline in the real rate o f interest. This should lower the unemployment rate, but it does not. Th ere fore, it would appear that real interest rate tar geting is not promising. In column four, w e add the rate o f M2 grow th [6 = 1]. The results turn sensible fo r everything but the Treasury bill rate, which continues to remain insignificant. The conclusion is that the Treasury bill rate at ( t - 1 ) adds absolutely no information to what is obtained from the results in 'lable 2. use estimates of anticipated inflation that cannot be objec tively justified. It is not clear whether the spread between the bond rate and Treasury bill rate is a more or less ac curate measure of anticipated inflation than is the recent ex post inflation. In either case, the onus of finding the true ex ante real rate is upon the advocates of interest rate targeting. 36lf there is real interest rate targeting, the only available in formation concerns observed, not anticipated, rates of infla tion. It requires prescience for the monetary authority to MARCH/APRIL 1994 200 Table 5 Inflation and Unemployment Rate Equations_________________ liti-D 1.99 [0.02] 1.39 U(t) 16= 1 1 -0 .2 4 [0.12] -0 .3 9 s 16=01 II tW t Equation 13.1 oi; ii Equation 12.1 Variable Constant U ( t-l) ■K(t-l) ilt-1 ) [0.12] 1.27 [0.05] 1.89 [0.00] [0.018] 0.56 [0.00] 0.72 [0.00] 0.226 [0.01] 0.95 [0.00] 0.858 [0.00] 0.12 [0.18] -0 .0 5 9 [0.61] -0 .0 0 8 [0.94] 0.15 [0.07] 0.079 [0.27] -0 .2 2 [0.00] Note: Sample period is 1958-92. N=35. The two-tail significance is shown in brackets. The Transm ission M echanism Th ere is a good reason w hy the Treasury bill rate is neither an indicator nor an intermediate target. This concerns the transmission mechan ism. Aggregate investment demand depends upon the Keynes-Tobin (/-ratio. M onetary policy exerts its effects upon the economy through this ratio. Keynes (1936, p. 151) explained the theory o f the (/-ratio: "...the daily revaluations o f the stock exchange, though they are prim arily made to facilitate transfers o f old investments between one individual and another, inevitably exert a decisive influence on the rate o f investment. For there is no sense in building up a new enter prise at a cost greater than that w hich a similar enterprise can be purchased; w hile there is an inducement to spend on a new project what may seem like an extravagent sum, if it can be floated o ff on the stock exchange at an immedi ate profit.”3 7 Formally, let q'.k be the market value o f k the existing capital and let p.k be the reproduction cost.3 Their ratio is the (/-ratio. 8 (17) q = q'.k / p.k The portfolio balance equation 18 is that the ra tio o f money to the market value o f capital M/q'.k depends upon L(i) w h ere i is a vector o f opportunity costs i= (i1 ,-.,in), and element / is . the perhaps controllable Treasury bill rate. Solve equation (18) fo r q=q'.k / p.k, which is associated with portfolio balance and obtain 19. Denote 37The role of financial markets in capital formation, along these lines, is the theme of Stein (1987, ch. 7; 1991, ch. 3). 38The period represents an inner product. Variables q\ k and p are vectors of market prices, physical quantities and reproduction costs, respectively. Capital and bonds are in vector k and the weighted sum is q'.k. This is definitely in the spirit of Keynes and Tobin. FEDERAL http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis m = M/p.k, the ratio o f real balances per unit o f capital. (18) M/q'.k = M / q pk = M/pk / q = Lti) (19) q = [M/p.k] / L (i) = m / L(i) M onetary policy, which changes reserves, operates as follows in the context o f equation 19. Let there be a rise in real bank reserves, w hich is a control variable. The higher ratio o f reserves to deposits induces banks to purchase financial assets, equity or debt. The greater w ill ingness to lend induces their customers to b o r row to purchase equity and debt. The money stock rises. Given the vector o f expected returns i on the n-assets, the prices o f existing assets, real and financial, rise. This is a rise in the Keynes-Tobin (/-ratio, the ratio o f the prices o f existing assets (stock prices, bond prices, physi cal plant), relative to their reproduction costs. This encourages the production o f investment goods and raises the excess demand fo r goods relative to current GDP.3 This is the logic o f 9 having m in equation 19 above: It reflects the qratio effect. The rise in q = q'.k/p.k need not be reflected in the Treasury bill rate or in any par ticular interest rate. Changing the Treasury bill rate without changing the grow th o f M2 has a negligible effect upon the (/-ratio, whereas changing the money stock has a large effect, assuming that both are controllable. Interest rate targeting o f the Treasury bill rate pro vided a misleading indicator o f w hat has been 39This is not the textbook transmission mechanism, but it is the one stressed by Keynes, Tobin and Friedman. 201 happening to the (/-ratio, or the stance o f m one tary policy as was stressed by Friedman and Schwartz in their account o f the Great Con traction.4 0 CONCLUSIONS In the long run, the GDP deflator is closely related to the quantity o f M2 per unit o f real GDP.4 The question examined in this paper con 1 cerns how the Federal Reserve should select ranges fo r m onetary grow th over the coming year to achieve a given rate o f change in the price level in the near future. Our conclusions w ere stated as propositions A-D at the beginning o f the paper. Friedman does not think that the inflation rate can be controlled finely: "...we cannot predict at all accurately just what effect a particular m onetary action w ill have on the price level and, equally important, just when it w ill have that effect. Attem pting to control directly the price level is therefore likely to make monetary policy itself a source o f econom ic disturbance because o f false stops and starts ...Accordingly, I believe that a m onetary total is the best currently available immediate guide or criterion fo r monetary policy—and I believe that it matters much less which particular total is chosen than that one be chosen (1969, p. 1089)...there seems little doubt that a large change in the money supply within a relatively short period w ill force a change in the same direction in income and prices...But w hen the money changes are moderate, the other factors come into their own. If w e knew enough about them 40One of the most vivid examples of the dangers of interest rate targeting, inspired by Friedman and Schwartz, is shown below, which compares 1929 with 1932. The data are from the U.S. Department of Commerce; the appropri ate series are noted. The first row is the S&P index (B85), the second row is the implicit price deflator P' for fixed in vestment (B68), the ratio of the two is an index of the qr-ratio. The fourth row is i1, the Treasury bill rate (B83). The variable /2 is the basic yield of 30-year corporate bonds (B75). Variable i3 is the Manhattan real estate mort gage rate (B78). The row labelled P is the implicit GNP deflator (B63) and M is the money supply (B110). The aver age annual rate of growth of P and M is in square brackets in the 1932 column. The movement in the treasury bill rate was a misleading measure of the extent that the qr-ratio changed. and about the detailed effects o f monetary changes, w e might be able to counter these e f fects by m onetary measures. But it is utopian given our present level o f knowledge. Th ere are thus definite limits to the possibility o f any fine control o f the general level o f prices by a fine adjustment o f m onetary change.” (p.181) Friedman’s argument should be qualified, in view o f the analysis in this paper. First, the choice o f the m onetary aggregate does matter. No aggregate has the same quality o f explana tory power as does M2, w ithin the context o f the dynamical system. Second, there is a serious question w hether the grow th o f M2 is controlla ble. From 1958 to 1975, the growth o f M2 was controllable. The equation fo r its grow th was a constant (which is the trend) plus the grow th o f reserves plus an error. From 1975 to 1992, the link betw een the grow th o f M2 and the growth o f reserves was no longer apparent. W hat should be the Federal Reserve’s control policy, since the link betw een M2 grow th and reserve grow th after 1975 is not apparent? We concluded that: (1) The grow th o f M2 is a good indicator within the context o f the dynamic model. However, it is doubtful that it is controllable in the medium run. (2) The Federal Reserve should place greater weight upon its control o f inflation, than upon the attempt to fine-tune the economy, because the inflation equation in the reduced form sys tem has more stability and predictability than does the unemployment equation in that system.4 2 4 See Figure 1. There is also a close long-run relation 1 between M2 and the quantity of adjusted reserves, and, hence, a long-run relation between the GDP deflator and the ratio of adjusted reserves per unit of real GDP. These relationships look similar to Figure 14. However, none of these three relationships passes the usual cointegration tests. 42Hall (p. 278) wrote: “ I conclude that established models are unhelpful in understanding this recession [1990-92] and probably most of its predecessors.” Insofar as the growth of M2 was controllable prior to 1975, the SM dynamic model does explain the recessions. See Figure 6 above. However, after 1975 it is not clear that the growth of M2 is controllable. Hence, the good fit in Figure 6 after 1975 does not contradict Hall. The Great Depression Period 1932 1929 variable 26.02 6.93 S&P index 39.4 31.6 Price investment good 0.22 0.66 qr-ratio index 0.88% pa 4.42% pa treasury bill (i1) 4.22 4.7 30 yr corp (i2) 5.75 5.92 mortgage rate (i3) 0.2[-7.67% pa] GNP deflator P 50.6 26,419 20,689[-8.15% pa] Money stock MARCH/APRIL 1994 202 (3) Friedman’s admonitions concerning fine tuning with respect to money, which is not obvi ously controllable, should apply to fine tuning o f the reduced form system using the controllable grow th o f reserves. Mathematically Friedman’s argument is that given the uncertainty concern ing the values o f the parameters in T&ble 2 as reflected in their standard errors, the central bank should be most reluctant to vary its con trol variable in pursuing its objective o f price stability lest grow th be adversely affected. However, an optimal control policy in this con text has not as yet been established.4 3 Garfinkel, Michelle, and Daniel L. Thornton. “ The Link Between M1 and the Monetary Base in the 1980s,” this Review (September/October 1989), pp. 35-52. Greenspan, Alan. 1993 Monetary Policy Objectives: Summary Report of the Federal Reserve Board. Board of Governors of the Federal Reserve System, 1993. Hall, Robert E. “ Macro Theory and the Recession of 1990-91,” The American Economic Review (May 1993), pp. 275-79. Infante, E.F., and Jerome L. Stein. “ Money Financed Fiscal Policy in a Growing Economy,” Journal of Political Economy (April 1980), pp. 259-87. Issing, Otmar. “ Theoretical and Empirical Foundations of the Deustche Bundesbank’s Monetary Targeting,” Inter economics (November/December 1992), pp. 289-300. Keynes, J.M. The General Theory of Employment, Interest and Money. Harcourt Brace, 1936. REFERENCES Barnett, W.A., E. Offenbacher and P Spindt. “ The New . Divisia Monetary Aggregates,” Journal of Political Economy (December 1984), pp. 1049-85. Beiongia, Michael T. “ Measurement Matters: Recent Results from Monetary Economics Re-examined,” mimeo, Universi ty of Mississippi, 1993a. _______ . “ Consequences of Money Stock Measurement: Evi dence from Three Countries,” presented at meetings of the American Statistical Association, August 8-10, 1993b. _______ , and Dallas S. Batten. “ Selecting an Intermediate Target for Monetary Policy When the Goal is Price Stabili ty,” Federal Reserve Bank of St. Louis Working Paper 92-008A (October 1992). Carlson, Keith. “ Inflation, Unemployment and Money: Comparing the Evidence from Two Simple Models,” this Review (September 1978), pp. 2-6. Friedman, Milton. The Optimum Quantity of Money, and Other Essays. Aldine Publishing Co., 1969. _______ , and Anna J. Schwartz. A Monetary History of the United States 1867-1960. Princeton University Press, 1963. 43The reason is that the coefficients of the dynamical system — equations 11, 12 and Table 2 (columns three and four) — are stochastic, with significant standard errors which do not go to zero as the sample size increases. This problem is being studied at present by Wendell Fleming (Depart ment of Applied Mathematics, Brown University) and the author. FEDERAL http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis Modigliani, Franco, and L. Papademos. “ Monetary Policy for the Coming Quarters: The Conflicting Views,” Federal Reserve Bank of Boston New England Economic Review (March/April 1976), pp. 2-35. _______ , a n d ________ Targets for Monetary Policy in the Coming Year, Brookings Papers 1, (1975), pp. 141-63. Ritter, Joseph A. “ The FOMC in 1992: A Monetary Conun drum,” this Review (May/June 1993), pp. 31-49. Stein, Jerome L. “ Cobwebs, Rational Expectations and Futures Markets,” Review of Economics and Statistics (February 1992a), pp. 127-34. _______ . “ Price Discovery Processes,” Economic Record, special issue (1992b). ________International Financial Markets. Blackwell, 1991. _______ . The Economics of Futures Markets. Blackwell, 1986. _______ . Monetarist, Keynesian and New Classical Econom ics. New York University Press, 1982. Thornton, Daniel L. “ Targeting M2: The Issue of Monetary Control,” this Review (July/August 1992), pp. 23-35. United States Department of Commerce. Long Term Economic Growth 1860-1965. Bureau of the Census, U.S. Government Printing Office, 1966. 203 Appendix Use of Quarterly Data in Estimating the Dynamic SM Model The results in the table below indicate w hy w e used annual data in our empirical analysis. Inflation is measured relative to the previous quarter, but at an annual rate. The grow th o f M2 is measured in the same manner. The e f fects build up over time and quarterly move ments per se have no significance. Consider first columns one and two, which correspond to equations 10 and 11. In the infla tion equation (column one) only the lagged de pendent variable is significant at the 5 percent level. The lagged money grow th is significant at the 8 percent level, but the equation fails to satisfy the neutrality constraint. Theoretically, in an equation such as 10, regardless o f the time span, the sum o f the coefficients o f lagged inflation (bo= 0.75) and lagged money growth (b3= 0.08) should sum to unity. The null hypothe sis that b2+ b3= l has a probability level o f 0.014; hence, the neutrality (null) hypothesis is reject ed. In addition, there is very serious serial correlation o f the residuals. Th e LM statistic, using three lags, w here the null is no serial correlation, has a probability o f 0.00. The ARCH test rejects homoskedasticity at the 3 percent level. Column tw o relates to equation 11. A t first glance, the results are significant. However, there are difficulties. First, the coefficient o f the lagged unemployment rate (0.98) is not signifi cantly different from unity, and the constant is not significantly different from zero. Thus, if in flation equals money growth, the unemployment rate converges to zero. Second, there is serious serial correlation o f the residuals. Using lags up to tw o quarters, the LM test o f no serial correla tion has a probability o f 0.00. Third, the ARCH test o f no heteroskedasticity has a probability o f 0.00. So the unemployment equation in column tw o fails using these diagnostics. The conclusion is that w e cannot have confidence in the results o f columns one and two. Columns three and four consider tw o lags of inflation and money growth, w here time is measured in quarters. This means that a span of half a year is being considered. The main results are that nothing o f significance, other than the effects o f its own lagged variable, is ap parent by focusing upon quarters rather than upon annual data. The only significant variables in the inflation equation (column three) are the lagged inflation rates one and tw o quarters. The one-quarter lagged money grow th is not signifi cant. The lagged two-quarter money grow th is significant at the 8 percent level. So, nothing much shows up within tw o quarters. Second, in the unemployment rate equation (column four), the lagged dependent variable is significant. In flation during the previous tw o quarters is not significant. The money growth in the previous quarter is not significant. However, the money grow th tw o quarters earlier is significant. Com pare T&ble 2 in the text w ith the table above. These are the reasons w hy w e used annual data in the analysis in the paper. Table A1 Inflation it and Unemployment U Equations (10-11) Equation 10 Variable Constant U(t-1) T (t-2 ) A t- 2 ) 7T 0.91 [0.12] -0.07 [0.50] 0.75 [0.00] -----0.08 [0.083] ------ Equation 11 U 0.21 [0.12] 0.98 [0.00] 0.036 [0.00] ------0.032 [0.00] ------ Equation 10 7T 0.76 -0.15 0.42 0.44 0.01 0.09 [0.16] [0.11] [0.00] [0.00] [0.84] [0.08] Equation 11 U 0.26 [0.04] 0.99 [0.00] 0.02 [0.21] 0.03 [0.14] -0.002 [0.86] -0.05 [0.00] Notes: Quarterly data. M2 Growth is the input. The sample is 1956:4-1992:4 The two-tail significance is shown in brackets. MARCH/APRIL 1994 204 Frederic S. Mishkin Frederic S. Mishkin is the A. Barton Hepburn professor of eco nomics, Graduate School of Business, Columbia University. He is also a research associate, National Bureau of Economic Rsearch. Commentary j L HE TIT LE OF THE PAPER by Jerry Stein is somewhat misleading. A m ore accurate title would be "The Resurrection o f M2 as a M one tary Indicator/' or "M2 Lives; Long Live M2.” The paper is really about how w ell the M2 aggregate functions as a m onetary indicator to guide m onetary policy. Because the paper provides support fo r M2 as a m onetary indicator, Jerry concludes that the central bank can achieve price stability. I agree w ith Jerry that the cen tral bank can achieve price stability, but this is not really the focu s o f the paper because, as I discuss later, the success o f M2 as a monetary indicator is not required fo r a central bank to achieve the price stability objective. THE BASIC IDEA The basic idea behind the paper is a simple one: A m onetary indicator may by itself convey little information about future inflation or unemployment—even though it is actually an excellent indicator fo r these variables—if the dynamic interaction between inflation and un employment is ignored. This point may be a simple one, but it is important nonetheless be cause it has often been overlooked in the recent debates about w hether the Federal Reserve should use M2 as a m onetary indicator or tar get. What the model in Stein’s paper shows is that dynamic interactions betw een inflation and unemployment im ply that the effect o f M2 FEDERAL RESERVE BANK OF ST. LOUIS grow th on the econom y depends very much on the current state o f the economy. If the econo my is slack w ith high unemployment and M2 grow th rises, inflation is likely to fall at first, but then w ill rise in the long run. Similarly, if the economy is boom ing w ith unemployment low, a decline in M2 grow th may be follow ed by rising inflation at first rather than falling inflation. The relationship betw een M2 grow th and infla tion may thus not be very apparent, even though there is a close relationship betw een M2 g ro w th and inflation in the long ru n as the stan dard quantity theory o f money predicts. Recent research which finds that M2 is a poor m onetary indicator has looked solely at the direct relationship betw een M2 grow th and a particular economic variable such as inflation or real output. Stein's analysis indicates that this approach may be highly misleading and that, w hen the dynamic interactions betw een infla tion and unemployment are taken into account, M2 comes out very w ell as an appropriate indi cator fo r the monetary authorities. Stein’s resur rection o f M2 has advantages over other recent attempts to resurrect M2 as in Feldstein and Stock (1993). They find that M2 helps predict nominal output growth, but is not a good fo re casting variable fo r either inflation or real out put growth. It is not clear that the FeldsteinStock finding is all that com forting to M2 advo cates since w e do not directly care about nomi- 205 nal output growth, but are more interested in its components—inflation and real output growth. Indeed, Stein’s paper may help explain w hy Feldstein and Stock find that M2 grow th works w ell in forecasting nominal output growth, but does poorly in forecasting its components. W hen M2 grow th rises, Stein's dynamic system indi cates that real output rises at first but this rise does not continue, w hile inflation rises later. In both the short and the long run, nominal out put growth is higher when M2 grow th rises, and this may explain the link between M2 growth and nominal output grow th that Feld stein and Stock find. On the other hand, the different response o f real output grow th and in flation in the short and long runs make it hard er to find a link between M2 grow th and real output growth and inflation. SOME CRITICISMS Stein’s paper provides a useful perspective on how to interpret the evidence on m onetary indi cators and provides new evidence that M2 might have a useful role as a monetary indicator. This paper suggests that the abandonment o f M2 by the Fed outlined in Alan Greenspan's recent tes timony in Congress may be premature. Despite finding value in this paper, like any good discus sant I have to poke some holes in its arguments and raise some criticisms. One serious problem w ith the evidence in the paper is that the favorable findings fo r M2 growth as a m onetary indicator only appear with annual data. Jerry deserves to be com mended fo r being very forthright in indicating that M2 and his model do not fare w ell with the quarterly data in Appendix 2. Jerry attributes the problem with the quarterly data to the nois iness o f this data. I continue to be quite dis turbed, however, that the results w ith quarterly data are so poor. A key point o f his analysis is that it focuses on dynamic interactions. W hen w e are interested in dynamic interactions, w e are particularly interested in looking at data ob served at short intervals such as a quarter be cause data averaged over longer intervals such as a year may not reveal much about the dy namics. The disappointing results w ith quarterly data are thus very troubling, because this is the data that would seem to be m ore suited to tests o f his model. Another issue about the robustness o f M2 as a monetary indicator arises w hen the paper uses the M2 divisia index instead o f M2 in the esti mated equations. In K. Alec Chrystal's paper in this volume, divisia M2 tends to outperform simple-sum M2 in the forecasting equations, and yet Stein’s results w ith divisia M2 do not satisfy the theoretical restrictions o f his model. Since there are some theoretical restrictions argu ments fo r divisia indices over simple-sum ag gregates, the lack o f robustness o f the results using divisia M2 is somewhat disturbing, par ticularly because other researchers such as Chrystal find that divisia M2 does pretty well. I also have some problems with the paper’s evidence on the poor forecasting perform ance o f real interest rates in the dynamic model. The way the effect o f real interest rates is tested is to add one lag o f the nominal interest rate as an explanatory variable in the regressions, which also include one lag o f the inflation rate. A rise in the lagged nominal interest rate is thus equiva lent to a rise in the lagged ex-post real interest rate in these regressions. Although the co effi cient on the lagged nominal interest rate there fore reflects the effect o f the lagged ex-post real interest rate, it is the effect o f the ex-ante real interest rate, a forward-looking variable, that is m ore relevant to the debate on w hether real in terest rates should be used as a m onetary indi cator. One variable that researchers have looked at that is meant to represent the effect o f real interest rates is the spread between short- and long-term interest rates. The idea is that the long rate reflects expected inflation and so the short-long spread tells us something about the real short-term interest rate. The short-long spread does pretty w ell in forecasts o f real eco nomic activity [for example, see Hardouvelis (1991), Bernanke and Blinder (1992) and Bernanke and Mishkin (1992b)], and it might be w orthwhile to look at how w ell it does in Stein’s fram ework. I also have some questions about the paper’s evidence on the controllability o f M2. The paper provides evidence that the coefficient on adjust ed reserves in an M2 regression is not signifi cant after 1975, thus casting doubt on the controllability o f M2 in recent years. Although the conclusion that M2 is uncontrollable might be correct, I think the jury is still out on this one. Despite Stein’s evidence, M2 might be more controllable than his evidence suggests because there are a lot o f other factors that affect the relationship betw een M2 and adjusted reserves that are left out o f his regression. I f these fac tors are predictable by the monetary authorities, MARCH/APRIL 1994 206 then the monetary authorities might be able to offset them and exercise far tighter control o f M2 than Stein’s regression equation suggests. POLICY IMPLICATIONS The paper indicates that since M2 growth works w ell in the dynamic model, it is a good long-run indicator fo r inflation. In addition, Jerry comes to the conclusion that since his inflation equation has m ore stability than the unemployment equation, the Federal Reserve should focus on price stability rather than un employment as the goal o f m onetary policy. I strongly agree that the prim ary focus o f central banks should be price stability rather than the business cycle. Th e uncertain effects o f m one tary policy on real output is one reason, as Jerry points out. Another important reason, however, relates to the expectations created by a particular strategy fo r monetary policy. Stein’s paper does not emphasize this second reason fo r focusing on the price stability objec tive because it does not make use o f rational ex pectations. W hether you buy into it completely or not, the rational expectations revolution has taught us important lessons about the problems that face central banks w ho attempt to manipu late real output or unemployment. I f a central bank tries to reduce business cycle fluctuations, models such as Barro and Gordon (1983) indi cate that this strategy w ill lead to high inflation w ithout necessarily achieving any reduction in the degree o f business cycle fluctuations. The problem is that attempts to reduce business cycle fluctuations destroy the credibility o f the central bank and so create expectations that high inflation w ill be accommodated, which results in a self-fulfilling prophecy. This lesson from rational expectations models has had an important impact on the economics profession. Most macroeconomists take the issue o f credibility very seriously w hen discussing monetary policy and, as a result, tend to sup port the view that m onetary policy should focus almost exclusively on price stability. Thus, w hether macroeconomists are monetarist or not, or w hether they accept the evidence in Stein’s paper resurrecting M2 as a m onetary indicator, they tend to agree with Stein’s view that price stability should be the prim ary goal o f a central bank. The importance o f credibility and expectations about m onetary policy suggests an important FEDERAL RESERVE BANK OF ST. LOUIS reason w hy m onetary targeting might be useful fo r monetary policymakers. As Bernanke and Mishkin (1992a) point out, targets fo r growth rates o f monetary aggregates might help signal the public about the long-run intentions o f a central bank regarding inflation. Adherence to a m onetary target may low er the public’s inflation expectations, which helps keep inflation from getting out o f hand. Stein’s paper lends some support to the use o f M2 targeting in the United States because it suggests that M2 grow th is a good indicator fo r inflation in the long-run and, thus, can provide an appropriate signal to the public. To finish my comments, I want to return to the issue o f w hy I think the title o f Stein’s paper is misleading. Jerry’s paper is not really about w hether the central bank can achieve price sta bility. It is true that having M2 be an accurate m onetary indicator makes it easier fo r a central bank to achieve price stability both because it provides a m ore accurate guide to m onetary policy and because it enables the central bank to signal the public about its anti-inflationary stance. However, even if M2 or any other m one tary aggregate is a poor m onetary indicator, cen tral banks can achieve price stability. Indeed, this is exactly w hat w e have seen over the last 10 years in the United States. The way I would characterize the Federal Reserve’s strategy fo r the conduct o f m onetary policy in recent years is that it has not made much use o f any specific monetary indicator. Instead, it has operated in the follow ing man ner: W henever the economy has been getting close to full employment o r inflation has risen, the Fed has stood ready to slam on the brakes by restricting reserves grow th and raising in terest rates until inflationary pressures subside. This strategy is not too different from nominal GDP targeting, although the weights on real out put grow th and inflation may not be equal as in nominal GDP targeting. This strategy seems to w ork pretty w ell in the United States and in other countries as long as the central bank pursues the follow ing rule-like behavior: It creates expectations that when infla tionary pressures increase, it w ill pursue tighter m onetary policy and then lives up to these ex pectations by actually carrying out this policy. The outcome o f this policy in the United States has been a low inflation rate with very little variability. Since the success o f this policy has not been based on the use o f any m onetary in 207 dicator, it should be clear that price stability can be achieved without it. '[’bus, even if w e are unable to find a satisfactory monetary indicator, there is still a strong case fo r rule-like behavior on the part o f the central bank to control inflation. _______ , and Frederic S. Mishkin. “ Central Bank Behavior and the Strategy of Monetary Policy: Observations from Six Industrialized Countries,” NBER Macroeconomics Annual (1992a), pp. 183-228. _______ , a n d ________ “ The Predictive Power of Interest Rate Spreads: Evidence from Six Industrialized Countries,” mimeo, Princeton University, December 1992b. REFERENCES Chrystal, K. Alec., and Ronald MacDonald. “ Empirical Evi dence on the Recent Behavior and Usefulness of SimpleSum and Weighted Measures of the Money Stock,” this Review (March/April 1994). Barro, Robert, and David B. Gordon. “ A Positive Theory of Monetary Policy in a Natural Rate Model,” Journal of Politi cal Economy (August 1983), pp. 589-610. Estrella, Arturo, and Gikas A. Hardouvelis. “ The Term Struc ture as a Predictor of Real Economic Activity,” Journal of Finance (June 1991), pp. 555-76. Bernanke, Ben, and Alan Blinder. “ The Federal Funds Rate and the Channels of Monetary Transmission,” The Ameri can Economic Review (September 1992), pp. 901-21. Feldstein, Martin, and James H. Stock. “ The Use of Mone tary Aggregates to Target Nominal GDP,” NBER Working Paper no. 4304 (March 1993). MARCH/APRIL 1994 209 A Conference Panel Discussion Michael J. Boskin Michael J. Boskin is Tully M. Friedman professor of economics and senior fellow, Hoover Institution, Stanford University. He is also an adjunct scholar with the American Enterprise Institute. The Role o f Rules in Monetary Policy J . W IL L TRY TO ASSUME m y comparative ad vantage on this panel and put a broader-brush perspective on m onetary aggregates, interm edi ate targets, rules versus discretion, and the re cent history o f m onetary policy-making. Many o f my positions have been stated in various parts o f several o f the recent Econom ic Reports o f the President. A bove all, m onetary policy ought to be forwardlooking. It should be rule-like, or rules-based, but not necessarily mechanical as in a Friedman or Shaw fixed m oney grow th rule. Let me state a fe w propositions that support m y position and which a fair reading o f history w ould conclude are sensible even though there are persons at this conference w h o have argued contrary propositions over time. Th e first is that high inflation, indeed even high and stable inflation, can carry substantial cost to the economy. It was not uncommon in the late 1970s and early 1980s fo r people to ar gue that if w e could m ore or less stabilize infla tion so that the variance was much smaller than it had been, a high mean o f 10 or 12 percent might be far preferable to bearing the potential cost o f disinflation. The cost o f disinflation was view ed as inordinately high, and indeed w e did have a high cost, as Rick Mishkin stated, in the recessions o f 1980 and 1981-82. But that cost, according to any serious analysis, was far less than the simple models that many economists w ere using predicted, especially in terms o f lost output. Th e costs w e re perhaps a third, and certainly less than half, in terms o f lost real out put than w hat had been predicted fo r the amount o f disinflation engendered. The cost o f inflation stems from a variety of things but one o f the most important is that the fiscal rules that determine our tax system are not invariant to the rate o f inflation. W hile w e eventually in the early '80s indexed tax brackets fo r inflation, w e did not index the definition o f income. W e still have historic cost depreciation. W e still have nominal capital gains tax, tax nomi nal interest, and allow deductions fo r nominal interest. It is complex, but w hen you are look ing at investment decisions, those are important. This is part o f the reason w h y monetary policy in the late 1970s, likely the w orst episode in the post-World W ar II history o f m onetary policy, was so bad. Attention was being paid to nomi nal interest rates rather than, as difficult as they are to measure, expected long-run real nefof-tax interest rates. The second point I w ould make, and w ill come back to, is that those w ho argue that in deed m oney does m atter initially—and not just fo r prices but fo r real output—seem to have been correct. A tighter m onetary policy than the Fed envisioned in the early ’80s led to that costly (but not as costly as predicted) disinfla tion. I think that the simple monetarist proposi tions available at that time broke dow n w ith the collapse o f M l velocity in the early 1980s (and MARCH/APRIL 1994 210 again w ith the collapse o f M2 velocity in the early 1990s). The simplistic notions o f monetarists took a beating, even if the fundamental tenets w ere, and I think continue to be, m ore or less correct. Keynesian and neo-Keynesian arguments took a beating as w ell since w riters in the econom etric Keynesian tradition greatly overstated what the cost o f the disinflation w ould be in terms o f lost output. Those w ho focused on expectations and on credibility proved to be—and let me make sure I am careful about this—partially correct, in m y opinion. I think they w ere no m ore fully correct than the monetarists w ere or than the simple Keynesians and neo-Keynesians w ere. All o f these schools o f thought contained ele ments o f truth, but none was a sufficient descriptor o f the econom y or prescriber o f eco nomic policy. W e have learned through the w ork o f some people at this conference and others that some households and businesses in deed are liquidity-constrained and do respond to short-run cash flows. Hence, there is some scope fo r affecting the shorter term course o f the economy, if and w hen that proves to be desirable, w ith discretionary policy. Expectations certainly have been shown to matter. A large part o f the reason that the last decade has been substantially better than the previous decade, in terms o f macroeconomic perform ance and in a manner I w ill describe in a moment, stems from the fact the Fed has gradually built considerable credibility on reduc ing inflation and keeping inflation low and sta ble. Th e inflation expectations premium has been gradually abating. The next point I w ould make is that the eco nomics profession ought to be quite humble about both our ability to go from changes in m onetary policy to short-run changes in nomi nal GDP, and from the change in nominal GDP to the changes in inflation and real output. Humility is called fo r in far greater magnitude than has been evidenced by most economists; that w ill lead me back in a moment to the proposition that I w ill make about nominal GDP rules. The w eight o f the evidence accumulated dur ing the recent relatively successful disin fla tion first in the early '80s and later in the last few years, from double digits dow n to the 4 to 5 percent range and, later, from that range down to around 3 percent—suggests that after adjust FEDERAL http://fraser.stlouisfed.org/RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis ing fo r the state o f the econom y the disinflation was achieved in the context o f much low er un em ployment and much less lost output than had been expected. Some people claim that the 1970s was just as good a decade and that despite the long expansion in the 1980s, the grow th then was no higher. But the 1980s w ere a period when lots o f inflation was taken out o f the sys tem and the previous decade was a situation in which lots o f inflation was added to the system. Indeed, if you step back (and I know it is hard when you are doing technical research on a specific subject) and look at post-W orld W ar II history, w e w ere in this horrible situation w here at corresponding stages o f each cycle— the midpoint, trough or peak—inflation at that point was getting higher and higher. And perhaps the most remarkable thing is that not only was inflation stabilized but that relation ship was broken, hopefully fo r a considerable length o f time, fo r the foreseeable future. There w ere many people who, circa 1980, thought w e would have, as I mentioned, not only something close to a depression to get inflation dow n to low levels, but that inflation w ould then start to accelerate substantially once w e got w ell into the next expansion. Can w e do better? M y answer is yes. And I w ill get to that in a second. As I said earlier, the w orst episode was the late 1970s and I b e lieve that there w ere several fundamental mis takes. One was accommodation and, without getting into personalities, I’ll just say that it seemed to me w e had a Fed in the late 1970s that was really not responsible. W hatever modest impetus and modest cost-push supplyshock w e had, w hatever oil prices did, was a tiny fraction o f the total impact on the accelera tion o f inflation. Some people attribute up to 3 percentage points in the 13 percent rate to the oil shock. But the inflation was basically a m onetary phenomenon. The Volcker disinflation o f the late 1970s and early 1980s, if I can revert to a professor giving grades, gets a B + or A - . It was achieved at much less cost than anticipated despite the severe recession, but also I think Rick Mishkin is right that the Fed really wasn’t looking just at m oney as velocity was collapsing. I do believe that m onetary policy, ex post, proved to be much tighter than the Fed had imagined and they did want a m ore gradual disinflation (that is one reason they don't get an A). W hether a m ore gradual disinflation could have been 211 achieved at a low er cost is something w e w ill never know. I give the Fed an A - fo r its policy in the late 1980s to try and proactively head o ff an incipient, building inflation. And this gets back to a point several people have made that m onetary policy has to be forward-looking. The Fed rarely gets credit w hen it prevents the inflation rate from going from , say, 4.5 per cent to 6.5 percent, because people never see it get up to 6.5 percent and then go back down again. And so I think an A - because they prob ably w ent a little too far. W hile they couldn’t have foreseen the oil shock or anticipate the size o f the defense drawdowns and other things going on in the economy, they probably should have done better at understanding that the regulatory system o f financial institutions was going to take some steam out o f the economy. W heth er that was desirable is another story, but I think that you can't understand m onetary policy without also looking at the regulatory structure o f the financial system. I would give the Fed low er marks fo r easing too slowly and too timidly but, to be intellectually honest, had they eased as I thought desirable—a bit m ore rapidly and a bit m ore aggressively —how much o f that w ould have shown up in output and how much o f that in slower reduc tion in inflation is certainly an open question. I certainly give them much higher marks than most o f the academic economics profession— Samuelson, Tobin, Solow, Feldstein, Friedman, McCracken and others. Yet, by the end o f ’91 or early '92, they got to about w h ere they should be, and I think the Fed is pretty close to w h ere it ought to be, although it probably w ill need to m ove to a less accomodative policy as 1994 progresses. W hat have they been doing? At various times, the Fed has announced or listed in prime direc tives that they have been looking at interest rates, reserves, M l and M2, comm odity prices, exchange rates, and so on. I think it is very clear that on the Federal Open Market Commit tee (FOMC) people are looking at different things but that, in general, the prim ary concern is and has been reducing inflation. They have been somewhat opportunistic about doing that. Th ey get concerned w hen it appears that infla tion looks like it may accelerate or over bad news in contemporaneous data about inflation. It is an interesting issue how much information that it is conveying and its potential as a leading indicator o f future inflation. Th ey have tended to take advantage o f opportunities to try to take another round out o f inflation w hen that seems desirable. W hen the econom y happens to be slack, they tend to try to help the economy somewhat in the short run. W hile there was not a lot o f discussion in the last year or tw o about price stability, there was a lot o f discus sion o f that as the prim ary goal a fe w years ago—they vie w their job as to try to keep infla tion low and steady and try to avoid doing any thing that leads to an unnecessarily large swing in output. I echo the lender-of-last-resort, avoida-financial-panic issue. They have operated under some big structural changes in the economy, in cluding the declining fraction o f credit extended by the banking system, the fact that far less of broad m onetary aggregates is reserved against any more, changes in the international arena w hich leads to far m ore m obility o f capital, and so on. W hat I infer from all o f this is that the Fed has to be a compass, not a w eather vane, laying out a basic path that they are trying to achieve fo r their policy. I think they have done that, although at times less than clearly. In general, they have laid out a course o f what they are trying to achieve that has generally been fairly reasonable, with a couple o f exceptions in the last decade or so. It is a rules-based policy, not one that is a fixed rule, but one that basically lays out a policy path that is deviated from only rarely and tem porarily, fo r contingencies that are generally well-understood by the public to be rare events. The basic rules-based fram e w ork is the proper one fo r m onetary policy, and I think it is probably the w ay to under stand what the Greenspan Fed has been trying to do, and perhaps the Volcker Fed up to a point as well. A far m ore difficult question is what do you do about specific indicators. I personally do not believe that M2 is a sufficient intermediate indi cator. I don’t believe nominal GDP is either, since w e still have the problem o f separating out real grow th and inflation. I believe the list o f indicators must include m ore than one sim ple measure such as M2, or adjusted reserves, or M l. That is not necessarily a disingenuous intellectual exercise to th row the Congress o ff their backs, although that may be a valuable purpose. I think that there is information con tained in a variety o f indicators and the Fed is going to have to look at all o f them. MARCH/APRIL 1994 212 Secondly, I believe that it is desirable fo r the Fed to lay out parameters, broadly speaking, despite Rick Mishkin’s argument that the Bun desbank and the Swiss have often been w ay o ff in m oney grow th targets. The Fed w ill continu ally face episodes such as w e had in the early '80s and the early '90s w hen relationships be tw een reserves and rates, betw een one or another m onetary aggregate and nominal GDP, and among nominal GDP, real GDP and infla tion, w ill be far less stable than they are at other times. Nevertheless, I do believe it is desirable fo r the Fed, in the context o f the http://fraser.stlouisfed.org/RESERVE BANK OF ST. LOUIS FEDERAL Federal Reserve Bank of St. Louis rules-based policy, to lay out w hat it is trying to achieve and how it is trying to achieve it in a w orld o f incomplete information, rapid structur al change and inaccurate data. That is not a simple task, but one the Fed has perform ed, by any fair evaluation, quite w ell fo r the past decade-and-a-half. REFERENCES Council of Economic Advisors. Economic Report of the Presi dent. U.S. Government Printing Office. 213 Philip H. D ybvig Philip H. Dybvig is Boatmen’s Bancshares professor of banking and finance, Oiin School of Business, Washington University, St. Louis. I am grateful for helpful comments from Kerry Back, Jim Bullard, Ning Gong, Hyeng Keun Koo, Mahesh Maheswa ran, and participants in the Conference on Monetary Ag gregates at the Federal Reserve Bank of St. Louis. All comments are the author’s and may not represent the position of the Federal Reserve Bank. What Is the Fed’s Decision Problem ? W ■ 'H A T IS THE BEST m odel o f a piece o f iron? I f it is to be thrown, the best m odel might be a uniform mass o f fixed density and shape. I f it is to conduct electricity, thinking o f the piece o f iron as a hollow tube like a pipe that carries water, is illuminating. For purposes o f studying its magnetic properties, it may be best to consider the piece o f iron as a collection o f rigidly located magnetic dipoles that can be aligned or not. In general, the best model de pends on the use to which the model is put. In an economic setting, the best economic model is one that helps us understand the choices made by econom ic agents. Unfortunately, the specific nature o f the Fed's decision problem remains obscure in most discussions o f Federal Reserve policy. In these remarks, I look at the Federal Reserve through the lens o f decision theory. W hile I'm not necessarily suggesting that the Fed must or should specify an explicit objective function, I do think that decision theory is nonetheless a very useful fram ew ork fo r thinking about the economy, m onetary ag gregates and the Fed’s policy role. This should be a com fortable notion fo r economists, virtually all o f w hose models are based on decision theory. The O b je c tiv e Function Many purely political attacks on the Fed are posed in terms o f the objective function. None theless, its specification is a substantive issue. Focusing on the Fed’s role in m onetary policy, there seems to be some consensus within the Fed that there is a lexicographic preference to keep inflation down, and given low inflation, to stimulate economic growth. Separate criteria are applied to crisis management such as the injec tion o f cash to help illiquid specialists during a crash. None o f this is entirely satisfactory: Lexicographic preference fo r reducing inflation is certainly not the ultimate objective o f the Fed, which might ultimately seek a good out come fo r the econom y given the complex inter action betw een the Fed, the Congress, the rest o f governm ent, and the rest o f the economy. In order to achieve a good outcome, part o f the Fed’s objective should be political survival w ith pow ers (including independence) intact. It seems that the lexicographic objective to keep inflation dow n is intended to do some good in the econom y subject to political survival and given inherent limitations on what the govern ment can do to help the economy. This narrow view o f the Fed does not seem ideal, but is sur ely better than what w ould com e under the po litical control that w ould result from any loss o f the Fed’s independence. C on trol Variables Although control variables include such things as reserve requirements and discount w indow policy, the most comm only used control variable is the open market operation. I continue to be puzzled as to w h y the Fed confines its open m arket operations to trading only once each day in a ve ry limited set o f securities, most MARCH/APRIL 1994 214 often repurchase agreements in short-term Treasuries. At the same time, the Fed seems to be very interested in the behavior o f long rates, apparently believing that movements in long rates signal changes in expectations o f inflation. In other words, the Fed is trading short-term in struments w hile judging the success or failure o f its actions, relative to maximizing its objective function, by watching long-term rates. This choice o f control variable, given the objective function, seems puzzling. Since the Fed is not the only econom ic actor in the econom y that looks to long rates to think about inflation, perhaps a better w ay to influence expectations o f inflation is by trading long-term bonds them selves. W h y doesn’t the Fed trade long-term bonds? One reason often cited is, in truth, ir relevant: Operation Tw ist in the ’60s was a bad idea imposed on the Fed from outside and it didn’t work. A m ore serious suggestion is that the Fed may not be big enough to affect long rates or, in other words, that long-term bonds may not in fact be a feasible control variable. The reasons w hy it may be infeasible fo r the Fed to trade enough to m ove long rates, h ow ever, aren't self-evident and usually are left unstated. In addition to long-term bonds, there are numerous other financial instruments such as futures and options on Treasuries that might be used as control variables. One reason fo r considering these instruments as control varia bles arises from the recent finance literature on h ow price pressure—the amount prices m ove in response to trading volum e—varies across m ar kets. Price pressure is a lot like walking down the demand curve as a monopolist: W hen your early trades have a big effect on price, you get a much less favorable price on subsequent trades. Most agents w ho take a position with respect to market interest rates want to mini m ize price pressure. The Fed actually may p refer the opposite perspective. If the Fed's m o tive fo r trading is an attempt to change expecta tions (say, o f future inflation) without taking on too large a risky position, the Fed may want to maximize (not minimize) price pressure fo r a given level o f exposure. Trading long-term in struments may be a feasible w ay to do so. Constraints W hat are the constraints faced by the Fed in maximizing its objective function? Almost every discussion o f Fed policymaking hinges on some implicit constraint. If the Fed is, in fact, too http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis small to m ove long rates, fo r example, then there must be some limitation to the Fed's ability to short T-bills and go long Treasury bonds or vice versa; otherwise, it seems that they surely could take positions that w ould m ove long-term rates. It should be interesting to specify explicitly such restrictions. Other constraints may arise from the Fed’s charter. Does the Federal Reserve Act constrain the amount o f risk the Fed is permitted to absorb? It might seem not. A fter all, what is interest rate risk to an agent w h o can always print m oney to satisfy a claim? Several central banks have learned the hard w ay the limitations on their ability to influence foreign exchange markets. The possibility o f large losses (or even profits) seems less likely in domes tic markets, given the printing o f m oney and possible deferral o f paper losses. Nonetheless, given the 1993 magnitude o f $16 billion returned to the Treasury by the Fed, it seems that trading gains or losses o f $5 billion could cause severe political damage. If the Fed misjudges its capacity to bear risk, it can cause significant damage by being either too bold or too timid. The In fo rm a tio n Set W e have discussed the objective function, the controls, and the constraints. W e cannot under stand a decision problem w ithout knowing the decision maker's inform ation set. In finance, w e routinely gather a great deal o f inform ation by m onitoring m ore or less continuously the m ar ket prices o f securities. Macroeconomists simi larly often m onitor high-frequency data such as market interest rates as indicators o f expected inflation and the level o f the stock market fo r expectations o f economic activity. Th ere are, how ever, many other variables that should be considered. Option prices, such as Standard & Poor's 100 index options and T-bond futures op tions, may be used to infer the types and amount o f risk people perceive in the market. These data permit us to separate the degree o f investors' uncertainty about the level o f future inflation from investors’ expectations o f the level o f inflation. This is important because it is the degree o f uncertainty about inflation, not the level itself, that makes planning difficult fo r businesses using nominal contracts. Similarly, the stock index options measure investors' uncer tainty about the overall level o f future economic activity. Other data, such as information on the money stock or unemployment, are available at an 215 intermediate frequency. These intermediate frequency data provide some independent in for mation beyond what is available in security prices, although how much is really an empirical question. And then there are the low-frequency time series, which are very important, such as inflation or industrial productivity. This plethora o f variables raises the difficult question: W hen w e can't look at 16 things at once, how do w e summarize the information in a w ay that is useful fo r policymaking? This is the type o f question that is implicit in the choice o f a m onetary aggregate or any other policy indicator or target. In principle, w e should not throw anything away. How ever, if w e put too many variables in our statistical analysis, the loss o f p ow er w ill reduce the quali ty o f fit, especially w hen some ultimate objec tives such as production and inflation are available only at low frequencies. Although it seems sensible to focus on a subset o f the avail able data, it is unclear what should be the crite rion fo r combining data or fo r deciding which data to th row away and which data to retain. This b rie f look at the Fed’s decision problem suggests several interesting avenues fo r research. It w ould be useful to have a careful and apoliti cal analysis o f the Fed's objectives. W e should quantify the Fed's constraints on trading, base m oney creation, and risk-bearing. Empirically, w e should have m ore w ork w ith high-frequency data (daily and intra-day) and m ore examination o f the Fed’s actual controls (trades) and their direct impact on markets. It w ould be interest ing to understand better how to aggregate lowand high-frequency data. Keeping the Fed’s deci sion problem in mind w ill help to guide our re search tow ard the most important policy issues. MARCH/APRIL 1994 216 Bennett T. McCallum Bennett T McCallum is H.J. Heinz professor of economics at . the Graduate School of Industrial Administration, Carnegie Mellon University. He is also a research associate, National Bureau of Economic Research. Monetary Policy Without Monetary Aggregates J L HE PAPERS PRESENTED at the conference represent a useful step in the ongoing search fo r im proved ways o f measuring m onetary ag gregates. Th eir basic idea, o f w eighting com po nents o f the aggregates by a measure o f the extent to which they serve as media o f ex change, should be rather appealing to anyone w ho view s the medium-of-exchange function as the defining characteristic o f money. And I don’t know o f any other potential defining characteristic (for example, the store-of-value function) that makes any sense. So, to repeat, I find quite promising the idea that some indices and w eighted sums might do a better job than the simple-sum aggregates in measuring the quality o f money. But w hile this type o f study seems potentially useful fo r the purpose o f studying m oney de mand behavior, building econom etric models and judging the historical record, I am not en thusiastic about the developm ent from the per spective o f monetary targeting. The reason—as some o f you w ill have heard me argue b e fo re— is that I believe that there is a good w ay o f con ducting m onetary policy that does not rely on any targeted m onetary aggregate. Instead, it uses as its target variable nominal GDP, or GNP, or domestic demand, or some such measure of aggregate nominal spending. Th ere are several ways o f arguing that nomi nal GDP (or whatever) is a m ore appropriate target variable than any m onetary aggregate. The simplest and most blatant is to just assert that it is obvious that a central bank's main job FEDERAL RESERVE BANK OF ST. LOUIS is to keep total nominal spending grow ing smoothly at a noninflationary rate. But one can proceed m ore circumspectly by arguing instead that from the perspective o f hitting price level or inflation targets, on average over the next decade or so, w e know w ith much greater ac curacy what grow th rate o f nominal GDP w ill do the job than w e do fo r M l or M2. And even if the task o f developing an im proved index o f m oney is successful, it w ill still be true that w e w ill know w ith m ore accuracy what rate o f grow th is needed (to deliver a chosen inflation rate) fo r nominal GDP. T o the foregoing one might naturally respond, w h y not make inflation the target directly rather than indirectly? But to this there are tw o answers. One is that, because the price level usually responds m ore slowly to policy actions than does nominal GDP, a policy feedback rule is m ore likely to generate dynamic in stabilityso-called instrument instability—if it responds to target misses fo r the price level rather than nominal GDP. And the second argument is that generating a smoothed path fo r nominal GDP is likely to result in smaller fluctuations o f real GDP— that is, reduced cyclical variability. (I am, o f course, aware that w e cannot be certain about the latter, given current knowledge, and also that it is not desirable to smooth out responses to all types o f shocks. But I w ill stand by the statement nevertheless.) T o return to the issue concerning m onetary aggregates, the only advantage that I can see 217 fo r them (as targets), relative to nominal GDP, is that observations are available m ore often and m ore promptly. But w e could certainly devise other measures o f nominal aggregate spending that w ould be available m ore frequently and promptly. Furthermore, it is not clear that hav ing measurements m ore frequently is terribly important. O ver the last year or so, w e have ex perienced quarter after quarter o f rapid M l and base grow th at the same time as very slow M2 growth. These aggregates w ere suggesting either excessively loose or excessively tight monetary policy, depending on which one you utilized. But nominal GDP grow th chugged along reasonably close to 5 percent (per annum) in almost every quarter, which is just about enough fo r 3 percent real grow th and 2 percent inflation. So if 2 percent is the Fed's concept o f "ze ro inflation,” which seems defensible, then policy behavior has been just about right from a medium-term perspective. And the point, rela tive to the issue regarding the frequency and promptness o f measurements, is that these vari ous grow th rates have differed in the manner described above fo r many months in succession. One objection that is sometimes raised against nominal GDP targets is that they might make it appear to the public that the Fed is controlling real GDP—that it is attempting a role that is greater than is actually feasible. But I w ould not presume that these targets w ould be publicly announced. The role fo r targets that I have in mind is as significant inputs that the FOMC w ould use in making its decisions, as proposed by Taylor (1993). Announcements are much less important, I believe, than behavior. Having appropriate targets is, o f course, not the w hole story; to conduct m onetary policy successfully it is also necessary to have a policy feedback process—among friends I w ould call it a "ru le”—that specifies instrument settings, that is, settings o f a variable that the central bank can control directly or w ith great accuracy. In m y ow n studies,1 which have been designed to see if a simple rule w ould succeed in hitting nominal GNP targets w ith reasonable accuracy in a variety o f (small) econom etric models, I have usually used the St. Louis adjusted m one tary base as the instrument variable. The rea son fo r that choice is that the base’s grow th rate provides a nice measure o f the pace at which open market purchases (or sales) are b e ing conducted, and if the adjusted base is used the measure takes account o f changes in reserve requirements as well. So it seems to be the most natural aggregate among those that are highly controllable —which the base is since it appears on the Fed's ow n balance sheet and so could be m onitored daily (and thereby kept close to the specified values). The other main contender fo r the role o f the instrument variable is, o f course, the federal funds rate (or some other short-term interest rate). But interest rates seem quite unattractive because a high interest rate suggests tight money from a short-term perspective but easy m oney from a long-term perspective. Or, as I say to m y students, if a central bank wants in terest rates to be low er, then it needs to raise interest rates. That strikes me as an extrem ely undesirable feature fo r an instrument variable. In addition, I have tried in m y simulation w ork to design interest rate rules and have found that they perform much m ore poorly than ones w ith the base instrument.2 These results, at the quarterly frequency, are not definitive but they are supportive o f the belief that the base is the better instrument from a m acroeconom ic per spective. Most actual central banks are, o f course, ex trem ely resistant to proposals fo r accurate base control, on a short-term basis, and have accord ingly been rather unreceptive to such policy rule suggestions. One important reason fo r that resistance, I believe, is the b elief that exerting short-term base control w ould generate more financial market instability and w ould also re quire the central bank to give up its role as the lender o f last resort. But I w ould like now to ar gue against that belief. There is a fairly w ell-known paper b y Goodfriend and King (1988) that emphasizes that functioning as the lender o f last resort does not necessarily require the provision o f discount w indow loans; what is necessary is that the cen tral bank makes available additional base money at times o f financial crisis. And they argue that this response w ould come about automatically if interest rate smoothing w e re being practiced. Some critics have described the Goodfriend-King scheme as calling fo r a constant rate o f base money grow th during times o f financial crisis, 'These include McCallum (1988, 1990, 1993a). 2See McCallum (1990, pp. 61-6; and 1993a, Section VII). MARCH/APRIL 1994 218 but that is an entirely incorrect description o f what their argument or proposal actually is. Consequently, in a paper that I have very re cently w ritten fo r a Bank o f Japan conference (McCallum, 1993b), I have tried to follow up on the Goodfriend-King idea by exploring the possi bility o f using a nominal GNP targeting rule to generate implied quarterly settings o f the m one tary base, and then to combine that w ith a higher-frequency rule that calls fo r w eekly ad justments o f a federal funds rate instrument that are designed to achieve the specified quart erly base values. This w eekly rule can be made to imply a lot o f week-to-week smoothing o f the funds rate and thereby automatically to provide lender-of-last-resort support to the financial sys tem. But can it do that w hile simultaneously hit ting the quarterly base settings w ith reasonable accuracy? That is clearly an empirical question whose answer depends upon the size o f shocks that occur and the strength o f w eekly responses o f the base to funds rate adjustments. But I have begun to study that question in this new paper, and the results obtained are quite encouraging. http://fraser.stlouisfed.org/RESERVE BANK OF ST. LOUIS FEDERAL Federal Reserve Bank of St. Louis I w ould like to conclude by expressing my appreciation to the St. Louis Fed’s Research Department fo r continuing their long-running program o f searching fo r ways to im prove the conduct o f m onetary policy. REFERENCES Goodfriend, Marvin, and Robert G. King. “ Financial Deregu lation, Monetary Policy, and Central Banking,” Federal Reserve Bank of Richmond Economic Review (May/June 1988), pp. 3-22. McCallum, Bennett T. “ Robustness Properties of a Rule for Monetary Policy,” Carnegie Rochester Conference Series on Public Policy (autumn 1988), pp. 173-203. _______ . "Targets, Indicators, and Instruments of Monetary Policy,” in William S. Haraf and Phillip Cagan, eds., Mone tary Policy for a Changing Financial Environment. American Enterprise Institute, 1990, pp. 44-70. _______ . “ Specification and Analysis of a Monetary Policy Rule for Japan,” Bank of Japan Monetary and Economic Studies (December 1993a). ________“ Monetary Policy Rules and Financial Stability,” un published working paper (October 1993b). Taylor, John B. “ Discretion Versus Policy Rules in Practice,” Carnegie-Rochester Conference Series on Public Policy 39 (December 1993), pp. 195-214. F e d e r a l R e s e r v e B a n k o f St. L o u i s Post Office Box 442 St. Louis, Missouri 63166 T h e R e v ie w is p u b lis h e d s ix tim e s p e r y e a r b y th e R e s e a r c h a n d P u b lic In fo r m a tio n D e p a r tm e n t o f th e F e d e r a l R e se rv e B an k o f S t. L o u is . S in g le - c o p y s u b s c r ip tio n s a r e a v a ila b le to th e p u b lic f r e e o f c h a r g e . M a il r e q u e s t s f o r su b sc r ip tio n s, b a c k issu e s, o r a d d r e s s c h a n g e s to : R e se a rc h a n d P u b lic In fo r m a tio n D e p artm e n t, F e d e ra l R e se rv e B an k o f S t . L o u i s , P .O . B o x 4 4 2 , S t. L o u is , M is s o u r i 6 3 1 6 6 . T h e v ie w s e x p r e s s e d a r e th o se o f th e in d iv id u a l a u th o r s a n d d o n o t n e c e s sa r ily r e fle c t o ffic ia l p o s itio n s o f th e F e d e r a l R e se r v e B an k o f S t. L o u is o r th e F e d e r a l R e s e r v e S y s te m . A r tic le s h e r e in m a y b e r e p r in te d p r o v id e d so u rce th e is c r e d ite d . P le a s e p r o v id e t h e B a n k ’s R e s e a r c h In fo r m a tio n a n d P u b lic D e p a r t m e n t w ith a c o p y o f r e p r in te d m a te r ia l.