The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Does the Exchange Hate Regim e Affect the Economy? Central Bank Independence and Economic Perform ance Hypothesis Testing w ith Near-Unit Roots: The Case of Long-Run P u rch asin g-P ow er Parity The Effect of Mortgage Refinancing on Money Demand and the Monetary Aggregates THE FEDERAL A RESERVE lU N k o t ST.I/H IS 1 F e d e r a l R e s e r v e B a n k o f St. L o u is R eview July/August 1993 In This Issue . . . D o e s th e E x c h a n g e R a te R e g im e A ffe ct th e E c o n o m y ? Terence C. Mills and G eoffrey E. W ood Teren ce C. Mills and Geoffrey E. Woods exam ine what makes pegged ex change rates so attractive; and consider empirically the relationship b e tw een the exchange rate regime and a num ber of key m acroeconom ic variables, such as output, prices and interest rates, to see w h eth er any system atic relationship exists betw een the behavior of these variables and the exchange rate regim e. They focus their study on the United Kingdom, because the U.K. experienced a wide variety of exchange rate regim es over the period covered by the data, and because Britain has not en dured hyperinflation or recessions as severe as those in some oth er coun tries. Their findings support the conclusion that the exchange rate regime has not been a source of volatility fo r the m acroeconom ic perform ance of the British economy. 21 C e n tr a l B a n k I n d e p e n d e n c e a n d E c o n o m ic P e r f o r m a n c e Patricia S. Pollard Central bank independence is becom ing popular, as evidenced by the num ber of countries that have recently enacted legislation removing their central banks from governm ent control. Examining the econom ic rationale for this popularity, Patricia S. Pollard employs em pirical studies to reveal that countries with independent central banks tend to experience low in flation with no loss of econom ic grow th. On the other hand, theoretical studies illustrate that an independent bank may increase policy conflicts within a country, resulting in poor econom ic perform ance. W eaknesses in both types of studies, how ever, may limit their ability to prove or disprove the usefulness of central bank independence regarding econom ic p erfor m ance. Pollard concludes that the relationship betw een central bank in dependence and the econom y is not fully understood. 37 H y p o th e s is T e s tin g w ith N ear-U n it R o o ts : T h e C ase o f L o n g -R u n P u rc h a s in g -P o w e r P a rity Michael J. Dueker As a principle, it has long been asserted that the quantity of goods one can buy with a given currency, such as the dollar, should be equal across countries, at least in long-run equilibrium. This condition, known as longrun purchasing pow er parity (PPP), has been subjected to num erous em pir ical tests. One source of disagreem ent in statistical tests of PPP has been the choice of null hypothesis: Tests whose null hypothesis is that PPP holds often fail to reject PPP, while tests whose null hypothesis is that PPP fails JULY/AUGUST 1993 2 often com e to the opposite conclusion. Thus, there is a danger in testing only one null hypothesis fo r a broad set of countries, failing to reject it, and concluding that the evidence is clearly for or against PPP. In this article, Michael J. Dueker tests post-1973 monthly data from major countries using a long-memory model. The advantage of this approach is that one can test both null hypotheses with the model and dem onstrate that it is unclear w h eth er long-run PPP holds, because real exchange rates have near-unit roots, w hich may preclude strong conclusions as to w hether real exchange rates are m ean-reverting. 49 T h e E f f e c t o f M o rtg a g e R e f in a n c in g o n M o n ey D e m a n d a n d th e M o n e ta ry A g g r e g r a te s Richard G. Anderson During the last two years, lower interest rates have stimulated extensive refunding of long-term debt, sharply increasing the relative volume of financial transactions. Mortgage refinancing has been a highly visible part of those transactions. Richard G. Anderson examines the effect of recen t waves of mortgage refinancing on the demand for liquid deposits and grow th of the m onetary aggregates. Mortgage servicers may hold unscheduled principal paym ents received following a refinancing in liquid deposits as long as six weeks prior to rem ittance to the investors who own the underlying m ortgage-backed securities. In addition, the grow th of oth er checkable deposits also ap pears to have been affected by fluctuations in m ortgage refinancing, perhaps because of households converting hom e equity to cash. The persistance of these increased demands for liquid balances illustrates that all transactions are not completed instantaneously, as is implicitly assumed. Anderson finds that the increased m ortgage refinancing accounts for a great deal of the volatility of M l’s grow th during the last two years, although not for its continued strong underlying trend. All non-confidential data and programs for the articles published in Review are now available to our readers. This information can be ob tained from three sources: 1. FRED (F e d e r a l R e s e r v e E c o n o m ic Data), an e le c tr o n ic b u lle tin b o a rd s e rv ic e . You can access FRED by dialing 314-6211824 through your modem-equipped PC. Parameters should be set to: no parity, word length = 8 bits, 1 stop bit and the fastest baud rate your modem supports, up to 14,400 bps. Information will be in directory 11 under file name ST. LOUIS REVIEW DATA. For a free brochure on FRED, please call 314-444-8809. FEDERAL RESERVE BANK OF ST. LOUIS 2 . T h e F e d e ra l R e s e r v e B a n k o f St. L o u is You can request data and programs on either disk or hard copy bv writing to: Research and Public Information Division, Federal Reserve Bank of St. Louis, Post Office Box 442, St. Louis, MO 63166. Please include the author, title, issue date and page numbers with your request. 3 . I n t e r - u n i v e r s i t y C o n s o r tiu m f o r P o l i t i c a l a n d S o c ia l R e s e a r c h (IC P S R ). M em ber institu tions can r e quest these data through the CDNet O rd er facility. N onm em bers should w rite to: ICPSR, In stitu te fo r Social R esearch, P.O. Box 1248, Ann A rbor, MI 4 8 1 0 6 , or call 3 1 3 -763-5010. 3 Terence C. Mills and G eo ffrey E. W ood Terence C. Mills is a professor at the University of Hull. Geoffrey E. Wood, a professor at City University Business School in London, was a visiting scholar at the Federal Reserve Bank of St. Louis. Kevin White provided research assistance. The authors would also like to thank Mervyn King and Paul Mizon for their helpful comments. Does the Exchange Rate Regime Affect the Economy? J . T SEEMS TO BE a general rule that countries wish to peg their exchange rates but sometimes have floating rates thru st upon them. On three occasions during the tw entieth cen tury—the breakup of the international gold standard in the 1930s, the breakup of the Bretton Woods system in the 1970s and m ost recently the exo dus of countries (notably Britain) from the ex change rate m echanism (EBM) of the European Econom ic Community (EEC)—external pressures led to the demise of fixed rate schem es and their replacem ent by some degree of exchange rate flexibility. In each case, the passing of the fixed rate schem e was m ourned and within relatively short periods a new fixed rate plan was advanced to replace its fallen predecessor. In view of these failures, how ever, it is reason able to ask: W hat makes pegged exchange rates so attractive? Recently, in the context of the ERM, tw o argu m ents have been advanced. Exchange rate fixity is, as David Hume described in 1752, a way of importing another country’s m onetary policy.1 In the case of the ERM, the deutsche mark served as the system ’s anchor currency, and 'See Hume (1970). 2We do not consider whether a single currency really is a natural development of fixed exchange rates. Germany’s low inflation rate was supposed to spread throughout the EEC. M oreover, the ERM’s m em ber nations believed that the Bun desbank’s reputation would provide some credi bility to the anti-inflation com m itm ent of other central banks and th erefo re reduce the costs of lowering inflation throughout the EEC. A se cond motive fo r adopting fixed exchange rates has been the claim that they, and ultimately a single currency, are im portant to the EEC’s Sin gle M arket Program m e.2 The logic is that the full benefits that could accrue from the free intra-European movem ent of goods, labor and capital will be realized only with a fixed ex change rate regim e.3 A third argum ent, not em phasized recently but im portant on earlier occasions, is that econom ic perform ance— grow th, inflation or any oth er im portant m easure—is b etter under a fixed exchange rate system .4 This third argum ent differs from the second in that it identifies no specific causal chain from exchange rate regime to economic perform ance. But does the exchange rate regime m atter for econom ic perform ance? That is the question “This was an important motivation for Britain’s return to the gold standard in 1925, for example. 3This argument usually takes as axiomatic that trade crea tion will outweigh trade diversion. JULY/AUGUST 1993 4 addressed in this paper. W e exam ine empirically the relationship betw een the exchange rate re gime and a num ber of key m acroeconom ic vari ables to see w hether any system atic relationship exists betw een the behavior of these variables and the exchange rate regime. W e have chosen to investigate this question for the United King dom because data over long periods are availa ble for the variables we wish to exam ine and because the United Kingdom experienced a wide variety of exchange rate regim es over the period covered by these data. TRADE AND THE EXCHANGE RATE REGIME The claim that exchange rate flexibility ham pers international trade in goods and in capital and thus depresses w elfare and perhaps grow th is based on the existence of uncertainty. It is argued that rem oving the possibility of exchange rate change will rem ove an im portant nontariff b arrier, because the possibility of ex change rate changes will deter some traders and investors altogether, w hereas others will have to pay a substantial cost to fix the domes tic value o f their foreign curren cy receipts. Floating exchange rates, in oth er words, are b e lieved to impose additional volatility, and hence costs, on international m arkets. If this is co r rect, a case fo r pegged exchange rates exists, and the case is particularly strong fo r any group of countries (such as the EEC) that wants to encourage mutual international trade and in vestment. The proposition seem s unexceptional, and for a num ber of years studies supported the propo sition. For example, Cushman (1983) and deGrauwe and deBellefroid (1987), w hich are representative of the early literature, found that floating rates did impede trade. But as time passed, an increasing num ber of studies sup ported it.5 By the early 1990s, not only had evidence shifted to support the notion that floating ex change rates do not impede trade, but Feldstein (1992) even w ent so fa r as to suggest that float 5Examples of these studies are Gotur (1985); the IMF’s (1984) extension of Cushman (1983) to cover the bilateral trade of the seven largest industrial countries; Bryant (1987), Bailey, Tavlas and Ulan (1986 and 1987), Bailey and Tavlas (1988) and Ascheim, Bailey and Tavlas (1987). 6Haberler (1986) suggested the same thing some years earlier. FEDERAL RESERVE BANK OF ST. LOUIS ing rates are m ore favorable to trade than are fixed rates.6 Attention is thus directed to other reasons fo r favoring pegged exchange rates. T h ere are two rath er distinct types of effects of exchange rate fixing. The first arises because if a fixed exchange rate is in place, it is unlikely to stay fixed without policy actions. These can take several form s. Most com mon are foreign exchange intervention and short-term interest rate manipulation. Accurate figures on official intervention or the stock of foreign exchange reserves are not always available. Interest rate figures, how ever, are available, and several authors have found that unpredictable interest rate variability increased after exchange rates are pegged.7 These actions in tu rn m ake money grow th m ore volatile, and this can have im por tant consequences for the economy. It may c re ate additional u ncertainty about the future behavior of the price level and thus about real rates of retu rn, which would affect investment. If future prices w ere uncertain, wage bargain ing would be m ore complex because it would be harder to judge the future purchasing pow er of an agreed money wage. This u ncertainty would also affect nominal variables. Bisk-averse investors would be m ore reluctant to buy governm ent bonds because they would be un certain what the coupons would be w orth and what the capital would be w orth at m aturity. This would raise nominal interest rates, the cost of debt service and thus the taxes necessary to service the debt. All these factors could have an adverse effect on long-term grow th, depressing its trend. In summary, the choice of exchange rate re gime could affect the long-run behavior of the economy, influencing trends or cycles in im por tant m acroeconom ic variables. If the choice of exchange rate regim e does not have these long-run consequences, then in term s of m acroeconom ic effects, all that the choice of exchange rate regime does is shift the distribution of short-run fluctuations from one m arket to another. This is the second type of effect noted above. The question we examine is w h eth er any as7See Batchelor and Wood (1982), Wood (1983) and Belongia (1988). Wood and Belongia’s research was conducted in the context of the ERM. In Wood (1983) there was an ex ception to this—Erie (South Ireland) after it joined the ERM. Unpredictable interest rate variability fell in that country, although it increased in every other ERM member country. 5 sociation exists betw een the exchange rate re gime and the trend or cyclical behavior of some key m acroeconom ic variables—in other words, w hether there is any evidence for the first type of effect. If no such association exists, then the only m acroeconom ic consequence of the choice of exchange rate regim e is the change in the distribution of short-term volatility betw een the foreign exchange m arket and the short-term money m arkets. If, in contrast, such an associa tion exists, then the choice of exchange rate re gime may be a m acroeconom ic policy decision of considerable im portance for national well being.8 It is now appropriate to present the data we use fo r exploring this question. W e then exa mine the properties of those data in light of the preceding discussion. THE STOCHASTIC PROPERTIES OF U.K. MACROECONOMIC SERIES ACROSS EXCHANGE RATE REGIMES In this section we consider the stochastic properties of five m ajor U.K. m acroeconom ic series since the mid-nineteenth century. T he ex change rate regim es since then have encom passed every possible type except the crawling peg. Until 1914, the United Kingdom was on the gold standard. That was suspended (that is, the United Kingdom left the standard but with the declared intention of returning) at the outbreak of W orld W ar I in 1914. A fter the war, the United Kingdom implemented a deliberate, dis cussed and announced policy of a retu rn to the gold standard at the prew ar parity. M onetary policy and foreign exchange intervention w ere used to this end, and the policy succeeded in 1925. The United Kingdom left the gold stan dard in 1931, how ever, and the exchange rate floated with varying degrees of intervention un til the outbreak of W orld W ar II in 1939.9 The rate was then pegged to the U.S. dollar. A fter 8lt is, of course, possible that the exchange rate regime is a product of the behavior of the economy; it need not be an exogenous choice. 9For a review and evaluation of explanations that have been advanced to explain the United Kingdom’s abandonment of the gold standard, see Capie, Mills and Wood (1986a). the war, the United Kingdom joined the Bretton Woods system. Several sterling devaluations oc curred under Bretton Woods, but sterling did not finally float until 1972. Again, there w ere varying degrees of intervention under this re gime of dirty floating, but the United Kingdom did not form ally peg sterling until it joined the ERM in 1990 after shadowing the deutsche m ark in 1988 and 1989. The United Kingdom subsequently left the ERM in 1992 to float once m ore. The series we exam ine across these vari ous regim es are output, prices, money, and short- and long-term interest rates. Our particular interest, and the focus of the em pirical w ork that follows, is the trend and the cycle in output and prices primarily, but also in money and interest rates. W e look to see how these variables have behaved over our close to a century-and-a-quarter of data, seeking changes in trend and changes in cyclical pat tern. W hen these are identified, we examine w hether any of these changes are associated with exchange rate regim e changes and, if so, consider why this might be. Output Annual output in the United Kingdom (meas ured in logarithms) over the period 1 8 5 5 -1 9 9 0 is shown in figure 1. Detailed econom etric ana lyses of this series in Mills (1991) and Mills and Wood (1993) show that it can be represented as the sum of a segmented linear trend, with breaks at 1918 and 1921 and a stationary, au toregressive, cyclical com ponent.10 Thus these results indicate that if output can be decom posed as y t = fut + n(, then the trend function is (1) = o + Bt + X lt + A,D2(, t tD w here Dit = ( t - T ) if / > T and zero otherw ise. The identified breakpoints are at Tt = 64 and T2 = 67, w hich coincide with 1918 and 1921. The cyclical com ponent, n(, on the other hand, is found to be adequately modeled as an AR(2) separate. The cycle comprises fluctuations about a horizon tal average; growth is all in the trend. This separation is consistent with most views of the cycle, but it should be noted that some scholars see the cycle as an integral part of the growth process. For an example, see Schumpeter (1950). 1 0Testing for stationarity has no direct economic significance. Rather, it lets us separate the cycle from the trend. The notion is that the trend and the cycle are economically JULY/AUGUST 1993 6 Figure 1 Annual U.K. Output (1855-1990) Logarithms process, leading to the fitted model (standard erro rs shown in parentheses), (2) Yt = 3.474 + 0.01961 - 0.1137D U + (0.026) (0.0007) (0.0103) 0.1170D 2I + n ( 0 . 01 0 1 ) nt = 1.099nt t - 0.346nl 2 + a, (0.083) (0.083)' on the output series in figure 1, and we thus conclude that, apart from the th ree years im mediately after W orld W ar I, during w hich the series fell dramatically, the stochastic process generating output has rem ained rem arkably sta ble. Output is a trend stationary process, ir respective of the exchange rate regime in force. Prices This model has some simple properties. Trend grow th is 1.96 percent per y ear until 1919 and 2.29 percent per year from 1922 on, with the level of trend output falling 28.3 percent in the intervening three years. The com ponent n, im plies that output exhibits stationary cyclical fluc tuations around the trend grow th path, with cycles averaging 8.1 years. The residual stan dard erro r of the equation is 2.33 percent. Figures 2 and 3 present plots of the (logarith mic) U.K. price level annually from 1870 to 1990 and monthly from Jan uary 1922 to May 1992, respectively, excluding the w ar years from 1940 to 1945. Unit root tests, calculated over various sample periods, provide little or no evidence against the hypothesis that prices are difference stationary, that is, 1(1).11 The post-1973 era may differ and is discussed later. The trend com ponent is shown superimposed Tw o aspects of price behavior are w orth fur- 1 Details of these tests and similar tests for the other series 1 investigated are reported in Mills and Wood (1993). http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis 7 Figure 2 Annual U.K. Price Level (1870-1939) Logarithms Annual U.K. Price Level (1946-1990) Logarithms JULY/AUGUST 1993 8 Figure 3 U.K. Price Level (1922-1939) Logarithms U.K. Price Level (1946-1992) Logarithms FEDERAL RESERVE BANK OF ST. LOUIS 9 th e r investigation. The first is the behavior of the price level before the United Kingdom aban doned the gold standard in 1931. Mills (1990) analyzes the long gold standard period from 1729 to 1931 and obtains an estim ate of the lar gest autoregressive root of 0.93, identical to that obtained for the sh orter sample beginning in 1870. The corresponding unit root test, though, rejects the unit root null hypothesis at the 5 percent significance level, and the process found to generate the price level (an autoregres sion of order two) yields cycles of around SO years, close to the long swings thought to have characterized prices during this period.12 The second aspect concerns the post-1946 b e havior of prices. Figures 2 and 3 show the ser ies to have undergone slope changes around 1973 and 1983; possible explanations fo r these are discussed in the next paragraph and in the Interpretation and Conclusions section. Statisti cally, this behavior is typical of an 1(2) process, and repeating the unit root tests for the (log arithmic) price changes, that is, for inflation, yields some evidence that postw ar prices can be modeled as an 1(2) process (evidence that infla tion is nonstationary), particularly fo r the postBretton W oods era beginning in 1973. The results are th erefore suggestive of the U.K. price level undergoing two shifts in its generating process. The first might be associat ed with the abandonm ent of the gold standard, shifting the series from 1(0) to an 1(1) process. (From figure 3 it is in fact clear that prices did not start a secular increase until m id-1933, some two years after the move from the gold stan dard.)13 A stable price level is certainly in acco r dance with what would be expected under the gold standard (or, in principle, any commodity standard). T h ere w ere fluctuations in the supply of gold, but in countries such as the United Kingdom, which had developed and stable bank ing systems, these fluctuations had only modest price level effects. The system was to some ex tent self-stabilizing. If prices w ere falling (the value of m oney rising) because the supply of gold was falling short of demand, th ere was an incentive to produce m ore gold. And if prices w ere rising (the value of m oney falling), then as the costs of gold production rose relative to what the m onetary authorities would pay for 12See Cagan (1984) for an extended discussion of this view. 13See the discussion in Capie, Mills and Wood (1986a). gold, the incentive to produce gold would diminish.14 The second shift is around 1973 and could be associated w ith both the move to float ing exchange rates and the first oil price shock. M on ey Figure 4 plots annual observations of the logarithms of M3 from 1871 to 1912 (the only aggregate apart from the m onetary base availa ble for this period), and figure 5 plots monthly observations of M3 from 1922 to 1989, exclud ing the w ar years. From a battery of unit root tests, we found that, for all sample periods in vestigated, the null hypothesis of a unit root cannot be rejected. M oreover, the series is in deed 1(1) because we could not establish that fu rth er differencing was required for stationarity. Interest Rates Figures 6 through 8 plot m onthly observations of short-and long-term interest rates from 1870 to 1992, excluding w ar years and related peri ods of interest rate restrictions. From the results of unit root tests, we find that since the lifting of restrictions after W orld W ar II both short- and long-term interest rates have been 1(1) processes, but their behavior b e fore 1939 is rath er different. Both are station ary betw een 1932 and 1939, but during the 1920s long-term rates are stationary (1(0)) and short-term rates are 1(1), w hereas before 1914 the orders of integration are reversed.15 Trend and Cycle Decom positions Has the variability about trend of the series altered across regimes? This is an im portant question because o f the widespread belief that floating exchange rates increase volatility in prices, interest rates, and econom ic activity and are in some general sense destablizing. To an sw er this question, w e need to decompose each series into trend and cycle components. There are many ways to do this, ranging from using a predeterm ined moving average to calculate trend to designing a signal extraction filter based on the stochastic process generating the data and a set of assumptions relating to the b e havior of the unobserved components. 1 5Capie, Mills and Wood (1986b) provides an extended dis cussion of the behavior of these two interest rate series in relation to the Stock Conversion of 1932. 14See Barro (1979) and Rockoff (1984) for a discussion of this. JULY/AUGUST 1993 10 Figure 4 Annual U. K. Money Supply: M3 (1871-1912) Logarithms For output, equation (1) provides the appropri ate decomposition. Table 1 thus reports the standard deviations of the cyclical com ponent nt fo r a variety of sample periods. The sample periods shown w ere chosen by two quite dis tinct criteria—output trend change and ex change rate regim e alteration. The 1922 break was used because after the 1 9 1 9 -2 2 discontinui ty, output resum ed a new trend, 1 8 5 5 -1 9 1 3 w ere gold standard years, and 1 9 2 5 -3 1 w ere years during which the United Kingdom was either on or com m itted to returning to the gold standard. The period comprising 1 8 5 5 -1 9 1 3 and 1 9 2 2 -3 1 is the same period omitting w ar and the postw ar years of the break in output's trend. The period 1 9 2 2 -3 1 has a stable output trend com bined with com m itm ent to gold; the period 1 9 2 2 -3 9 has a stable output trend with a change in exchange rate regime. The period 1 9 3 2 -9 0 is our whole sample period after gold. The years 1 9 3 2 -3 9 and 1 9 4 6 -9 0 are, of course, the same period excluding the W orld W ar II years. The period 1 9 4 6 -9 0 is simply postwar; 194 6 -7 2 is Bretton Woods; and 1 9 7 3 -9 0 is the period of various degrees of float. (Further sub division of the series to examine the association http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis with various exchange rate regim es more minutely, although appealing, is ruled out by many of these regim es having too few output observations for our statistical techniques.) From all these statistics, one gets the im pression that variability about trend has increased during the tw entieth century. In particular, the aban donment of the gold standard in 1931 seems to have been accompanied by an increased varia bility of output about trend, even after the w ar years are excluded. In summary, the standard deviation almost doubled (from 2.87 percent to 5.49 percent) after 1931. But it should be noted that variability fell after the pound floated in 1972. From 1946 to 1972 the standard deviation was 4.45 percent; from 1973 to 1990 it was 3.64 percent. For the other series, we have presented evi dence of shifts in the stochastic processes generating them, so signal extraction techniques would be rath er difficult to apply. W e have chosen therefore to use a technique that has proved popular in recen t years for re-examining the stylized facts of m acroeconom ic time series, namely the detrending filter proposed for use in econom ics by Hodrick and Prescott and used, 11 Figure 5 U.K. Money Supply: M3 (1922-1939) Logarithms U.K. Money Supply: M3 (1946-1989) Logarithms JULY/AUGUST 1993 12 Figure 6 U.K. Interest Rates (1870-1913) Percent Figure 7 U.K. Interest Rates (1922-1939) Percent http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis 13 Figure 8 U.K. Interest Rates (1954-1992) Percent for example, in Kydland and P rescott.16 This is an alternative to the method used earlier in the paper for separating a series into trend and cy cle. It is described in the appendix, which also contains a summary of w hen this method is ap propriate and w hen it may be misleading. Tables 2 through 5 report statistics assessing the variability of the trend and cycle com po nents of the price level, money supply and short- and long-interest rates, and figures 9 through 12 present graphs for these com po nents. Although these tables report results from the exam ination of monthly data, the b reak points are at year ends except for 1992, w hose data end with June. This choice of breakpoints reflects two con siderations. The first relates to w hen an ex change rate regime changed. Does chan ge for our purposes relate to w hen the change was formally announced or to w hen it becam e ex pected and affected behavior? The latter is the m ore significant, but it is not clear a p riori 16See Hodrick and Prescott (1980) and Kydland and Prescott (1990). w hen it would be. Nor as it turns out does detailed exam ination of the data case by case give clear-cut answ ers.17 Accordingly, the simple expedient of using calendar years as b reak points was adopted, on the grounds that using other dates close to these would not change the results. For the interw ar years, the trend of the price level was relatively flat, with a slow decline un til 1933 and an upward drift th ereafter. The cy clical component, in contrast, is relatively volatile, no doubt, in view of the unchanged be havior of money, reflecting the changes in ex change rate policy in the United Kingdom, as well as the disturbed external environm ent. Not only did the interw ar years include the Great Depression in the United States, with the as sociated severely depressing effects on the prices of commodities, but in continental Eu rope there w ere inflations—hyperinflations in some cases—civil w ar and revolutions. M ean while Britain's exchange rate regim e was chang ing rapidly. Betw een 1919 and 1925 there was a 17See Mills and Wood (1993). For a subset covering the years 1870-1939, see Capie and Wood (forthcoming). JULY/AUGUST 1993 14 Table 1 Table 3 Variability of the Cyclical Component of Output_________________________ Component Variability of Money 1922.01-1939.12 1946.01-1992.05 1946.01-1972.12 1973.01-1989.06 Standard Deviation Period 1855-1913 1855-1931 1855-1913 and 1922-1931 1922-1931 1922-1939 1932-1939 1932-1990 1932-1939 and 1946-1990 1946-1990 1946-1972 1973-1990 X 2.69 2.87 2.95 3.45 3.93 3.58 5.49 4.27 4.42 4.45 3.64 x: sx ST sc 7.90 10.04 9.27 11.29 0.10 1.08 0.27 0.67 0.10 1.08 0.27 0.67 0.02 0.02 0.02 0.02 sample mean sx: sample standard deviation sT: sample standard deviation of trend component sc : sample standard deviation of cycle component Table 4 Component Variability of Short-Term Interest Rates Table 2 Component Variability of Prices x X 1922.01-1939.12 1946.01-1992.05 1946.01-1972.12 1973.01-1992.05 x: sx ST sc 3.12 4.60 3.88 5.61 0.09 0.94 0.30 0.51 0.09 0.94 0.30 0.51 0.02 0.01 0.01 0.01 sample mean 1870.01-1913.12 1922.01-1931.12 1932.01-1939.12 1954.01-1972.12 1973.01-1992.04 x: sx ST sc 2.79 3.75 0.86 5.42 11.43 1.21 1.07 0.71 1.76 2.44 0.77 0.68 0.51 1.54 1.84 0.84 0.76 0.53 0.66 1.22 sample mean sx: sample standard deviation sx: sample standard deviation sT: sam p le standard deviation ot trend com ponent sT: sample standard deviation of trend component sc: sample standard deviation of cycle component sc : sample standard deviation of cycle component com m itm ent to retu rn to gold at the prew ar parity, and the exchange rate rose steadily toward that. Gold was abandoned in 1931, and the exchange rate th ereafter floated with vari ous degrees of intervention until the outbreak of w ar in 1939. A fter 1946, the trend is smooth and m onoton ic, and the cyclical com ponent is less volatile than before. Trend money is rath er similar to trend prices. Its variability is stable throughout the sample period, supporting the suggestion that external factors w ere im portant in interw ar price volatility. Pre-1914 trend interest rates fluctuate around 18We have noted this result in a series of previous papers. Mills and Wood (1982) suggested it was due to the stable price expectations provided by the gold standard. Mills FEDERAL RESERVE BANK OF ST. LOUIS 3 percent, although the far greater stability of long-term rates is reflected in the almost cons tant com ponents of this series relative to short term rates.18 Volatility is indeed fairly stable un til 1972, after which both trend and cycle com ponents becam e considerably m ore variable. INTERPRETATION AND CONCLUSIONS W hen discussing the preceding findings, it is convenient to consider the trend and cyclical behavior of each series together. W e start with output. As noted previously, the trend grow th and Wood (1992) was unable to reject this hypothesis after exhaustive testing. 15 Table 5 Component Variability of Long-Term Interest Rates X 1870.01-1913.12 1922.01-1931.12 1932.01-1939.12 1954.01-1972.12 1973.01-1992.04 x: sx ST sc 2.94 4.45 3.31 6.35 11.32 0.23 0.13 0.35 1.71 1.91 0.22 0.11 0.31 1.67 1.74 0.04 0.13 0.14 0.24 0.61 sample mean sx: sample standard deviation sT: sample standard deviation of trend component sc: sample standard deviation of cycle component o f output changed from 1.96 percent per year to 2.29 percent per year betw een 1919 and 1922. Speculating on what produced that w el come change is outside the scope of this paper. W hat we would note is the stability of the post-1922 trend in the face of a wide variety of m onetary experiences and exchange rate re gimes, a finding clearly consistent with the longru n neutrality of money. In contrast to that long-run neutrality, the cy clical behavior w as affected. The variability of output rose substantially w ith the abandonm ent of the gold standard. The significance of this is discussed later. Turning now to prices, what do we find? The first notable feature is the essentially flat trend, with long swings around it, under the gold stan dard. More dram atic and equally revealing about the nature of the m onetary regim e is the post-1946 period. The trend of prices was posi tive after 1946, accelerated sharply around 1973 and slowed around 1983. T he United Kingdom w ent to a floating exchange rate in 1972, but at around the same time there was also the first oil price shock and the H eath-Barber m onetary expansion. That the acceleration of prices was the result of these factors ra th er than the new exchange rate system is suggested by the slow ing of prices around 1983, w hen the United 1 9The role of the exchange rate regime in the 1970s episode is also discussed in Williamson and Wood (1976). The con clusion that the exchange rate regime was not at fault was also, by different means, argued there. Kingdom was still under a floating rate regime but had a governm ent strongly committed to reducing inflation by introducing money supply targets and a com m itm ent to budget balance over the cycle.19 The cyclical com ponent of prices becam e m uch sm oother and was un affected by the exchange rate regim e; its varia bility was unchanged from 1946 to 1992 and identical >ver subperiods and the period as a whole. And finally, interest rates. The striking con trast is betw een the behavior in the pre-W orld W ar II period, w hen long-term rates w ere sta ble and short-term rates w ere volatile, an obser vation usually interpreted as reflecting expectations o f long-run price level stability and behavior in the post-1972 period, w hen inflation first accelerated and then slowed, and both in terest rate series displayed markedly increased variability.20 How do these findings as a whole b ear on the hypothesis that the exchange rate regime is not a source of volatility? They support it. Of the variety of exchange rate regim es after 1913 (we turn to the gold standard in a moment), none seemed to increase the volatility of any series examined to any significant extent. T he policy changes necessary to hold rates pegged may have appeared in foreign exchange reserves, a series that we did not examine because reliable data w ere not available. The policy changes did appear in movements that had higher frequ en cies than the trends and cycles we isolate.21 In terest rate cyclical variability did increase with the move to floating exchange rates in 1972, but there arc num erous other factors to explain this. Sho ks to the price of oil disturbed finan cial m arkets very substantially in this period. Two other shocks w ere superimposed on the oil price shocks. T h ere was a com m itm ent to reduce inflation—particularly after 1979. W hat this meant in term s of the operation of m one tary policy was unknow n, so the commitm ent increased u ncertainty for a time. And further, m onetary targets w ere adopted. These affected how the authorities used short-term interest rates; and as com m itm ent to m onetary targets planations of the Gibson paradox that depend on slowmoving price expectations. 21See Batchelor and Wood (1982), Wood (1983) and Belongia (1988). 20See Mills and Wood (1982). Fisher (1930), Friedman and Schwartz (1982), and Mills and Wood (1992). All have ex JULY/AUGUST 1993 16 Figure 9 Price Level Trend and Cycle (1922-1992) Logarithms Logarithms Figure 10 Money Supply: M3 Trend and Cycle (1922-1989) Logarithms Logarithms 1922 26 30 34 38 42 46 50 54 58 62 66 70 74 78 821986 FEDERAL RESERVE BANK OF ST. LOUIS 17 Figure 11 Short Interest Rate Trend and Cycle (1870-1992) Percent Percent 3H------- 1------ 1------ 1------ 1------ 1------ 1------ 1------ 1------ 1------ 1------ 1------ r ~ -50 1870 80 90 1900 10 20 30 40 50 60 70 80 1990 Figure 12 Long Interest Rate Trend and Cycle (1870-1992) Percent 20 15- 10- --3 .0 Trend 1 ------1-----1---- 1-----1-----1-----1---- 1-----1-----1-----1---- r^-4.o 1870 80 90 1900 10 20 30 40 50 60 70 80 1990 JULY/AUGUST 1993 18 becam e increasingly credible, the relationship betw een m ovements in short- and long-term rates changed.22 It cannot but be observed that there was greater stability of output, interest rates and prices under the gold standard than under any subsequent exchange rate regime. But, of course, the gold standard was m ore than an ex change rate regim e. It was a system, a set of rules, for the conduct of m onetary policy. As Bordo (1993) w rote, "The gold standard rule can be viewed as a form of contingent rule or a rule with escape clauses. The m onetary authori ty maintains the standard—that is, keeps the price of the cu rren cy in term s of gold fixed— except in the event of a well-understood em er gency, such as a m ajor w ar or a financial crisis. In wartim e it may suspend gold convertibility and issue paper m oney to finance its expendi tures, and it can sell debt issues in term s of the nominal value of its cu rren cy on the und er standing that debt will eventually be paid off in gold. The rule is contingent in the sense that the public understands that the suspension will last only fo r the duration of the w artim e em er gency plus some period of adjustment. It as sumes that afterw ard the governm ent will follow the deflationary policies necessary to re sume payments at the original parity.” It may be consistent with this interpretation of the gold standard that with the floating exchange rate of the 1970s, output variability fell, but not to w here it had been under the gold standard. The argum ent would be that m onetary policy was now clearly focused on internal objectives and not subject to the vicissitudes of a multitude of shocks from the outside world. All in all, then, it appears clear that the ex change rate regim e in the United Kingdom has not been a source of volatility for the main m acroeconom ic variables. For that reason we need not consider why exchange rate regim es might affect real econom ic perform ance—in the United Kingdom they did not. The case for a fixed rate regim e in the United Kingdom ap parently m ust depend only on its traditional source of support—the desire to import price level perform ance. It is, of course, im portant to consider w hether these results generalize to other economies. 22lnitially, rises in short-term rates produced rises in long term rates. But as markets became convinced that the authorities were serious, short- and long-term rates started to move much more independently of each other. FEDERAL RESERVE BANK OF ST. LOUIS There is virtually no feature of the U.K. econo my to indicate that they should not.23 The com position of output is not unusual; the U.K. econom y has always been fairly open. It was a dominant econom y internationally for only a modest part of our period, and it has not gone through hyperinflation or recessions as severe as those in some other economies, so such problems cannot have biased our results. Though we would not claim that our findings are m ore than those of a case study, we would suggest that they are findings we would not be surprised to see roughly repeated in studies of other countries. REFERENCES Aschheim, Joseph, Martin J. Bailey and George S. Tavlas. “ Dollar Variability, the New Protectionism, Trade and Finan cial Performance,” in Dominick Salvatore, ed., The New Protectionist Threat to World Welfare (NorthHolland 1987), pp. 424-49. Bailey, Martin J., George S. Tavlas and Michael Ulan. “ Ex change Rate Variability and Trade Performance: Evidence for the Big Seven Industrial Countries,” Weltwirtch Archiv (Volume 122,1986), pp. 466-77. _______ “ The Impact of Exchange-Rate Volatility on Export Growth: Some Theoretical Considerations and Empirical Results,” Journal of Policy Modeling (Spring 1987), pp. 225-43. Barro, Robert J. “ Money and the Price Level Under the Gold Standard,” Economic Journal (March 1979), pp. 13-33. Batchelor, Roy A., and Geoffrey E. Wood. “ Floating Exchange Rates: The Lessons of Experience," in Roy A. Batchelor and Geoffrey E. Wood, eds., Exchange Rate Policy (St. Martin’s Press, 1982), pp. 12-34. Belongia, Michael T. “ Prospects for International Policy Coor dination: Some Lessons for the EMS,” this Review (July/August 1988), pp. 19-27. Bordo, Michael D. “ The Gold Standard, Bretton Woods, and Other Monetary Regimes: A Historical Appraisal,” this Review (March/April 1993), pp. 123-87. Bryant, Ralph C. International Financial Intermediation (The Brookings Institution, 1987). Cagan, Phillip. “ Mr. Gibson’s Paradox—Was it There?” in Michael D. Bordo and Anna J. Schwartz, eds., A Retro spective on the Classical Gold Standard, 1821-1931 (Univer sity of Chicago Press, 1984), pp. 604-10. Capie, Forrest H., Terence C. Mills and Geoffrey E. Wood. “ What Happened in 1931?” in Forrest H. Capie and Geoffrey E. Wood, eds., Financial Crises and the World Banking System (St. Martin’s Press, 1986a), pp. 120-48. 23This is also suggested by the similar structure of models used to explain and predict the economies of a wide range of countries. 19 _______ “ Debt Management and Interest Rates: the British Stock Conversion of 1932,” Applied Economics (Volume 18, 1986b), pp. 1111-26. King, Robert G., and Sergio Rebelo. “ Low Frequency Filtering and Real Business Cycles,” University of Rochester Work ing Paper No. 205 (1989). Capie, Forrest H., and Geoffrey E. Wood. “ Money in the Economy, 1870-1939,” in Roderick Floud and Donald McClosky, eds., An Economic History of Britain (Cambridge University Press, forthcoming). Kyland, Finn E., and Edward C. Prescott. “ Business Cycles: Real Facts and a Monetary Myth,” Federal Reserve Bank of Minneapolis Quarterly Review (Spring 1990), pp. 3-18. Cushman, David O. “ The Effects of Real Exchange Rate Risk on International Trade,” Journal of International Economics (August 1983), pp. 45-63. deGrauwe, Paul, and Bernard deBellefroid. “ Long-Run Ex change Rate Variability and International Trade,” in Sven Arndt and J. David Richardson, eds., Real Financial Link ages Among Open Economies (MIT Press, 1987), pp. 193-212. Feldstein, Martin. “ The Case Against EMU,” The Economist (June 13, 1992), pp. 19-22. Fisher, Irving. The Theory of Interest (Macmillan, 1930). Friedman, Milton, and Anna J. Schwartz. Monetary Trends in the United States and the United Kingdom (University of Chicago Press, 1982). Gotur, Padma. “ Effects of Exchange Rate Volatility on Trade: Some Further Evidence,” IMF Staff Papers (September 1985), pp. 475-512. Haberler, Gottfried. “ The International Monetary System,” The AET Economist (July 1986). Harvey, A.C., and Alfred Jaeger. “ Detrending, Stylized Facts and the Business Cycle,” London School of Economics Discussion Paper No. EM/91/230, 1991. Hodrick, Robert J., and Edward C. Prescott. “ Postwar U.S. Business Cycles: An Empirical Investigation,” CarnegieMellon University Discussion Paper No. 451 (1980). Hume, David. “ Of the Balance of Trade, ” in Eugene Rotwein, ed., David Hume, Writings on Economics (The University of Wisconsin Press, 1970), pp. 60-77. International Monetary Fund. “ Exchange Rate Volatility and World Trade,” Occasional Paper No. 28 (1984). Mills, Terence C. "Are Fluctuations in U.K. Output Transitory or Permanent?” Manchester School of Business (Volume 59, 1991), pp. 1-11. _______ “A Note on the Gibson Paradox During the Gold Standard,” Explorations in Economic History (July 1990), pp. 277-86. Mills, Terence C., and Geoffrey E. Wood. “ Capital Flows and the Excess Burden of the Exchange Rate Regime,” Hull Economic Research Paper (1993). _______ “ Econometric Evaluation of Alternative Money Stock Series, 1880-1913,” Journal of Money, Credit and Banking (May 1982), pp. 265-77. _______ “ Money and Interest Rates in Britain From 1870-1913,” in Nick Crafts and Stephen Broadberry, eds., Britain in the International Economy 1870-1939 (Cambridge University Press, 1992), pp. 199-217. Rockoff, Hugh. “ Some Evidence on the Real Price of Gold, Its Costs of Production, and Commodity Prices,” in Michael D. Bordo and Anna J. Schwartz, eds., A Retrospective on the Classical Gold Standard, 1821-1931 (The University of Chicago Press, 1984), pp. 613-44. Schumpeter, Joseph Alois. Capitalism, Socialism, and Democracy, 3rd ed. (Harper, 1950). Williamson, John, and Geoffrey E. Wood. “ The British Infla tion: Indigenous or Imported?” American Economic Review (September 1976), pp. 520-31. Wood, Geoffrey E. “ The European Monetary System: Past Developments, Future Prospects and Economic Rationale,” in Richard Jenkins, ed., Britain and the EEC (Macmillan, 1983). Appendix The Hodrick-Prescott Filter The filter proposed by Hodrick and Prescott (1980) has a long tradition as a m ethod o f fitting a smooth curve through a set of points, ver sions of it being used as an actuarial graduation formula. Given the traditional decomposition y, = M + n,> the trend series /i( is obtained as the , solution to the problem of minimizing T (2) y, = A[^/+2- 4 M + r6 + A-J)M ,+J , + M ,J Using the lag operator B, defined such that = this can be w ritten as ( 3 ) Yt = X[B 2- 4 B 1+ ( 6 + X 1 - 4 B + B % t ) = m - B f a - B 1 )2+ \ T ( 1) so that if an infinite series of y values w ere available, would be given by the two-sided moving average co with respect to \2, x \ir The first ord er con dition for this minimization problem is ( 4) u — Lj j. oo or Y j , a < -j ’ j — or -j JULY/AUGUST 1993 20 w here the weights can be calculated from (5) a(B) = [A ( 1 - B H 2 -B _1)2+ 1 ] " !. by the structural model (6) y t = nt + n, M = M,-. + v ,-, , King and Bebelo (1989) provide expressions for the ay w hich do not take a simple form . For tunately, Hodrick and Prescott (1980) provide an algorithm that rem oves the need to calculate the moving average weights and so allows the trend to be computed w hen only a finite num ber of y observations are available. This al gorithm was employed to com pute the decompositions used here, noting that the cycli cal com ponent can be obtained by residual as nl = y t-\xt. Typically, following Hodrick and Pres cott, A is set at 100 if annual data are used or 1,600 if quarterly or monthly data are used.1 Harvey and Jaeger (1991), fo r example, show that the filter a(B)yt can be interpreted as being the optimal estim ate of w hen y t is generated 1See Hodrick and Prescott (1980). FEDERAL RESERVE BANK OF ST. LOUIS V, = n'NID (0,0;), £~MD (0,ct|), A = o 2lo* n An observed series may not be generated even approximately by such a model, and even if it is, the ratio of the tw o innovation variances may be very different from the assumed value of A Harvey and Jaeg er argue that the Hodrick. Prescott filter may create spurious cycles, dis tort the estim ates of the com ponents or both. King and Bebelo argue in similar vein, although they focus on the calculation of sample moments of the estim ated trend and cycle components. Given these strictures, we emphasize that our use of the filter is purely for exploratory pur poses outside the confines of any explicit model. 21 Patricia S. Pollard Patricia S. Pollard is an economist at the Federal Reserve Bank of St. Louis. Heather Deaton and Richard D. Taylor provided research assistance. Central Bank Independence and Economic Performance J n RECENT YEARS MANY countries have adopted or made progress tow ard adopting legislative proposals removing their central banks from governm ent control, that is, making them independent. Between 1989 and 1991, New Zealand, Chile and Canada enacted legisla tion that increased the independence of their central banks. T he 1992 Treaty on European Union (M aastricht Treaty) requires European Community (EC) m em bers to give their central banks independence as part of establishing the European M onetary Union. As a result, EC countries that do not yet have strongly indepen dent central banks have introduced legislation or announced their com m itm ent to make their central banks m ore independent.1 Furtherm ore, in recen t months the governm ents of Brazil and Mexico have announced their intentions to in troduce legislation to create m ore independent cen tral banks. In view of these developments, it might seem reasonable to conclude that unambiguous links had been established betw een econom ic p erfor 1To meet the level of independence prescribed by the Maas tricht Treaty, a central bank must be prohibited from taking instructions from the government. The term for central bank governors must be set at a minimum of five years, although it can be renewed. In addition, the central bank must be prohibited from purchasing debt instruments directly from the government (that is, in the primary market) and from providing credit facilities to the govern ment. Both Denmark and the United Kingdom have reserved the right to decline membership in the European Monetary Union. Thus neither country has introduced m ance and the degree of central bank indepen dence. Interestingly, how ever, the two postW orld W ar II star perform ers among the indus trialized econom ies—Germany and Japan—have d ifferent levels of cen tral bank independence. The German Bundesbank is viewed as one of the most independent central banks in the world, w hereas the Bank of Japan is seen as m ore subject to governm ent control. Thus the contrast betw een the movement to grant central banks m ore independence and widely different degrees of independence across the m ajor econ omies raises several questions. Among these are: W hy is the idea of an independent central bank popular? Are th ere econom ic benefits of having an independent central bank? This paper exam ines em pirical and theoretical studies of central bank independence to address these questions. Empirical research ers have de vised m easures of independence to focus on the relationship betw een central bank independence and a country's econom ic perform ance. T h eo retical studies have modeled the strategic belegislation to ensure conformity of their central banks with the Maastricht provisions. For a detailed analysis of the institutional status of the central banks of the EC countries, see the Committee of Governors of the Central Banks of the Member States of the European Economic Community (1993). JULY/AUGUST 1993 22 havior of m onetary and fiscal policym akers to be able to com pare an econom y’s perform ance w hen policym akers cooperate in setting policies with its perform ance w hen they do not cooperate. The next section of this paper presents a sur vey and evaluation of em pirical studies.-Next, theoretical studies are presented and evaluated. The final section exam ines the extent to which these studies eith er explain the cu rren t move ment tow ard greater central bank independence or highlight unresolved questions in this debate. EMPIRICAL STUDIES: CENTRAL HANK INDEPENDENCE AND ECONOMIC PERFORMANCE Inflation and Central Bank Independence As a broad generalization, interest in central bank independence was motivated by the belief that, if a central bank was free of direct politi cal pressure, it would achieve low er and m ore stable inflation.2 Bade and Parkin (1985) con ducted one of the first em pirical studies of this link. The authors used data for 12 Organization for Econom ic Cooperation and Development (OECD) countries in the post-Bretton W oods era and m easured the degree of central bank in dependence according to the extent of govern m ent influence over the finances and policies of the central bank.3 The degree of financial in fluence on the central bank was determ ined by the governm ent’s ability to set salary levels for m em bers o f the governing board of the central bank, to control the cen tral bank’s budget and to allocate its profits. The degree of policy in fluence was determ ined by the governm ent’s ability to appoint the m em bers of the central bank governing board, governm ent rep resen ta tion on this board, and w h eth er the govern ment or the central bank was the final policy authority. Countries w ere given a rank of one through fou r in each category, with fou r being the highest level of central bank independence. Bade and Parkin concluded that the degree of financial independence of the central bank was 2Buchanan and Wagner (1977) point out that even an in dependent central bank may not be immune from political pressures and thus exhibit an inflationary bias. 3The 12 OECD countries are Australia, Belgium, Canada, France, Germany, Italy, Japan, Netherlands, Sweden, Swit zerland, United Kingdom, and United States. http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis not a significant determ inant of inflation in the post-Bretton Woods period. Policy indepen dence, how ever, was seen as an im portant de term inant of inflation because the two countries with the highest degree of policy independence (Germany and Switzerland) had inflation rates significantly below those of all other countries in the sample. They found no significant d iffer ences in inflation perform ance among countries with low er rankings of independence in the post-Bretton Woods era. Alesina (1988) used the Bade and Parkin (1985) index but added the following fou r countries: Denm ark, New Zealand, Norway and Spain. He found, as hypothesized, that th ere was generally an inverse relationship betw een average infla tion rates and the level of central bank in dependence. Grilli, Masciandaro and Tabellini (1991) created two indexes of central bank independence—one based on econom ic m easures of independence (with a scale ranging from zero to eight), and the other based on political m easures of in dependence (with a scale ranging from zero to seven).4 The political factors w ere similar to those identified by Bade and Parkin. The eco nomic factors considered w ere the ability of the governm ent to determ ine the conditions under w hich it can borrow from the central bank and the m onetary instrum ents under the control of the cen tral bank. The data set com prised 18 OECD countries over the period 1 9 5 0 -8 9 .5 For the period as a whole, Grilli, M asciandaro and Tabellini found that econom ic independence was negatively related to inflation. Political in dependence also had a negative correlation with inflation, but the relationship was not statistical ly significant. Breaking the data into fou r decade-long subperiods, they found that neither m easure of independence had a significant e f fect on inflation in the first tw o decades. In the 1970s both m easures of independence w ere sig nificant, w hereas in the 1980s only the econom ic independence m easure was significant. Alesina and Summ ers (1993) calculated a m easure of central bank independence by aver aging the indexes created by Bade and Parkin, 4ln both measures the scale is increasing in the level of in dependence. 5Grilli, Masciandaro and Tabellini add Austria, Denmark, Greece, New Zealand and Portugal to Bade and Parkin’s group of countries and eliminate Sweden. 23 Figure 1 Average Inflation: 1955-1988 Percent •I■Spain 8- New Zealand 7- 6 Italy ^United Kingdom Australia * * Denmark *** France / Norway / Sweden - .j. Japan Canada 54- Belgium •h The Netherlands United States Switzerland Germany 3- -----------------1 ------------------------- 1---------------- 2 3 Index of Central Bank Independence Source: Alesina and Summers (1993). and Grilli, Masciandaro and Tabellini.6 The countries included w ere the same as in Bade and Parkin with the addition of Denm ark, New Zealand, Norway and Spain. The sample period was 1 9 5 5 -8 8 .7 As in the previous studies, they found a negative correlation betw een the level of central bank independence and the rate of inflation (figure 1). They also found that the m ore dependent a central bank was, the greater the variability in inflation (figure 2). This, they argued, was a result of a correlation betw een the level and variability of inflation. Cukierman (1992) provided an extensive analy sis of central bank independence and its rela tionship to inflation perform ance using data for 1 9 5 0 -8 9 . Unlike previous studies, he used not only legal m easures of central bank indepen dence, but also practical m easures of the level of independence. One such m easure was the frequency of turnover of the central bank governors. Another m easure of practical in dependence was based on answ ers from a ques tionnaire completed by qualified individuals at the central banks.8 Cukierm an’s analysis is the most com prehensive to date, not only because it incorporates inform ation about the actual level o f independence a central bank enjoys in prac tice, but also because it includes a sample of 70 countries.9 Cukierman concluded that "central 6See Bade and Parkin (1985) and Grilli, Masciandaro and Tabellini (1991). 7See Alesina (1988). Alesina and Summers report that the results of their study are the same if the data period is res tricted to 1973-1988, the post-Bretton Woods era. 8The sample period for the questionnaire data was 1980-89. 9The questionnaire data were available for only 24 countries. JULY/AUGUST 1993 24 Figure 2 Variance of Inflation: 1955-1988 Percent 40 •fr Italy 30Spain + •I* United Kingdom Australia / France •I* Japan New Zealand 20- j,Sweden Norway j . 4 Canada B e lg iu m * Denmark 10 - •fr United States *T h e Netherlands ^Switzerland Germany 0 0 1 2 3 Index of Central Bank Independence 4 5 Source: Alesina and Summers (1993). bank independence affects the rate of inflation in the expected direction.”10 This result was also found by Cukierm an, W ebb and Neyapti (1992).11 Central Bank Independence and the Real E conom y dence and econom ic output. If an independent central bank can produce low er inflation than a dependent central bank, does this com e at the cost of low er output? Conversely, are dependent central banks attem pting to exploit a short-run Phillips Curve relationship, accepting higher in flation in order to achieve higher output? Although most of the em pirical w ork focused on the relationship betw een central bank in dependence and the rate of inflation, some studies examined the link betw een indepen- Grilli, Masciandaro and Tabellini (1991) found no system atic effect of central bank indepen dence (using either of their two indicators) on the grow th rate of real output. Alesina and 10Cukierman did not actually use the rate of inflation, but the rate of depreciation of the real value of money, defined by the following formula: d, = n, 1 + n, where n, is the inflation rate in period t. The use of d, as noted by Cukierman, moderates the effects of hyper inflation on the results. 11Capie, Mills and Wood (1992) also studied the link between inflation and central bank independence. Their data set consisted of 12 countries, with the data series beginning between 1871 and 1916 and ending in 1987. Central FEDERAL RESERVE BANK OF ST. LOUIS banks were classified as either dependent or independent according to the extent of their control over monetary poli cy. The authors examined the relationship between the sta tus of the central bank and inflation over the entire sample period and four subsample periods—pre-World War I, the Interwar Years, Bretton Woods and post-Bretton Woods. Periods of hyperinflation, however, were excluded from the data. In all sample periods, the countries with independent central banks were in the low inflation group. Nevertheless, some of the dependent central banks were also in this group. The authors concluded that independence may be a sufficient condition for low inflation but not a necessary one. 25 Figure 3 Average Real GNP Growth: 1955-1987 Percent * Japan 6- S p a in * 4- Australia/ Norway * Canada " a 'v * * F ran ce j . The Netherlands .j. Germany Denmark ?Sw eden * United States •I* Switzerland •I* United Kingdom n . i . i , i m i T< 3H New Z e a la n d * 2 “ I-------------------------1 ------------------------- 1 ------------------------- 1 ------------------------- 1 ---------------------- 0 1 2 3 Index of Central Bank Independence 4 Source: Alesina and Summers (1993). Summers (1993) likewise found no correlation betw een average econom ic grow th or the varia bility of grow th and the level of cen tral bank independence (figures 3 and 4).12 De Long and Sum m ers (1992) looked at the relationship betw een central bank independence and output per w ork er while trying to eliminate differences betw een countries that w ere due solely to convergence effects.13 To do this, they exam ined the grow th rate of real gross domes tic product (GDP) per w ork er during 1 9 5 5 -9 0 , controlling for the level of GDP per w orker in 1955 .14 This procedure showed a positive rela1 2The results are the same if per capita gross national product (GNP) is used rather than GNP. ,3Standard neoclassical growth models suggest that growth rates of economies tend to converge over time. Thus given two countries, the one with the lower per capita output will have a higher growth rate than the other until their levels of real output per capita converge. 14GDP per worker levels are based on the Summers and Heston (1991) estimates, which use purchasing power parity conversions. tionship betw een central bank independence and econom ic grow th.15 More precisely, they found that holding constant the 1955 level of real output per w orker, a unit increase in their index of cen tral bank independence was as sociated with a 0.4 percentage point increase in grow th per y ear.16 In contrast, Cukierman, Kalaitzidakis, Summers and W ebb (1993) found that output grow th in industrialized countries was unrelated to central bank independence even after controlling for structural factors that might influence grow th. The factors they considered w ere the initial level 15This study does not take into account that the degree of in dependence of the central bank of New Zealand changed dramatically in 1989. Furthermore, all of the studies, with the exception of Alesina (1988), do not take into account that there was an institutional change in the structure of the Bank of Italy in 1981 that increased its independence. The latter change, however, was not as substantial as the former. 16De Long and Summers regress the average growth rate of GDP per worker over the period 1955-90 on GDP per worker in 1955 and the central bank independence index. JULY/AUGUST 1993 26 Figure 4 Variance of Real GNP Growth: 1955-1987 Percent * Japan •I* Spain •I" Switzerland -T h e Netherlands ^ Denmark ^. New Zealand •b . . .. * Belgium Australia . United Kingdom / F ra n c e * Canada •I* Sweden *!• Germany United States Norway ------------------------- 1 ------------------------- 1 ------------------------- 1 ------------------------- 1------------------------ 0 1 2 3 Index of Central Bank Independence 4 5 Source: Alesina and Summers (1993). of a country’s GDP, its initial enrollm ent rates for primary and secondary education, and changes in its term s of trade. The authors did find, however, using the turnover rate of central bank governors as a proxy for independence, that central bank independence did have a posi tive effect on grow th in developing countries. The difference in the results for industrialized countries versus developing countries, they ar gue, may imply that “dependence on political authorities is bad for grow th only when the lev el of independence is sufficiently high."17 Cen tral bank independence is higher in all the industrialized countries than in most of the de veloping countries. Central Bank Independence and Fiscal Deficits Another area of empirical study has been the relationship betw een central bank independence 17See Cukierman, Kalaitzidakis, Summers and Webb (1993), p. 42. http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis and fiscal deficits. The motivation for these studies is the belief that independent central banks should be b etter able to resist govern ment efforts to have them monetize deficits. Thus governm ents realizing that th ere may be some limit on their ability to issue bonds con tinuously to finance deficits may decide to limit deficit spending. Parkin (1987) investigated this question for the same 12 countries as Bade and Parkin for the period 1 9 5 5 -8 3 .18 He found that there was some evidence of a negative relationship betw een cen tral bank independence and the long-run be havior of governm ent deficits as a percent of gross national product (GNP). The deficits of Switzerland and Germany, the countries w ith the highest levels of central bank independence, had long-run equilibrium values n ear zero w ith little variance. However, other countries, notably France, that had low levels of central bank in18See Bade and Parkin (1985). 27 dependence also had small long-run deficits as a p ercent of GNP. Masciandaro and Tabellini (1988) looked at fis cal deficits as a percent of GDP in Australia, Canada, Japan, New Zealand and the United States during the period 1 9 7 0 -8 5 .19 They found that New Zealand, w hich had the lowest level of central bank independence of the five countries during this period, had the highest fiscal deficit as a percent of GDP. The United States, how ever, w ith the highest level of central bank independence among this group of countries, had a deficit/GDP ratio similar to those of the other countries. Grilli, Masciandaro and Tabellini (1991) found that th ere was generally a negative correlation betw een the deficit/GNP ratio and the degree of central bank independence. However, if political factors, as well as central bank independence, w ere included in their regression, the latter variable was insignificant.20 Thus they conclude that an independent m onetary authority appar ently does not discourage the governm ent from running fiscal deficits. A fu rth er exam ination of the relationship b e tw een fiscal deficits and central bank indepen dence, w hich is consistent with the w ork done by Alesina and Sum m ers and De Long and Sum m ers, is presented h ere.21 Using the same index of central bank independence and the same 16 countries as these previous papers, th ere is some evidence of a negative correlation b e tw een average deficits as a percent of GDP and central bank independence for the period 19 7 3 -8 9 , as shown in figure 5 .22 The degree of independence, however, is not a statistically significant (at a = .05) determ inant of the deficit/GDP ratio. The variability of deficits as a percent of GDP is also negatively correlated w ith central bank independence (figure 6) and this relationship is statistically significant. EVALUATION OF THE EMPIRICAL STUDIES At first glance, these studies seem to indicate that a country that wants to low er its inflation rate and do so w ithout hurting grow th should create an independent cen tral bank. Such a cen tral bank apparently could also help reduce fis cal deficits and increase output. These benefits would explain the recen t popularity of indepen dent central banks. Thus Grilli, M asciandaro and Tabellini commented: Having an independent central bank is almost like having a fre e lunch; th e re are b enefits but no apparent costs in term s o f m acroeconom ic perfo rm ance.23 Alesina and Sum mers (1993) w ent a step fu rth er in concluding their findings: "Most obviously they suggest the econom ic perform ance m erits of central bank independence.”24 A m ore careful analysis of these studies, however, indicates w eaknesses that highlight the need fo r fu rth er evidence before one should believe that creating an independent cen tral bank will improve a country's economic perform ance. The following four weaknesses are considered: 1) the difficulty in m easuring central bank independence; 2) the possibility of a spurious relationship betw een independence and econom ic perform ance; 3) the possible en dogeneity of central bank independence; and 4) the inclusion of the fixed exchange rate period in the sample data of some of the studies. The m easures of central bank independence used in em pirical studies have been determ ined by establishing a set of factors thought to be relevant for independence and then analyzing central bank ch arters and laws for compliance with these factors. W ith the exception of the in- 19The deficits are as a percent of GNP for Japan. 23See Grilli, Masciandaro and Tabellini (1991), p. 375. 20These political factors include the frequency of government changes, significant changes in the government and the percent of governments in a given period supported by a single majority party. 24See Alesina and Summers (1993), p. 159. Even the press has picked up the banner of central bank independence. A recent headline in The Washington Post proclaimed: “ More Independence Means Lower Inflation, Studies Show.” See Berry (1993). 2 See Alesina and Summers (1993) and De Long and Sum 1 mers (1992). 22The 1989 ending date was chosen because of the change in the status of the Bank of New Zealand, which occurred in 1989. All data are from the International Monetary Fund, International Financial Statistics. JULY/AUGUST 1993 28 Figure 5 Average Deficit as a Percent of GDP: 1973-1989 Percent 12.5 Italy 10 •{• Belgium 7.5 New Zealand^ T 5 , . * Japan Spain + 2.5 AThe Netherlands j, t Canada , Sweden / * United States United Kingdom Australia / France .j. Germany Norway*^ * Denmark -I* Switzerland 0 0 1 2 3 Index of Central Bank Independence 4 5 Figure 6 Variance of Deficit as a Percent of GDP: 1973-1989 Percent •I* Sweden *1* Denmark New Zealand •i' *** Norway 10 •fr Belgium United Kingdom . Spain * * + T h e Netherlands 5 Italy Japan A ustria* Canada + France 0 0 1 http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis j,United States ^.Germany 4* Switzerland 2 3 Index of Central Bank Independence 4 5 29 dex created by Cukierman, all of the indexes of independence apply equal w eight to each factor. For instance, the Grilli, M asciandaro and Tabelli ni index based on political m easures of indepen dence gives a country one point if no one on the central bank board is appointed by the governm ent and one point if the policy form u lated by the central bank does not require ap proval by the governm ent. Although the latter certainly places a greater constraint on the ac tions of the central bank than the form er, the two are treated the same empirically. Another concern is that the studies are based on a legal m easure of independence that may not reflect a bank's d e fa c t o level of indepen dence. If th ere is a difference betw een legal and practical independence, studies based on the form er type of m easures may provide m islead ing results. Cukierman (1992), in an attem pt to address this possibility, uses central bankers' responses to a questionnaire to determ ine the actual degree of independence in the 1980s. He finds that the correlation betw een the legal in dex and this practical index of independence is 0.33 for developed countries, 0.06 for devel oping countries and 0.04 overall.25 This finding indicates, as Cukierman notes, that a legal index of independence is not useful for studying de veloping countries. It also indicates that a legal index may be a weak m easure of actual in dependence fo r the developed countries. T h ere also may be bias in the factors selected to m easure independence. For example, Grilli, M asciandaro and Tabellini include: “statutory re quirem ents that central bank pursues m onetary stability amongst its goals” in their index.26 Like wise, a central bank is m ore independent under Cukierm an’s system if price stability is its only objective than if price stability is one of a num b er of objectives or not an objective at all. Us ing the goal of price stability as a m easure of central bank independence may result in a bias betw een the m easure of independence and the inflation rate. Table 1 Comparison of Relative Rankings of Central Bank Independence________ Alesina Australia Belgium Canada Denmark France Germany Italy Japan Netherlands New Zealand Norway Spain Sweden Switzerland United Kingdom United States Alesina and Summers Cukierman 14 5 5 5 5 1 13 3 5 14 5 14 5 1 5 3 8 8 4 4 8 1 14 4 4 16 8 15 8 1 8 3 7 14 5 4 9 2 12 15 6 10 16 13 10 1 7 3 relative rankings as given by Alesina, Cukierman, and Alesina and Sum m ers.27 All agree that Swit zerland and Germany have the m ost independent central banks of the countries studied. T h ere are, however, a few countries w hich are ranked quite differently by the authors. For example, Japan has the second lowest level of indepen dence of all 16 countries, according to Cukier man, w hereas Alesina, and Alesina and Summers give it a m uch higher level of independence. The problem s in developing precise m easures of central bank independence are less im portant, however, if there is a consensus in ranking cen tral banks within broad levels of independence. Table 1 lists 16 OECD countries along with their This discrepancy over the degree of indepen dence of the Bank of Japan is not due solely to differences in factors considered in m easuring independence. The index used by Alesina is based on the criteria of independence created by Bade and Parkin (1985). The index used by Alesina and Sum mers is constructed by averag ing the indexes created by Alesina, and Grilli, Masciandaro and Tabellini. Bade and Parkin claim that the Bank of Japan is independent from the governm ent in form ulating and im plementing m onetary policy, and Grilli, M ascian daro and Tabellini claim that th ere are no provisions for handling policy conflicts betw een the Bank of Japan and the governm ent. In con trast, Cukierman claims that the Bank of Japan and the governm ent form ulate policy jointly and 25The correlations are based on the weighted indexes. Giving each factor related to independence an equal weight in the indexes results in a correlation of 0.01 for de veloped countries and 0.00 for developing countries. 27The measure of independence developed by Cukierman is based on more factors than the measure used by Alesina, and Alesina and Summers. Thus Cukierman’s rankings are more delineated than the other two. 26See Grilli, Masciandaro and Tabellini (1991), p. 368. JULY/AUGUST 1993 30 fu rth er notes that in the case of a policy con flict, the executive bran ch of the governm ent has final authority.28 Since most of the em pirical studies consider only central bank independence as a d eter m inant of econom ic perform ance, it is possible that if oth er factors are accounted for, these results could be spurious. Grilli, M asciandaro and Tabellini attem pt to account for other fac tors that could affect the rate of inflation by in cluding political variables. They find that after accounting fo r political factors, central bank in dependence was still negatively related to infla tion in the countries studied over the period 1 9 5 0 -8 9 . The incorporation of political variables is a step in the right direction, but other factors also should be considered. As noted by Cukier man, “m onetary policy is generally sensitive to shocks to governm ent revenues and expendi tures, employment, and the balance of pay m ents.”29 The types of shocks that a country experienced over the sample period and the reaction of the central bank to these shocks can affect its econom ic perform ance. A study by Johnson and Siklos (1992) found that the reac tions of central banks (as m easured by changes in interest rates) to shocks to unemploym ent, in flation and w orld in terest rates w ere not closely related to standard m easures of central bank in dependence. Empirical use of these indexes may be proble matic if central bank independence is an en dogenous variable in the sense that countries with a com m itm ent to price stability may have a g reater propensity fo r independent central banks. If this is true, the m ere establishm ent of an independent bank without a com m itm ent to price stability will not bring inflation benefits to a country. In fact, a public aversion to inflation predates the establishm ent of many independent central banks. This was true fo r the creation of the Bundesbank and m ore recently with respect to central banks in Chile and New Zealand. New Zealand had one of the highest inflation rates of all industrialized countries in the 1980s. In 1989 legislation was passed to increase the indepen dence of its central bank substantially. This 28Aufricht (1961) reproduces the Bank of Japan charter and subsequent changes in its governing regulations, which support the conclusion reached by Cukierman. 29See Cukierman (1992), p. 438. 30See McCallum (1989), pp. 285-88, for an explanation of the limitations on monetary policy under a fixed exchange rate system. http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis change is often credited with bringing inflation down to near zero. Though the legislation c e r tainly formalized the country’s com m itm ent to price stability, New Zealand had succeeded in reducing its inflation rate from nearly 16 p er cent in 1987 to 6 percent b efore the creation of an independent central bank. In theory, the degree of independence of a central bank should not be a determ inant of a country’s inflation perform ance under a fixed exchange rate system because m onetary policy cannot be set exogenously.30 During the BrettonWoods era, it is not clear that any central bank (with the possible exception of the U.S. Federal Beserve) could be considered independent in the sense of an ability to pursue an independent m onetary policy.31 Thus the em pirical finding of a negative relationship betw een independence and inflation w hen the sample period extends over both the Bretton Woods and post-Bretton Woods eras may indicate a flaw in these studies. To assess the effect of central bank indepen dence on inflation, the data used in these studies could be divided into tw o periods. If no evidence of a relationship betw een indepen dence and inflation is found in the Bretton Woods period, this would strengthen the under lying argum ent of these studies that central bank independence is a prim ary determ inant of a country’s inflation perform ance.32 If, how ever, evidence is found of a relationship betw een cen tral bank independence and inflation in the Bretton W oods period, this would conflict with theory and could indicate that the em pirical findings are spurious. THEORETICAL MODELS OF FIS CAL AND MONETARY POLICY IN TERACTIONS In contrast to the em pirical studies, the theo retical studies of cen tral bank independence and econom ic perform ance concentrate on the con flicts that can arise w hen m onetary and fiscal policy are delegated to independent institutions. In this literature an independent central bank is one that does not cooperate with the fiscal au3 Indeed, the primary argument in favor of a flexible ex 1 change rate system was that such a system would permit individual countries to pursue independent monetary poli cies. See, for example, Friedman (1953) and Johnson (1969). 32This is Grilli, Masciandaro and Tabellini’s finding (1991). 31 thorities in setting econom ic policy. A depen dent central bank is one that cooperates with the fiscal authority in setting policy. In examining the theoretical implications of central bank independence, this paper focuses on models in w hich the policymaking process is decentralized.33 The basic fram ew ork of these models is as follows. The governm ent controls fiscal policy, and the central bank controls m onetary policy. Both parties set goals for the econom y (generally inflation and output targets) and assign priority to these goals. The goals and priorities may differ across the policymakers. Each institution uses the instrum ents available to it in an attempt to reach its goals. In most models the central bank controls the grow th rate of the m onetary base and the governm ent controls fiscal spending. T h ere is an underlying model of the econom y that indicates how fiscal and m onetary policy will affect the relevant eco nomic variables. All of the models assume that th ere are no stochastic shocks to the economy. The governm ent and the central bank can either cooperate in implem enting their policies or choose not to cooperate. If they do not co operate, they either can set policies simultane ously, or one party can set its policies first and the other then adopts its policies in reaction to these. Consider Andersen and Schneider’s (1986) sim ple model in w hich the governm ent and the central bank establish targets for inflation and output.34 The fu rth er the actual level of output and rate of inflation are from their respective targets, the m ore disutility each authority r e ceives. Thus, using the following equations, each authority can be modeled as setting policy to minimize its respective loss functions:35 (1) Lf = a J y - Vj)2 + bfln - nfl2 (2) L m = a m ( y - y m )2 + b m ( n - n m )z V J af > bf bm — am > (3) nf > nm, yf > y m 33There have been studies concentrating solely on monetary policy that have shown that better economic outcomes result from the policymaker placing a greater weight on in flation than society as a whole. Rogoff (1985) argues that these results indicate the economic benefits of central bank independence. These studies ignore the interaction of fiscal and monetary policy in determining economic out comes and thus are not discussed here. 34Generally it is assumed that the government places more weight on meeting its output target than its inflation target, whereas the opposite holds for the central bank. Further where: L f is the fiscal authority’s loss function L m is the m onetary authority’s loss function y is output n is inflation yf is the fiscal authority’s output target y m is the m onetary authority's output target nf is the fiscal authority's inflation target nm is the m onetary authority’s inflation target a is the weight placed on the output target b is the weight placed on the inflation target Andersen and Schneider com pare the economic outcomes under cooperation vs. noncooperation given th ree different models of the economy. The first model is Keynesian in nature. This is a short-run model with price sluggishness so that even anticipated changes in policy affect ag gregate demand. The level of output and the rate of inflation prevailing in the econom y are affected by both fiscal and m onetary policies, which can be shown in a simple reduced form model with the following equations: w here f is the fiscal policy instrum ent and m is the m onetary policy instrum ent.36 ( 4 ) y = r 0f + y i m 0< y,< y0 (5) n = dQ + #,m f 0 < 6 0< 6 V In the second model, which Andersen and Schneider re fe r to as Keynesian-New Classical, anticipated m onetary policy is neutral; it can af fect only inflation. Thus in a world of certainty, equation (4) becom es the following: (V y = Y ,f In the third model, the econom y is New Classical in nature, characterized by p erfect price flexibi lity and rational expectations. Anticipated policy, both fiscal and m onetary, affects only inflation, not output. T he econom y is modeled by the fol lowing equations: ( 7) n = 1 7 / + r ) ,m more, it is generally assumed that the inflation and output targets set by the government are greater than or equal to the targets set by the central bank. 35The quadratic nature of the loss functions, which is stan dard in the macroeconomic game theory literature, implies that deviations on either side of the targets produce an equal loss to the policymaker. 36The restrictions in equations (4) and (5) imply that fiscal policy has a greater (lesser) effect on output (inflation) than does monetary policy. JULY/AUGUST 1993 32 (8) y = 7i - tT (9) n - ne = r j j f - f e) + r ] ^ m - m e), w here y now refers to output relative to capaci ty and the superscript e refers to the expecta tion of the variable. Output can be increased above capacity only through unanticipated infla tion, and unanticipated inflation can occu r only through unanticipated changes in fiscal policy, m onetary policy or both. The relevant issue for policy is the size of the loss to each policym aker under cooperation and noncooperation. Cooperation in the determ ina tion of m onetary and fiscal policies is modeled by the governm ent and the central bank choos ing the policy variables (f and m) to minimize a weighted average of their loss functions: (10) min Lc = pLf + ( l - p ) L m 0<p<l = p\afi y - V j f + b fln - t t/I + (1 - p )[a m( y - y j + b j n - n j ] , w h ere the w eight placed on each loss function is determ ined by the relative bargaining strength of the two parties. Solving this minimization problem yields the equilibrium values for output and inflation, w hich can be substituted into the loss functions for the governm ent, equation (1), and the central bank, equation (2), to determ ine the loss to each. As noted above, noncooperation can be model ed in two ways. In the first, fiscal and m onetary policies are chosen simultaneously; that is, the governm ent selects a level of spending to mini mize its loss function, equation (1), taking as given the actions of the central bank. At the same time, the central bank chooses the grow th rate of the m onetary base to minimize its loss function, equation (2), taking as given the actions of the governm ent. This stru ctu re is referred to as a Nash game and the resulting equilibrium is called a Nash equilibrium. In a Nash equilibrium, neither authority, taking the actions of the other as given, can decrease its loss by unilaterally changing its policy. In the second model of noncooperation, one policy is set b efore the other is determ ined. This process is known as a Stackelberg game, and the policym aker who moves first is known 37See Andersen and Schneider (1986), p. 188. http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis as the Stackelberg leader, w hereas the other policymaker is known as the Stackelberg follow er. The leader chooses its policy, and the fol lower sets its policy in reaction. Furtherm ore, the leader, in choosing its policy, knows how the follower will react. Although the equilibrium level of output and the rate of inflation vary depending on which model of the econom y is used, in all three models the cooperative solution is Pareto superi or to the noncooperative solution. This result is invariant to the stru ctu re of noncooperation— Nash or Stackelberg. The perform ance of the econom y is b etter under cooperation in the sense that the losses to the governm ent and the central bank are each low er than they are un der noncooperation. This result holds even if the governm ent and the central bank each place the same weight on meeting their inflation tar gets relative to their output targets (af = a m and b f= b m but m aintain different targets. ) Andersen and Schneider sum marize these results by noting the following: When we have two independent authorities who act in their own selfish interest, then we quite often observe a conflict over the "right” policy direction. This result should be kept in mind when quite often the argument is put for ward that an independent monetary authority should be created. ... Two independent policy makers do not automatically guarantee a policy outcome which is preferred to other outcomes under different institutional solutions.37 Alesina and Tabellini (1987) show that adding one m ore target to the loss functions of the governm ent and the central bank also does not change the nature of the results. Noncooperation is once again suboptimal. Adding a time dimension to the model also does not change the basic result that coopera tion can improve the outcom e from the p er spective of both policym akers. Pindyck (1976) presents one of the first dynamic models analyzing the strategic interaction of m onetary and fiscal policy. He argues that the separation of monetary and fiscal control may considerably limit the ability of either authority to stabilize the economy, particularly when the conflict over objectives is at all significant.38 Petit (1989) exam ines the issue of policy coor dination in a continuous time model. The 38See Pindyck (1976), p. 239. 33 governm ent sets targets for output and inflation, giving higher priority to output. The central bank targets inflation and the level of in tern a tional reserves, giving higher priority to infla tion.39 As is standard, the governm ent sets the level of public expenditures to minimize its loss function, w hereas the central bank sets the grow th of the m onetary base to minimize its loss function. In this model, policies are set at the beginning and are unchanged over the period considered. Once again, cooperation is Pareto superior to the Nash and the Stackelberg equilibriums. F u rth er more, cooperation in this dynamic system leads to a decrease in the variability of the targets (particularly prices and international reserves), and raises the speed of adjustm ent of the sys tem. The latter indicates that, given a shock to the system, the econom y will retu rn more quickly to its long-run values of output and in flation if the governm ent and the central bank are coordinating their policies. Thus Petit con cludes that policym akers should coordinate their policies.40 Other studies con cen trate on the interaction of the governm ent and the central bank in financing fiscal deficits w here the deficit must be financed through bonds, seignorage or both.41 Under the assumption that th ere is some limit on the ability of a governm ent to continu ally issue bonds to finance its deficit, the need fo r inflation revenues becom es im portant.42 Sar gent and W allace (1981) conducted the seminal research on this question and showed that if the governm ent em barks on a path of unsus tainable deficits, the central bank might eventu ally be forced to inflate to fund the deficits. If the public realizes that the governm ent debt is on such a path, it will expect inflation to in crease, w hich may cause inflation to increase 39The target for international reserves reflects a balance of payments objective. 40Hughes Hallett and Petit (1990) also model the interaction of fiscal and monetary policy in a dynamic setting, reach ing this same conclusion. 4 Seignorage is the revenue received from the creation of 1 money. It occurs because base money costs only a fraction of its face value to produce. 42As the public debt grows, there may be increasing concern among bondholders that the government will be unable to repay the bonds. 43As Sargent and Wallace note, if money demand today de pends on inflationary expectations, then the price level to day is a function of not only the current money supply, but also expectations of the future levels of the money supply. well b efore the debt limit has been reached.43 This outcom e is a result of the governm ent b e ing able to set its policies and the central bank having to react to those policies (a Stackelberg game).44 In general, a conflict over the public debt can arise at any time w hen the governm ent and the central bank are allowed to adopt independent policies. Tabellini (1986) develops a dynamic model in w hich the central bank sets targets for changes in the m onetary base and the stock of outstanding public debt while the governm ent sets targets for the fiscal deficit net of interest payments and the stock of outstanding public debt. The target value of public debt is the same for both authorities. In choosing the level of the m onetary base and the fiscal deficits, the two authorities are constrained by the govern ment's dynamic budget constraint.45 The stock of public debt as a proportion of income is con sidered too high by both the fiscal and m one tary authorities. In the noncooperative setting, however, each authority ignores the benefit to the other of its own actions to reduce the level of debt. In the cooperative setting these benefits are internalized, resulting in a lower level of debt. Tabellini (1987) and Loewy (1988) provide two m ore examples of models examining the conflict betw een central banks and governm ents over fiscal policy. Both show that such a conflict can lead to an increase in governm ent debt. As not ed by Blackburn and Christensen (1989), a con flict will always arise betw een a central bank whose goal is to maintain price stability and a governm ent whose objective is to increase out put and is pursuing this goal by running a stream of large deficits. Such a m acroeconom ic program is infeasible; one party will have to re vise its strategy (give in). The conflict creates 44The concern that undisciplined fiscal policies could result in inflation was recognized by the EC in drafting the Treaty on European Monetary Union. In the regulations concern ing the proposed European Central Bank, the bank is pro hibited from financing fiscal deficits of the member countries. As pointed out by Sargent and Wallace, and expounded on by Darby (1984), the need for the central bank to mone tize government debt through an inflationary policy is based on the assumption that the rate of growth of the real economy is less than the real rate of interest. 45Note that monetary base and fiscal deficits in this model are both instruments and targets. JULY/AUGUST 1993 34 problem s for the econom y because of the un certainty over the future course of policy: the public can expect higher inflation or higher tax es, depending on w hich policym aker gives in.46 EVALUATION OF THE THEORETICAL LITERATURE The theoretical studies indicate that noncoor dination of fiscal and m onetary policies will result in a suboptimal econom ic perform ance from the perspective of both the governm ent and the central bank. Policy targets are m ore closely met w hen coordination occurs. Thus an independent central bank is not conducive to achieving b etter policy outcomes. However, the theoretical w ork, like the em pir ical studies, has its weaknesses. One criticism is that the models are too simplistic. Neither the p referen ce structures of the two authorities, nor the models of the economy, are completely specified. Furtherm ore, most of the models operate in a world o f certainty. Policy, however, is not made in a world of certainty. Extrinsic u ncertainty—shocks to the econom y—can drive a wedge betw een the im plem entation of policy and its outcom e. Intrinsic u ncertainty—lack of knowledge of the p referen ces of a policym aker— is incorporated only in Tabellini and Loew y’s models.47 As these two models illustrate, adding uncertainty can increase the policy conflict b e tw een an independent central bank and fiscal authority. In addition to assuming certainty, the models also omit one im portant player in these policy gam es—the public. Public perception of the credibility of a m acroeconom ic program is im portant to its results because the public can limit the ability of policym akers to take advan tage of an inflation/output tradeoff. If an in dependent central bank can increase the public perception of the credibility of policy, this in turn should produce b etter econom ic results.48 Another deficiency of this literature is its failure to address the feasibility of the policymak e rs’ goals. The output goals set by the govern 46A government may adopt a strategy of running deficits, through decreasing taxes, to force future governments to cut expenditures. Under this strategy, the government would prefer an independent central bank, which will re fuse to monetize the deficits and thereby increase the likelihood that fiscal spending will be reduced. See Sargent (1985) for a discussion of this type of strategy. FEDERAL RESERVE BANK OF ST. LOUIS m ent, fo r example, may not be sustainable w ithout accelerating inflation. Tax and expendi tures plans, which lead to a stream of deficits, may also raise questions about the sustainability of fiscal policy. In this environm ent, an indepen dent central bank could be useful if its credible com m itm ent to price stability forced the govern ment to evaluate the sustainability of its policy goals. In contrast, centralization of policies might reduce the long-run econom ic p erfo r m ance of a country w hen the governm ent’s fo cus is short-run perform ance. CENTRAL BANK INDEPENDENCE AND THE ECONOMY— W H AT DO WE KNOW? This paper began with two questions: W hy is the idea of an independent central bank as pop ular as it is? Are th ere econom ic benefits to be gained from having an independent central bank? Unfortunately, the em pirical and theoretical studies surveyed do not provide clear answ ers. The em pirical studies find that th ere is a nega tive correlation betw een central bank indepen dence and long-run average inflation. They also show a negative correlation betw een indepen dence and long-run average governm ent deficits as a percent of GDP. In general, they find no evidence of a positive correlation betw een out put grow th and central bank independence. These results all point in the same direction yet do not provide unequivocal evidence that an in dependent central bank will low er inflation and governm ent deficits and raise a cou ntry’s output. In sum, these empirical studies provide evi dence of a negative correlation betw een central bank independence and inflation and central bank independence and fiscal deficits, but they do not provide evidence of causality. Countries with an aversion to inflation may formalize this aversion through the creation of an independent central bank. If this is true, it is the inflation aversion, not the independence of the central bank, that is the prim ary causal factor behind the low inflation result. The empirical m easures them selves are biased tow ard the finding that 47See Tabellini (1987) and Loewy (1988). In Tabellini’s model the government is initially unaware of the preferences of the central bank. In Loewy’s model both parties are initially unaware of the preferences of the other. 48This issue has been studied in the literature that focuses only on monetary policy. See Blackburn and Christensen (1989) for a survey of this literature. 35 independence prom otes low inflation. This is b e cause the m easures place m uch weight on legal requirem ents that a central bank pursue price stability and place this goal above all others. Cukierman is explicit in stating that his m easure of independence: is not the independence to do anything that the central bank pleases. It is ra th er the ability of the bank to stick to the price stability objective even at the cost o f oth er short-term real ob jec tives.49 Given such a definition of independence, it is not surprising that independence is equated with low inflation. Theoretical studies indicate that an indepen dent central bank can increase policy conflicts with the governm ent w henever the preferen ces of the two differ and, in so doing, w orsen the econom ic perform ance of a country. These studies, how ever, do not provide overwhelming support for the idea that countries should place m onetary policy in the hands of the executive or legislative branches of governm ent. The simple stru ctu re of these models ignores some factors that affect the outcom e of policy decisions— for example, the role of the public and the overall credibility of policy. Central bank in dependence may enhance credibility and thus the overall effectiveness of a policy program. In sum then, in the em pirical studies, em pha sis on price stability and freedom to pursue this goal are prim ary determ inants of independence. In the theoretical studies independence is eq uated with noncooperation betw een the fiscal and m onetary authorities in policy im plem enta tion. These different definitions of indepen dence may partly explain the d ifferent results. Fu rtherm ore, countries that may be classified as independent using the em pirical definition may be classified as dependent using the theoretical definition. New Zealand is one such example. T he 1989 Reserve Bank of New Zealand Act made price stability the only goal of the central bank, and the central bank is free to adopt poli cies to achieve that goal. Thus according to the em pirical definition of independence, the 1989 act created an independent central bank in New Zealand. The central bank’s inflation target, however, is established by the governm ent fo r a multi-year period. The governor of the central bank signs an agreem ent pledging the bank to adopt policies to m eet this target. Such coopera tion betw een the m onetary and fiscal policy m akers is consistent with a dependent central bank in the theoretical models. Altogether these studies indicate that we are far from fully understanding the role of central bank independence in producing favorable eco nomic outcomes. REFERENCES Alesina, Alberto. “ Macroeconomics and Politics,” in Stanley Fischer, ed. NBER Macroeconomics Annual (MIT Press, 1988), pp. 13-52. , and Lawrence H. Summers. “ Central Bank In dependence and Macroeconomic Performance: Some Comparative Evidence,” Journal of Money, Credit and Bank ing (May 1993), pp. 151-62. _______ , and Guido Tabellini. “ Rules and Discretion with Noncoordinated Monetary and Fiscal Policies,” Economic Inquiry (October 1987), pp. 619-30. Andersen, Torben M., and Friedrich Schneider. “ Coordination of Fiscal and Monetary Policy Under Different Institutional Arrangements,” European Journal of Political Economy (February 1986), pp. 169-91. Aufricht, Hans. Central Banking Legislation (International Monetary Fund, 1961). Bade, Robert, and Michael Parkin. “ Central Bank Laws and Monetary Policy,” unpublished manuscript, University of Western Ontario, 1985. Berry, John M. “ More Independence Means Lower Inflation, Studies Show,” The Washington Post (February 17, 1993). Blackburn, Keith, and Michael Christensen. “ Monetary Policy and Policy Credibility: Theories and Evidence,” Journal of Economic Literature (March 1989), pp. 1-45. Buchanan, James M., and Richard E. Wagner. Democracy in Deficit (Academic Press, 1977). Capie, Forrest H., Terence C. Mills, and Geoffrey E. Wood. “ Central Bank Dependence and Inflation Performance: An Exploratory Data Analysis,” Centre for the Study of Monetary History, City University Business School, Discussion Paper No. 34 (March 1992). Committee of Governors of the Central Banks of the Member States of the European Economic Community. Annual Report 1992 (April 1993). Cukierman, Alex. Central Bank Strategy, Credibility, and In dependence: Theory and Evidence (MIT Press, 1992). _______, Steven B. Webb, and Bilin Neyapti. “ Measuring the Independence of Central Banks and Its Effect on Policy Outcomes,” The World Bank Economic Review (September 1992), pp. 353-98. _______ , Pantelis Kalaitzidakis, Lawrence H. Summers, and Steven B. Webb. “ Central Bank Independence, Growth, In vestment, and Real Rate,” Carnegie-Rochester Conference Series on Public Policy (autumn 1993). Darby, Michael R. “ Some Pleasant Monetarist Arithmetic,” Federal Reserve Bank of Minneapolis Quarterly Review (spring 1984), pp. 15-20. De Long, J. Bradford, and Lawrence H. Summers. “ Macroeconomic Policy and Long-Run Growth,” Federal Reserve Bank of Kansas City, Economic Review (fourth quarter 1992), pp. 5-30. 49See Cukierman (1992), p. 370. JULY/AUGUST 1993 36 Friedman, Milton. “ The Case for Flexible Exchange Rates,” in Milton Friedman, ed. Essays in Positive Economics (University of Chicago Press, 1953), pp. 157-203. Petit, Maria Luisa. “ Fiscal and Monetary Policy Co-ordination: A Differential Game Approach,” Journal of Applied Economet rics (April-June 1989), pp. 161-79. Grilli, Vittorio, Donato Masciandaro, and Guido Tabellini. “ Po litical and Monetary Institutions and Public Financial Poli cies in the Industrial Countries,” Economic Policy 13 (October 1991), pp. 341-92. Pindyck, Robert S. “ The Cost of Conflicting Objectives in Policy Formulation,” Annals of Economic and Social Meas urement (May 1976), pp. 239-48. Hughes Hallett, Andrew and Maria Luisa Petit. “ Cohabitation or Forced Marriage? A Study of the Costs of Failing to Coordinate Fiscal and Monetary Policies,” Weltwirtschaftliches Archiv (1990), pp. 662-89. Johnson, David R., and Pierre L. Siklos. “ Empirical Evidence on the Independence of Central Banks,” unpublished manuscript, Wilfrid Laurier University (March 1992). Johnson, Harry G. “ The Case for Flexible Exchange Rates, 1969,” this Review (June 1969), pp. 12-24. Loewy, Michael B. “ Reaganomics and Reputation Revisited,” Economic Inquiry (April 1988), pp. 253-63. Masciandaro, Donato and Guido Tabellini. “ Monetary Regimes and Fiscal Deficits: A Comparative Analysis,” in H. Cheng, ed. Monetary Policy in the Pacific Basin Countries (Kluwer Academic Publishers, 1988), pp. 125-52. McCallum, Bennett T. Monetary Economics: Theory and Policy (MacMillan, 1989). Parkin, Michael. “ Domestic Monetary Institutions and Deficits,” in J. M. Buchanan et al., eds. Deficits (Basil Blackwell, 1987), pp. 310-37. http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis Rogoff, Kenneth. “ The Optimal Degree of Commitment to an Intermediate Monetary Target,” Quarterly Journal of Eco nomics (November 1985), pp. 1169-90. Sargent, Thomas J. “ Reaganomics and Credibility,” in Alberto Ando et al, eds., Monetary Policy in Our Times (MIT Press, 1985), pp. 235-52. Sargent, Thomas J., and Neil Wallace. “ Some Unpleasant Monetarist Arithmetic,” Federal Reserve Bank of Minneapo lis, Quarterly Review (fall 1981), pp. 1-17. Summers, Robert, and Alan Heston. “ The Penn World Table (Mark 5): An Expanded Set of International Comparisons, 1950-88,” Quarterly Journal of Economics (May 1991), pp. 327-68. Tabellini, Guido. “ Money, Debt and Deficits in a Dynamic Game,” Journal of Economic Dynamics and Control (July 1986), pp. 427-42. . “ Central Bank Reputation and the Monetization of Deficits: The 1981 Italian Monetary Reform,” Economic In quiry (April 1987), pp. 185-200. 37 Michael J. D uek er Michael J. Dueker is an economist at the Federal Reserve Bank of St. Louis. Richard I. Jako provided research assistance. Hypothesis Testing with Near-Unit Roots: The Case o f Long-Run Purchasing-Power Parity ' I HE HYPOTHESIS THAT the purchasing pow er of a given currency, like the dollar, will be equal across countries has strong appeal. If the hypothesis is true, then inflation and ex change rate m ovements will be such that a given cu rren cy will, over time, lose equal am ounts of its purchasing pow er in all countries. The se quence of events by w hich deviations from purchasing-pow er parity would be eliminated can best be illustrated by example: If the dollar could purchase m ore goods in other countries than in the United States, then U.S. consum ers would purchase m ore goods from abroad, w hich would raise the demand for foreign cu r rencies relative to the dollar and lead to a depreciation of the dollar and eventual equaliza tion o f the dollar’s purchasing pow er across countries. Despite the intuitive appeal of such argum ents for long-run purchasing-pow er pari ty, statistical tests have been mixed. This paper argues that previous test results have conflicted because tests of purchasing-pow er parity have relatively low pow er under both the null hypoth esis that it holds, and the null that it fails. Hence this paper contains tests of both null hypotheses and shows that frequently neither is rejected for monthly data from five m ajor industrialized countries. This result serves as a caution against testing only one null hypothesis, finding that the null hypothesis cannot be rejected for a broad set of countries and concluding that there is robust evidence for o r against the theory of long-run purchasing-pow er parity. The theory of long-run purchasing-pow er par ity (PPP) implies that a currency's purchasing pow er is equal across countries in long-run equilibrium, but does not specify how long devi ations from this equilibrium can last. Large and persistent departures from PPP in the last 20 years, how ever, have cast doubt on the validity of PPP. As we will discuss later, there is a liter ature w hich tests w hether long-run PPP holds, that is, w h eth er departures from PPP are transi tory. This article aims to reconcile some of the disparate results from previous studies by using a long-memory model, w hich can do m ore than classify deviations from PPP as tem porary or perm anent: it can provide specific m easures of their persistence. Such m easures are useful b e cause large, persistent differences in a cu rren cy’s purchasing pow er across countries can greatly affect trade flows and the allocation of resources. Empirically, long-run PPP holds if the real ex change rate, w hich equals the nominal exchange rate multiplied by the ratio of the domestic and foreign price levels, is m ean-reverting. This arti cle will conform with the m ajority of the em pir ical PPP studies by using consum er price indexes JULY/AUGUST 1993 38 to calculate the real exchange rate. If price in dexes m easured the prices of identical baskets of goods across countries, absolute PPP would hold if P* = S x P , w here P is the domestic price of the goods basket, P* is the foreign price and S is the exchange rate in term s of units of foreign cu rren cy per unit of domestic currency. Because consum er price indexes do not m easure the cost of identical baskets of goods across countries, how ever, relative PPP modifies the relationship to account for the ra tio of the values of the tw o distinct baskets of goods: P* = kSP, w here k is the ratio of the value of the foreign basket to the domestic basket. The domestic country’s real exchange rate with the foreign country is then 1/k and equals SP/P*.1 The conventional approach to testing for longrun purchasing-pow er parity consists of testing for a unit root in the real exchange rate: Longru n PPP holds if the real exchange rate is meanreverting but not if it has a unit root. Previous tests have shown little pow er to reject w hichever of the two null hypotheses is employed. Tests w hose null hypothesis is that the real exchange rate contains a unit root generally fail to reject, w hereas tests w hose null is that long-run PPP holds also often fail to reject. These disparate findings are reconciled, however, if th ere is long m em ory in the real exchange rate, which enables both acceptance and rejection of longrun PPP at conventional significance levels.2 This paper employs long-memory models to obtain estim ates of the orders of integration of real exchange rates on a continuous scale. The advantage of estim ating the order of integration on a continuous scale is that we can confirm that long-memory time series behavior in real exchange rates is a possible source of the dis crepancies betw een previous tests of long-run 'Summers and Heston (1991) tabulate the costs of nearlyidentical baskets of goods across countries, rather than use existing price indexes. They define the PPP nominal exchange rate to be P*/P and use this implied exchange rate, rather than the market exchange rate, to make cross country comparisons. The Summers and Heston measures of the price levels could take the place of commonly used consumer price indexes when testing long-run PPP, as could wholesale price indexes. The Summers and Heston data, however, are only available on an annual basis and include data extrapolated between five year data collection periods. The analysis in this paper will be limited to the use of consumer price indexes to facilitate comparison with previous studies. 2Long memory, as will be discussed later, means that the order of integration of a time series process is greater FEDERAL RESERVE BANK OF ST. LOUIS PPP. The finding of long m em ory in real ex change rates also allows us to judge w hether the real exchange rate reverts to its m ean within an econom ically meaningful tim e fram e. W HY PURCHASING-POWER PAR ITY MIGHT NOT HOLD Before discussing statistical tests of PPP, it is w orth repeating Engel’s (1992) list of possible reasons fo r the em pirical failure of PPP: 1. Barriers to trade such as tariffs and tran s portation costs. 2. D ifferent consumption preferen ces across countries. 3. The presence of non-traded goods in price indexes. 4. Prices which are sticky in term s of the cu r rency in w hich the good is consumed. B arriers to trade, such as tariffs, are an obvious reason why the same goods do not sell at the same price throughout the world. D ifferent con sumption p referen ces, on the oth er hand, would lead consum ers in each country to choose different baskets of goods. Because price indexes are constructed for baskets of goods designed to rep resent a particular cou ntry’s consumption, an apparent failure of PPP could be due to different rates of price inflation across two country’s distinctive baskets of consumption goods, rath er than different prices fo r the same goods across countries. W hen included in price indexes, non-traded goods can also muddle the interpretation of the real exchange rate, because non-traded goods can be idiosyncratic and are thus not directly com parable across countries. Nevertheless consum er price indexes will be used in this paper, despite the presence of nontradeables, because wholesale price indexes can fail to reflect the underlying rate of inflation ac curately.3 The fourth source of failure, sticky than zero. If the order of integration is greater than 0.5, the series is not covariance stationary and if the order of integration is greater than one, the series does not have a mean. Long memory is not the same as an autoregressive near-unit root, because a series with a near-unit au toregressive root is still integrated of order zero, and is not considered a long-memory process. 3For example the wholesale price index for Japan suggests that Japan has had deflation on average from 1980 to the present, whereas the GDP deflator and CPI show moderate inflation. 39 prices, can best be explained by an example: Japanese autos sold in Japan and also exported to the United States have sticky prices in yen w hen sold in Japan and sticky prices in dollars when sold in the United States. Any exchange rate fluctuations would cause the yen (or dollar) price of the same model of ca r to differ across the Pacific. Thus autos might contribute to the failure of PPP in the true sense: the same good being sold at d ifferent prices (net of taxes) across countries. PREVIOUS TESTS OF PURCHASING-POWER PARITY Tests of PPP in the literature can be classified according to many criteria. In this b rief review of a large literature, three features will receive attention: 20th century annual data vs. post-1973 monthly data; the use of consum er price index es versus wholesale price indexes in the calcula tion of the real exchange rate; and w hether or not price levels are assumed to be measured with e rro r.4 The aim of this review is to illus trate the lack of consensus that has em erged from studies of long-run PPP and identify which modeling choices might have influenced the out com es of those tests. Coughlin and Koedijk (1990) conduct unit-root tests on real exchange rates using post-1973 monthly data and consum er prices and find that the unit-root null hypothesis cannot be rejected. They also exam ine w h eth er the real exchange rates are cointegrated with factors thought to determ ine the real exchange rate. Cheung and Lai (1993b) use post-1973 monthly data on con sum er prices and allow for m easurem ent erro r in prices. They use a Johansen (1991) likelihood ratio test for cointegrating vectors, in which long-run PPP is the null hypothesis, and general ly fail to reject long-run PPP.5 Thus the studies o f Coughlin and Koedijk (1990) and Cheung and Lai (1993b) illustrate the im portance of the null hypothesis in testing long-run PPP. Edison and Fisher (1991) and Fisher and Park (1991) represent another pair of studies that differ in the null hypothesis employed and the general conclusions about long-run PPP. Cheung and 4For a thorough introduction to what the authors call the purchasing-power parity assumption, see Caves, Frankel and Jones (1990). Lai (1993a) study 20th century annual data and find evidence that the real exchange rate has long memory, but not a unit root for most countries studied. Pippenger (1993) uses w hole sale price indexes and finds evidence that PPP betw een Switzerland and various countries ap pears to hold in the long run. Overall a lack of consensus em erges from em pirical tests of long-run PPP. The choice of null hypothesis is one source of discrepancy, and it appears that results from a long-memory model can reconcile the apparently conflicting results o f several previous studies of long-run PPP, which differ prim arily in their choice of null hypothesis. Consequently, tests of each null hypothesis will be highlighted in the estim ation results below. TESTING PURCHASING-POWER PARITY Much research on w hether PPP holds in the long run consists of perform ing a unit-root test on the real exchange rate. It is well known, however, that unit-root tests, especially those having a unit root as the null hypothesis, like Dickey-Fuller, have little power against longmemory alternatives. Such unit-root testing con sists of classifying econom ic variables as either integrated of order zero [1(0)1 o r one [1(1)1.6 In contrast, we use a param etric long-memory model in w hich data series, like the real ex change rate, are modeled as integrated of order d, denoted 1(d), w here d does not have to be an integer. Any series that is integrated of order d < l will retu rn eventually to its mean (or its determ inistic trend), so shocks to the real ex change rate are not perm anent if the real ex change rate is integrated of order d < l . This paper also provides inform ation about the sources of PPP failure by examining the com ponents of the real exchange rate. T he ratio of the price levels may have a higher order of fractional integration than the nominal exchange rate, or vice versa. If r is the real exchange rate, s is the nominal exchange rate, p is the domestic price level, and p* is the foreign price level (all in natural logs), then r = s + (p-p’ ). If s is I(dl), (p-p *) is I(d2), then r will generally be in6lntegration of order one means that a variable’s first differ ences are stationary, whereas its levels are not. 5Provided that the hypothesis of zero cointegrating vectors has been rejected, the null hypothesis for subsequent hypothesis tests using the Johansen procedure is that there is at least one cointegrating vector. JULY/AUGUST 1993 40 tegrated of order m a x {d l,d 2 }. If r is 1(b) w here b < max {d1 ,d 2 }, then the nominal exchange rate and the price ratio (p-p*) are fractionally cointegrated, that is, they share the same stochastic trend to some exten t.7 T he real ex change rate does not have a unit root if b < l . W e can also exam ine the point estim ates of d l and d2 and see w hich com ponent appears to have the strongest trend. This com parison an sw ers critics of flexible exchange rates who ar gue that floating rates actually have caused the real exchange rate to be less stable than it would have been under fixed nominal exchange rates.8 If d l > d 2 , then shocks to the nominal exchange rate are m ore persistent than shocks to the relative price levels. The latter is som ewhat cu rious, because proponents of the switch to a flexible exchange rate regime envisioned flexible exchange rates as sources of real exchange rate stability in a world in w hich countries might have persistent d ifferences in inflation rates. Yet if shocks to the nominal exchange rate are m ore persistent than shocks to the relative price levels, then the nom inal exchange rate has persistence above w hat is potentially useful in reducing the variance of the real exchange rate. In the em pirical results that follow, the possibility of ex cess persistence in the nominal exchange rate will be examined. BACKGROUND ON LONGMEMORY MODELS For many time series, autoregressive movingaverage (ARMA) models serve as a parsimonious way to sum m arize the autocovariance stru ctu re of the data. One limitation of such models is that ARMA processes are integrated of order zero, and the autocovariances die o ff relatively quickly, even when a root in the autoregressive polyno mial is n ear one. Thus ARMA models can be called short-m em ory models, because a shock affects the level of the series fo r a relatively short time. Long-memory models, in contrast, are suitable for data that have slowly decaying coefficients 7For reasons outlined below, the restriction that the coeffi cients on s and (p-p*) equal one is relaxed, so that the order of integration of a general linear (cointegrating) com bination of s, p and p* is assumed to be the order of in tegration of the real exchange rate. The concept of cointegration has been generalized [Granger (1986)] to in clude cases in which series have stochastic trends that only partially offset each other. This is called fractional cointegration. Originally, cointegration meant that a partic FEDERAL RESERVE BANK OF ST. LOUIS in their moving-average representations. The fractional ARMA model can serve as a longm em ory model, yet it adds only one param eter to a standard ARMA model. To illustrate, we b e gin with a simple ARMA(1,1) applied to the first difference of a data series y, w here L is the lag operator, £ is a m ean-zero disturbance, p is the AR coefficient, and 0 is the MA coefficient: (1) (1 - pL )(l - L)y = ( 1 + d L k A fractional ARMA model is simply an ARMA model applied to fractionally differenced data: (2) (1 - pL )(l - L)dy = (1 + 0L)£ The fractional differencing operator is evaluated by taking a Taylor series expansion around L = 0:9 (3) (1 - L P = 1 - d L + d ( d - l ) L 2 2 d i d - l ) ( d - 2 ) L 3 + ... 3! Two characteristics of fractionally integrated data are w orth noting. First, a series that is in tegrated of order d (1(d)) with d < l reverts to its mean (or at least to its determ inistic trend). Second, if d < .5 , the series is covariance station ary. At first glance, it might seem counter-intui tive that a m ean-reverting series can fail to be covariance stationary. W ith long m em ory, how ever, the departures from the m ean can be sufficiently persistent that the variance of the series is infinite. Furtherm ore, two commonly assumed datagenerating processes fit within the subset of fractional integration: trend and difference stationarity. Fractional integration offers a bridge betw een the controversial assignment of a data series as either trend or difference stationary, so that questions about stationarity assumptions ular linear combination of two strongly trending series was 1 0 ). ( 8For example, Aliber (1993) notes that “ the U.S. dollar ap preciated from 1979 to 1985 even though the U.S. inflation rate was higher than the inflation rates in Germany and Japan.” 9The concept of fractional differencing was developed by Granger and Joyeux (1980) and Hosking (1981). 41 may be avoided. For example, if y is trend sta tionary, we might model y as the real exchange rates. The general form of the fractional ARMA model is (4) y, = M + £, f (8) A(L)(2 - L)d t = B(L)et y In differences, equation (4) looks like w here y is 1(d) and t is assumed to have zero mean, no serial correlation and variance a 2. A(L) is an autoregressive polynomial of order p and B(L) is a moving-average polynomial of order q: (5) (2 - L ) y t = fi + ( l- L ) £ , If y is difference stationary, then (6) ( 1 - L ) y t = /i + £( Now suppose that y is fractionally integrated of order d. The first differences of y are then equal to (7) (1 - L ) y t = ju + ( 1 - L Y X w here fl is proportional to )u. Clearly trend stationarity (d = 0) and d ifference stationarity ( d = l ) are bridged by fractional integration, which al lows for interm ediate cases. An intuitive way to understand why fractional integration is an in term ediate case betw een trend and difference stationarity is to interpret each shock in a difference-stationary process to be a perm anent shift away from any previous trend; a shock in a trend-stationary process is a short-lasting shift from the trend; a shock in a fractionally inte grated process is a long-lasting shift from the trend. This paper uses estim ates of the order of fractional integration to discrim inate, if possible, betw een long-lasting shifts from the mean real exchange rate and perm anent shifts in the real exchange rate (the unit-root case). (9) A(L) = l - p ^ L - p 2L 2- . . . - p L P (10) B(L) = 1 + 0,L + 0,L2 + ... + Q Li Estimation was carried out using the Fox and Taqqu (1986) frequency-dom ain estim ator of fractional ARMA models. The estim ator is based on an approxim ation to the likelihood. Dahlhaus (1988, 1989) has analyzed the Fox and Taqqu es tim ator and has shown that it shares the same asymptotic efficiency as exact maximumlikelihood estimation. Fu rth er details regarding the estim ator appear in the appendix. Before presenting estim ation results, we must discuss how the real exchange rate was calculat ed. Any m ism easurem ent of the price levels can lead to spurious changes in the mean of the real exchange rate and bias tests toward reje c tion of long-run PPP. To minimize the possibility of spurious rejections of long-run PPP, the real exchange rate was calculated by estimating a fractionally cointegrating relationship betw een the nominal exchange rate (s), the domestic price level (p) and the foreign price level (p*):11 ( 11) s ~ a 0- a p t + aj>*t + £ , ESTIMATES OF LONG MEMORY IN REAL EXCHANGE RATES The data used in this article consist of 234 monthly observations of the nominal exchange rates and consum er price indexes for the Unit ed States, Great Britain, Germany and Japan from June 1973 to November 1 9 9 2 .10 Thus six bilateral relationships will be examined. Fractional ARMA models are used to estim ate the orders of integration of the nominal ex change rates, the ratios of the price levels and 10Koedijk and Schotman (1989) find that the real exchange rates between 15 industrialized countries are fairly well spanned by the real exchange rates between the United States, Japan, Germany and Great Britain. The residuals from equation (11) w ere then treated as the real exchange rate for unit-root testing with the fractional ARMA model. Cheung and Lai (1993b) and Pippenger (1993) also esti m ate a general cointegrating relationship, rath er than impose a t = a , = 1. They both argue that, because of m easurem ent erro r in price indexes and unequal weights attached to the same good in different indexes, it is undesirable to impose unit coefficients on the price indexes w hen studying w hether the real exchange rate is m ean-reverting. The Phillips and Hansen (1990) "Cheung and Lai (1993a) discuss the asymptotic theory be hind estimating regressions where the residuals are frac tionally integrated. JULY/AUGUST 1993 42 Table 1 Estimated Orders of Integration_________________________ Real exchange rate (b) t-statistic for 1 -b Nominal exchange rate (d1) Price ratio (d2) United StatesUnited Kingdom 1.06 (.039) -1.54 1.12 (.031) 1.60 (.043) United StatesGermany .903 (.067) 1.45 1.17 (.028) 1.21 (.081) United StatesJapan 1.24 (.067) -3.58 1.22 (.070) 1.62 (.080) United KingdomGermany .872 (.047) 2.72 1.01 (.052) 1.57 (.079) United KingdomJapan 1.19 (.034) -5.59 1.15 (.039) 1.12 (.039) GermanyJapan 1.13 (.207) -.628 1.19 (.142) 1.50 (.091) Countries NOTE: Standard errors are in parentheses. method of estim ating cointegrating relationships was used for equation (11). This method ac counts for simultaneity in the determ ination of left- and right-hand side variables. Cheung and Lai (1993b) suggest using estim ation procedures that take into account interactions betw een leftand right-hand side variables. In fact, the Phillips-Hansen estim ates of a } and a 2 indicate that they should not be restricted to equal one. For example, for the nominal exchange rate b e tw een Britain and the United States, the estim ates of a 1 and a 2 are 1.45 and 1.22, respectively. The next point of focus is the null hypothesis to be tested: PPP holds as the null hypothesis if the hypothesis that b < 1, w here b is the order of integration of the real exchange rate, is not rejected; the alternative null hypothesis that PPP fails is not rejected if the null that b > l is not rejected. Using the fractional ARMA model, it is easy to test both null hypotheses and show how the results depend on the choice of the null. Table 1 contains the main results on the esti mated orders of integration of the relevant ser ies. Simple t-tests can be used to test for long-run PPP by dividing one minus the estim at ed order o f integration of the real exchange rate by its standard erro r. Doing this, we see that the null hypothesis that b < 1, w here b is 12Orders of integration greater than one for data in logs im ply that the growth rates of the series have long memory. http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis the order of integration of the real exchange rate, is rejected for only tw o of the six pairs: United States/Japan and Britain/Japan. These are the significantly negative t-statistics in table 1. Thus long-run PPP is not rejected as a null hypothesis in four of six cases. If we reverse the null, how ever, we can reject the null that b > 1 for only one pair: Britain/Germ any.12 This is the significantly positive t-statistic in table 1. The results for United States/Germany are b o r derline with a t-statistic of 1.45, but this is not significant at the usual 5 percent level of sig nificance in a one-tailed t-test, w here the critical value is 1.658. Overall, the orders of integration of real ex change rates are often close enough to one that neither null hypothesis is rejected. This explains some discrepancies betw een past tests of longrun PPP. Coughlin and Koedijk (1990) used Dickey-Fuller unit-root tests on real exchange rates and could not reject the null that long-run PPP fails to hold. Cheung and Lai (1993b), on the other hand, tested the null that long-run PPP holds and did not find many rejections of long-run PPP. The results from the long-memory model reconcile these findings. The estim ates from the long-memory models do more than give unit root tests, how ever, by 43 providing an estim ate of the order of integra tion of the real exchange rate on a continuous scale. W hichever null is used, one result is clear: Even if long-run PPP holds, it is very slow in developing. Assuming that the order of in tegration of the real exchange rate is 0.9, 73 percent of a shock is still present after 12 months; 68 percent after 24 months; 65 percent after 36 months; and 63 percent after 48 months. As a practical m atter, it seem s fair to conclude that PPP does not hold within a time horizon that is econom ically relevant. Uncovering this type of inform ation is the chief advantage of es timating the order of integration on a continu ous scale. W ith other unit-root testing methods, we are forced to view a series as being either 1(0) or 1(1). Such a polar characterization may not provide practical inform ation about the p er sistence of the shocks. Table 1 also provides inform ation about the persistence of shocks to the nominal exchange rate relative to shocks to the ratio of the price levels. In five of six cases, point estim ates sug gest that the nominal exchange rate has a lower order of integration than the price ratio.13 For Britain/Japan the point estim ate of the order of integration is 1.15 fo r the nominal exchange rate vs. 1.12 for the price ratio, but this d iffer ence does not appear to be statistically signifi can t.14 Thus the conjecture that nominal exchange rates in the post-Bretton W oods era have shown excess persistence appears to be false. In general, g reater persistence in the nominal rate would be needed to offset the p er sistence in the price ratios for the real exchange rate to be rendered 1(0). This is because no 1(0) linear combination can exist, for example, b e tw een a series that is I(.8) and one that is I(.2). The series that is I(.2) does not have enough of a tren d with w hich to offset the relatively strong trend in the variable that is I(.8). A nother finding from table 1 is that inflation differentials are fractionally cointegrated in some cases. For example, the estim ates indicate that (p us- p JP ) is 1(1.62) and (p LIS- p U ) is 1(1.60), N K but the d ifference (pUK- p JPf) is only 1(1.12). This means that the inflation differentials betw een the United States and Britain and betw een the 13An order of integration above one for variables in logs means that the growth rates display long memory. In the case of the price ratio, the corresponding growth rate is the inflation differential across the two countries. For the nominal exchange rate, it is the rate of exchange rate ap preciation. United States and Japan appear m uch m ore p er sistent than the inflation differential betw een Britain and Japan. In other words, inflation rates in Britain and Japan com e closer to sh ar ing a comm on trend with each other than with inflation in the United States. Tables 2 through 4 rep ort the fractional ABMA param eter estim ates fully only for the bilateral relationships for the United States for the sake of brevity. The key result in these ta bles is that in fractional ABMA models the fra c tional differencing param eter can capture the long-run behavior of the data, freeing AR param eters to match the short-run dynamics. If, on the other hand, an ARMA model instead of a fractional ABMA model w ere fit to the data, the autoregressive polynomial would be forced to have a near-unit root. In table 2 several AR param eters are negative, and the largest equals 0.53 in the fractional ABMA model of the ratio of the price levels b e tw een the United States and Britain. In table 3 both estim ated AB param eters for the nominal exchange rate betw een the United States and Germany are negative, implying that all positive dependence in the exchange rate beyond the first lag is due to the large positive value of the fractional-differencing param eter. The largest root in an AR polynomial in table 3 is 0.37, w hich is far from the unit circle, in the real ex change rate betw een the United States and Ger many. Estimates of the model of the real exchange rate betw een the United States and Japan, found in table 4, also show two negative AR coefficients. In fact all of the AR polynomi als in table 4 have roots with real parts that are very far from the unit circle. They are - .0 5 , - .0 3 , and .08, respectively, for the real ex change rate, the nominal exchange rate and the price ratio. W ith the inclusion of the fractional differencing param eter, the AR param eters can take values which allow the fractional ARMA model to capture both long-run dependence and short-run dynamics in the data. The shaded in sert and figures 1 through 3 provide a visual check of the m atch betw een the covariance stru ctu re of the data and that implied by the es timated fractional ARMA model. 14A formal test would require joint estimates of the two frac tional ARMA models, however. JULY/AUGUST 1993 44 Table 2 Fractional ARMA Models: United States-United Kingdom Real Exchange Rate Parameter Order of integration AR MA b ^1 p2 0, Parameter value 1.06 -.2 2 8 .012 .534 Standard error .039 .096 .063 .084 Nominal Exchange Rate Order of integration AR MA di P\ P2 01 1.12 -.161 .002 .452 .031 .100 .045 .090 Ratio of Price Levels Order of integration AR *2 **1 MA 01 1.60 .529 -.0 8 9 -.809 .043 .051 .043 .027 Table 3 Fractional ARMA Models: United States-Germany Real Exchange Rate Parameter Order of integration AR MA b *> 1 "2 «1 Parameter value .903 .508 -.050 -.176 Standard error .067 .132 .060 .140 Nominal Exchange Rate Order of integration AR MA d1 Pi P2 01 1.17 -.3 3 2 -.017 .463 .028 .099 .039 .105 Ratio of Price Levels Order of integration AR MA http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis d2 Pi P2 01 1.21 .326 .044 -.1 2 7 .081 .162 .071 .134 45 Table 4 Fractional ARMA Models: United States^lapan Real Exchange Rate Parameter Order of integration AR MA Parameter value Standard error b 1.24 -.105 -.191 -.2 2 4 .067 .126 .066 .172 Pz Nominal Exchange Rate Order of integration AR MA p-l Pz 1.22 -.065 -.174 .226 .070 .113 .073 .150 Ratio of Price Levels Order of integration AR MA d2 Pa Pz •i 1.62 .159 -.2 8 8 -.749 .080 .108 .058 .019 A Visual Check of the Results O nce a fractional ARMA model has been es timated, a visual diagnostic check of the ade quacy of the specification and estim ates can be obtained by looking at a plot of the periodogram of the data alongside a plot of the spectral density implied by the specified fra c tional ARMA model. If the model fits the data well, the two graphs will have the same general turning points. The area under the spectral density equals the variance. Thus the height of the spectral density at any point in dicates how much of the variance is due to cycles of that frequency. In this way, the spectral density sum m arizes the autocovari ance structure. Figures 1 through 3 provide a look at the three data series fo r the United States/Germany: the real exchange rate, the nominal exchange rate and the price ratio. The figures show that the estim ated fractional ARMA models roughly envelope the smoothed 1ln theory the distinction is clear: A series with a nega tive order of fractional integration will have a spectral density value of zero at frequency zero; a series with a positive order of fractional integration will have a spec tral density value of infinity at frequency zero. For figures 2 and 3, it is clear from the periodogram that the series have positive orders of integration. This is not clear for the real exchange rate in figure 1. periodograms. Figure 1 shows why the out come of the unit root test on the U.S./ Germany real exchange rate is borderline: It is unclear w hether the upturn in the smoothed periodogram at very low frequencies is sig nificant or w hether the point estim ate of a negative order o f integration for the differ enced real exchange rate is c o rre ct.1 One other thing to note is that without ARMA param eters the model would be able to fit o n ly s e r ie s that h a v e g lo b a lly c o n c a v e (o r globally convex) spectral densities.2 W ith ARMA param eters, how ever, figure 1 illus trates that a fractional ARMA can generate turning points in the spectral density. The best-fitting globally concave spectral density would obviously provide a much less satisfac tory fit of the U.S./Germany real exchange rate than the one with turning points shown in figure 1. 2Without ARMA parameters, the fractional ARMA model is called the fractional noise model. Its spectral density is globally concave if the fractional differencing parameter from equation (2) is negative; it is globally convex if the fractional differencing parameter is positive. JULY/AUGUST 1993 46 Figure 1 Spectrum of Differenced Real Exchange Rate: U.S./Germany ARFIMA (2,d,1) Spectrum Smoothed Periodogram Frequency Figure 2 Spectrum of Differenced Nominal Exchange Rate: U.S./Germany ARFIMA (2,d,1) Spectrum http://fraser.stlouisfed.org/ FEDERAL St. Louis Federal Reserve Bank ofRESERVE BANK OF ST. LOUIS Smoothed Periodogram Frequency 47 Figure 3 Spectrum of Differenced Price Level Ratio: U.S./Germany ARFIMA (2,d,1) Spectrum Smoothed Periodogram Frequency CONCLUSIONS This article illustrates the key role played by the null hypothesis in testing fo r unit roots in real exchange rates. If b is the order of integra tion of the real exchange rate, then the null that f a > l is difficult to reject, in which case one would presum e that long-run purchasing-pow er parity does not hold. W hen the null is that b < 1, we also find few rejections, so long-run PPP apparently holds. W hen this type of am biguity appears, it is helpful to estim ate the ord er of integration on a continuous scale. The fractional ARMA models presented here do this and the standard erro rs on b for the six real exchange rates studied show that even if b < 1, it is not fa r enough away from one to make a strong case that purchasing-pow er parity is em pirically relevant. Cheung, Yin-Wong. “ Long Memory in Foreign Exchange Rates,” Journal of Business and Economic Statistics (Janu ary 1993), pp. 93-101. Cheung, Yin-Wong, and Kon S. Lai. “A Fractional Cointegra tion Analysis of Purchasing Power Parity," Journal of Busi ness and Economic Statistics (January 1993a), pp. 103-12. _______ “ Long-Run Purchasing Power Parity during the Re cent Float,” Journal of International Economics (February 1993b), pp. 181-92. Chowdhury, Abdur R., and Fabio Sdogati. “ Purchasing Power Parity in the Major EMS Countries: The Role of Price and Exchange Rate Adjustment,” Journal of Macroeconomics (Winter 1993), pp. 25-45. Coughlin, Cletus C., and Kees Koedijk. “ What Do We Know About the Long-Run Real Exchange Rate?” this Review (January/February 1990), pp. 36-48. Dahlhaus, Rainer. “ Small-Sample Effects in Time Series Analysis: a New Asymptotic Theory and a New Estimate,” Annals of Statistics (Volume 1b, 1988), pp. 808-41. _______ “ Efficient Parameter Estimation for Self-Similar Processes,” Annals of Statistics (Volume 17, 1989), pp. 1749-66. REFERENCES Dickey, David A., and Wayne A. Fuller. “ Likelihood Ratio Statistics for Autoregressive Time Series with a Unit Root,” Econometrica (July 1981), pp. 1057-72. Aliber, Robert Z. “ The Case for Flexible Exchange Rates Revisited,” unpublished manuscript, University of Chicago (April 1993). Diebold, Francis X., and Glenn D. Rudebusch. “ Long Memory and Persistence in Aggregate Output,” Journal of Monetary Economics (September 1989), pp. 189-209. Caves, Richard E., Jeffrey A. Frankel and Ronald W. Jones. World Trade and Payments: An Introduction (Scott, Foresman, 1990). Edison, Hali J., and Eric O'N. Fisher. “A Long-Run View of the European Monetary System,” Journal of International Money and Finance (March 1991), pp. 53-70. JULY/AUGUST 1993 48 Engel, Charles. “ Real Exchange Rates and Relative Prices: An Empirical Investigation,” NBER Working Paper No. 4231 (December 1992). Johansen, Soren. “ Estimation and Hypothesis Testing of Cointegrating Vectors in Gaussian Vector Autoregressive Models,” Econometrica (November 1991), pp. 1551-80. Engel, Charles, and James D. Hamilton. “ Long Swings in the Dollar: Are They in the Data and Do Markets Know It?” American Economic Review (September 1990), pp. 689-713. Koedijk, Kees, and Peter Schotman. “ Dominant Real Ex change Rate Movements,” Journal of International Money and Finance,” (December 1989), pp. 517-31. Fisher, Eric O’N., and Joon Y. Park. “ Testing Purchasing Power Parity Under the Null Hypothesis of Co-Integration,” Economic Journal (November 1991), pp. 1476-84. McNown, Robert, and Myles S. Wallace. “ National Price Lev els, Purchasing Power Parity, and Cointegration: A Test of Four High Inflation Economies,” Journal of International Money and Finance (December 1989), pp. 533-45. Fox, Rort, and Murad S. Taqqu. “ Large Sample Properties of Parameter Estimates for Strongly Dependent Stationary Gaussian Time Series,” Annals of Statistics (Volume 14, 1986), pp. 517-32. Phillips, Peter C.B., and Bruce E. Hansen. “ Statistical Infer ence in Instrumental Variables Regression with 1(1) Processes,” Review of Economic Studies (January 1990), pp. 99-125. Granger, C.W.J. “ Developments in the Study of Cointegrated Economic Variables,” Oxford Bulletin of Economics and Statistics (August 1986), pp. 213-28. Pippenger, Michael K. “ Cointegration Tests of Purchasing Power Parity: The Case of Swiss Exchange Rates,” Journal of International Money and Finance (February 1993), pp. 46-61. Granger, C.W.J., and Roselyne Joyeux. “An Introduction to Long-Memory Models and Fractional Differencing,” Journal of Time Series Analysis (Volume 1, 1980), pp. 15-29. Sowell, Fallaw. “ Maximum Likelihood Estimation of Stationary Univariate Fractionally Integrated Time Series Models,” Journal of Econometrics (July-September 1992a), pp. 165-88. Hakkio, Craig S. “ Does the Exchange Rate Follow a Random Walk? A Monte Carlo Study of Four Tests for a Random Walk,” Journal of International Money and Finance (June 1986), pp. 221-29. _______ “ Modeling Long-Run Behavior with the Fractional ARIMA Model,” Journal of Monetary Economics (April 1992b), pp. 277-302. Hosking, J.R.M. “ Fractional Differencing,” Biometrika (Volume 1b, 1981), pp. 165-76. Summers, Robert, and Alan Heston. “ The Penn World Table (Mark 5): An Expanded Set of International Comparisons, 1950-1988,” Quarterly Journal of Economics (May 1991), pp. 327-68. Hsieh, David A. “ The Determination of the Real Exchange Rate: the Productivity Approach,” Journal of International Economics (May 1982), pp. 355-62. Taylor, Mark P “An Empirical Examination of Long-Run Pur . chasing Power Parity Using Cointegration Techniques,” Ap plied Economics (October 1988), pp. 1369-81. Huizinga, John. “An Empirical Investigation of the Long-Run Behavior of Real Exchange Rates,” Carnegie-Rochester Conference on Public Policy (Autumn 1987), pp. 149-214. Whitt, Joseph A. Jr. “ Nominal Exchange Rates and Unit Roots: A Reconsideration,” Journal of International Money and Finance (December 1992), pp. 539-51. Appendix: The Fox and Taqqu Estimator Dahlhaus (1988) discusses why the Fox and Taqqu (1986) frequency domain estim ator is an approxim ate maximum-likelihood estim ator, sharing the same optimality properties as exact maximum-likelihood estimation. The Fox and Taqqu estim ator is derived from the following minimization problem: v —i (12) min y /(A.) , (/n(cr/U ;0)) + ---------------] o 2/ ( Xk;6) w here /(Afc is the vector of periodogram o r ) dinates of the data and o 2 k) is the spectral f(\ http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis density function implied by the param eterized model. For the fractional ARMA model in equa tion (8), the spectral density equals (13) j U) = ^ (e" ^ (2 - e ‘x)~d (1 - e ~ ' x)~d, IA(e-*)|2 w here A and B are polynomials defined in equa tions (9) and (10). T he objective function is minimized with respect to 6 and o2- An intuitive description of the objective function is that one w ants to choose param eters that will m ake the spectral density function implied by the model look like the periodogram of the data. 49 Richard G. Anderson Richard G. Anderson is a research officer with the Federal Reserve Bank of St. Louis. Heather Deaton provided research assistance. The Effect o f Mortgage Refinancing on M oney Demand and the Monetary Aggregates . ONEY SERVES AS A medium of exchange fo r transactions involving financial instrum ents as well as real goods and services. Unfortunately, the total volume of transactions in the economy is not observable. As a result, econom ic analyses of money demand typically focus on the relation ship betw een the quantity of m oney demanded and the production of new goods and services, m easured by either gross domestic product or personal consumption expenditures. Because ag gregate volumes of financial and nonfinancial transactions likely move in parallel with the out put of new goods and services, the use of out put rath er than the volume of transactions may cost little in term s of understanding movements in the m onetary aggregates. In some periods, however, events occu r which rem ind us that this is not always the case. This article examines the effect of one such ongoing recen t event— the refinancing of residential m ortgages— on money dem and.1 'Other recent examples include the Tax Reform Act of 1986, which boosted household liquid deposits in late 1986 and early 1987, and the closure of large numbers of thrifts by the Resolution Trust Corporation. Recognizing that special factors can significantly distort growth of the monetary ag gregates, the Bach commission recommended that the Federal Reserve regularly undertake and publish studies of the effects of special factors; see Report of the Advisory Simple models of the demand for money as a medium of exchange often implicitly assume that the purchase or sale of a good or service is com pleted within a relatively b rief period. Unlike the transactions in these models, the refinancing of a residential m ortgage that has been securi tized in the secondary m arket initiates a sequence of transactions that may continue for four to six weeks, or m ore. During this time, the quantity of liquid deposits demanded increases. W hen the last transaction in the sequence is concluded, the quantity of deposits demanded falls back ceteru s p a rib u s to its earlier level. Mortgage refinancing is an im portant phenom e non in the United States because most homes are financed with long-term, fixed-rate amortized mortgages that contain a “put” option, allowing the borrow er to repay the outstanding principal amount of the loan at any tim e without penalty. Homeowners typically exercise that option when mortgage rates fall significantly (1-2 percentage Committee on Monetary Statistics (1976). The Bank of England regularly publishes such analyses; see Pepper (1992, 1993) and Topping and Bishop (1989). JULY/AUGUST 1993 50 points) below recen t previous levels by taking out a new m ortgage loan to repay the old. As shown in figure l , 2 extensive m ortgage r e financing has occu rred during two periods in the last decade, 1 9 8 6 -8 7 and 1 9 9 1 -9 3 . In the form er, an initial surge in refinancing during 1986 was interrupted by a pause, b efore fears of rising m arket rates launched a second round in 1987. In the latter, th ree waves of refinancing —of increasing m agnitude—m irrored the halting fall in long-term m arket interest rates. During 1992, fo r example, nearly one-fifth of all hom e ow ners refinanced their m ortgages.3 In 1993, the volume of refinancing activity will surpass 1992’s record pace. The next section of this article describes the changes in the grow th and volatility of liquid deposits and M l that have occurred during periods of extensive m ortgage refinancing. The article then exam ines the extent to w hich these changes may be related to increases in mortgage securitization. Finally, it explores w h eth er recen t fluctuations in the grow th of oth er checkable deposits (OCDs) since 1991 also may be related to m ortgage refinancing. during late 1991, the third q uarter of 1992 and the second quarter of 1993, liquid deposit grow th accelerates. As refinancings continue at the higher rate, deposit levels converge to the new desired level and deposit grow th slows. W hen refinancing activity subsides—as in mid-1992 and early 1993—liquid deposit grow th slows fu r th er and deposits may run off. Through its effect on liquid deposits, mortgage refinancing sharply increased the volatility of M l during both 1 9 8 6 -8 7 and 1 9 9 1 -9 3 , as shown in figure 3.4 At the same time, the volatility of the broader aggregate M2, shown in figure 4, apparently was only slightly affected. In large part, the low er sensitivity of M2 to mortgage refinancing reflects the m uch sm aller share of transaction deposits in M2 (about 20 percent) than in M l (about 70 percent). The small changes that do appear in the volatility of M2 closely resem ble changes in its non-M l com ponent.5 The increases in liquid deposits that have ac companied accelerations in mortgage refinan c ing since m id-1990 are shown in figure 2. The link betw een m ortgage refinancing and liquid deposit grow th is a stock adjustm ent process w herein the stock of liquid deposits responds to changes in the flow of refinancings. W hen the pace of mortgage refinancing increases, as it did The ability of increases in m ortgage refinancing to affect the level and volatility of liquid deposits and M l is in part due to the borrow ed reserves operating procedure used by the Federal Reserve to control the grow th of M2. During the last de cade, this operating procedure has largely evolved into one that closely stabilizes the fed er al funds rate about a level thought to be consis tent with the desired amount of discount window borrow ing and the grow th of M2. To maintain the desired levels of the federal funds rate and discount window borrow ing, transitory increases in the demand fo r reserves are auto matically accom m odated with increases in the supply of nonborrow ed reserv es.6 2ln the figure, the volume of refinancing activity is proxied by liquidations of mortgage-backed securities. This concept is explored further in this article. No such correlations between refinancing-related deposit inflows and nontransaction funding sources are apparent in the data, however. 3Nineteen percent of the homeowners interviewed in Fannie Mae’s 1993 national housing survey had last refinanced their mortgage between January 1992 and March 1993. An additional 3 percent had refinanced during 1991 and 1990. 6For an analysis of the borrowed reserves procedure and its relationship to federal funds rate targeting, see Thornton (1988). For a careful discussion of why and how reservesbased targeting procedures evolve into federal funds rate targets, see Meulendyke (1990). MORTGAGE REFINANCING AND MONEY DEMAND 4The coefficient of variation shown in the figure equals the ratio of the standard deviation to the mean of the series, each calculated from the most recent 12 months of data. The coefficient of variation indicates whether the variability of the data has increased or decreased over time relative to its average level. 5The volatility of M2 differs little from that of its non-M1 component. It is feasible that banks’ cash management practices might account for the insensitivity of M2 volatility. Increases in liquid deposits provide additional funds to banks. If bank cash managers respond by reducing their issuance of overnight repurchases (RPs), the change in the volatility of M2 might be considerably less than that of M1. FEDERAL RESERVE BANK OF ST. LOUIS 51 F ig u re 1 M ortgage Interest Rate and Refinancing Activity Percent Billions of dollars Monthly data, January 1984-September 1993 16 50 Contract rate on 30-year, fixed -40 -30 -2 0 -10 92 1993 Liquidations of federal-agency-guaranteed mortgage-backed securities. F ig u re 2 Refinancing Activity and Liquid Deposits Billions of dollars Monthly data, seasonally adjusted 50- Index of mortgage refinancing activity 40C h an g e in Liquid Deposits 3020- 10- I] [JU Liquid D e p o s its= D em and D e p o s its+ O th e r C h e c k a b le D e p o s its * S a v in g s and MMDA d e p o s its - 10- 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 i i i J ASON DJ FMAMJJASONDJ FMAMJJASONDJ FMAMJJ AS 1990 1991 1992 1993 Liquidations of federal-agency-guaranteed mortgage-backed securities. JULY/AUGUST 1993 52 Figure 3 Refinancing and the Volatility of M1 Billions of dollars Index Monthly data, seasonally adjusted Shaded areas are periods of heavy refinancing activity. Figure 4 Refinancing and the Volatility of M2 Billions of dollars Monthly data, seasonally adjusted 92 Shaded areas are periods of heavy refinancing activity. http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis 1993 53 THE ROLE OF MORTGAGE SECURITIZATION The increase in mortgage securitization during the last decade has increased the potential for mortgage refinancing to affect the grow th of the m onetary aggregates.7 The sale of m ortgages in the secondary m arket creates an additional finan cial instrum ent—the m ortgage-backed security, or MBS—and involves a num ber of additional firm s in the m ortgage process, including the originators of the mortgages, the assem bler of the m ortgage pool (who also issues the MBSs), the servicer of the mortgage pool (who collects monthly payments and disburses funds to inves tors) and, typically, at least one governm ent agency. The refinancing of securitized mortgages thus becom es a circuitous calling and refunding of relatively large amounts of long-term, publicly held debt. Elevated levels of liquid deposits may persist for four to six weeks or m ore, until all related transactions are settled. Legally, mortgage securitization entails com bining a fixed pool of m ortgages into a trust. The m ortgages serve as collateral for MBSs sold against the trust. The servicer of the MBSs, as a trustee, collects payments from hom eow ners and passes them through w ithout taxation to the holders of the MBSs. Liquidity of the MBSs is enhanced by obtaining a third-party guaran tee covering the payments that will be due to investors if hom eow ners pay at the scheduled, minimum contract rate. T h ree federal-government-sponsored enterprises, known as "agen cies,” dominate that business.8 For a fee, these agencies guarantee the paym ent of principal and interest on securities backed by pools of specified mortgages. The Governm ent National Mortgage Association (Ginnie Mae, or GNMA), 7See Duca (1990) for an analysis of the interactions between demand deposits and mortgage refinancing during 1986-87. 8 small amount of MBSs is issued without agency guaran A tees. Bank of America issued the first such private mort gage pool in 1977. In 1992, private mortgage pools represented only 8 percent of all outstanding pools. For background, see Downs (1985) and Pavel (1986). 9The precise nature of the guarantee varies somewhat by agency. GNMA and FNMA guarantee timely (within the month) payment of principal and interest, regardless of payments by the borrower. FHLMC guarantees timely pay ment of interest and eventual (within the year) payment of principal. In addition to issuing guarantees on MBSs backed by privately assembled mortgage pools, FNMA and FHLMC may purchase mortgages outright and market MBSs backed by pools of those mortgages. In 1992, for example, FNMA “ issued” (guaranteed) $194 billion in MBSs. Of that amount, about $13 billion were originated by FNMA itself; a part of the D epartm ent of Housing and Urban Development, guarantees payments on MBSs backed by pools of Federal Housing Administra tion (FHA) and Veterans Administration (VA) mortgages. The Federal National Mortgage Association (Fannie Mae, or FNMA), a federally chartered, privately owned stock corporation, and the Federal Home Loan Mortgage Corpora tion (Freddie Mac, or FHLMC), a wholly owned subsidiary of the federally chartered Federal Home Loan Bank System, guarantee payments on MBSs backed by pools of conventional m ortgages.9 Absent refinancings or home sales, MBS in vestors receive a monthly payment that includes the scheduled am ortization of the pool’s m ort gage principal plus the accum ulated interest. Refinancings, home sales and an occasional ex tra payment by a hom eow ner retu rn additional (or unscheduled) principal pro rata to the holders of the MBSs backed by that mortgage pool. The monthly liquidation fo r a mortgage pool is the sum of the scheduled and unscheduled principal payments retu rned to investors. Note that MBSs aren ’t "called" in the traditional sense associated with corporate bonds, but rath er are only proportionately liquidated or repaid. As shown in the upper panel of figure 5, the outstanding stock of MBSs increased about six fold during the last decade, m uch m ore rapidly than M l o r M2. W ith few changes in mortgage servicing rules and practices during the last decade, the rapid grow th of securitization sug gests that the transactions incurred in refinancing securitized m ortgages will have larger effects on the m onetary aggregates in the 1990s than they did in the m id-1980s. Annual liquidations of MBSs, shown in the low er panel of the figure, the balance was originated by private lenders under a FNMA guarantee plan. FNMA’s 1992 Annual Report em phasizes the off-balance-sheet contingent risk nature of these securities: “ MBS are not assets of the corporation [FNMA], except when acquired for investment purposes, nor are the related outstanding securities recorded as lia bilities. However, the corporation is liable under its guaran tee to make timely payment of principal and interest to investors. The issuance of MBSs creates guaranty fee in come with Fannie Mae assuming credit risk, but without assuming any debt refinancing risk on the underlying pooled mortgages.” In 1992, FNMA recorded $834 million in guaranty fees. JULY/AUGUST 1993 54 Figure 5 M ortgage-Backed Securities O utstanding atYear-E nd Billions of dollars 1400—-----------------------------------------------------------------------------------------------------------------------------------1 ----------------------Data through September 1993 r— ■ 1200- I■ GNMA GNMA 1000- ■ ■ FHLMC FHLMC 800- □ □ FNMA FNMA |-| 600- _ I H 400 200 -------------------------------------- 0 1972 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 1993 End-of-year level Annual Liquidations of M ortgage-Backed Securities Billions of dollars 350 D ata th ro u g h S e p te m b e r 1993 300 ■ GNMA 250 m FHLMC □ FNMA 200 150 100 50 0 1972 73 74 75 76 77 78 79 80 81 FEDERAL RESERVE BANK OF ST. LOUIS 82 83 84 85 86 87 88 89 90 91 92 1993 Annual total 55 have on balance increased in proportion to the outstanding stock except for significant surges during periods of refinancing. Annual liquida tions jumped to about 17 percent of the out standing stock of MBSs during 1 9 8 6 -8 7 and 19 percent during 1 9 9 1 -9 2 . More recently, liquida tions during Ju n e through Septem ber 1993 averaged nearly $44 billion a month, almost a 40 percent annual rate. Recent fu rth er decreases in mortgage rates portend continuing high liqui dation rates during late 1993 and early 19 9 4 .10 The increase in deposits that follows an in crease in mortgage refinancing activity may in part be traced to the m echanics of mortgage securitization and servicing. Mortgage servicers’ handling of the unscheduled principal payments associated with refinancings is governed by the rules of the federal agency that guarantees the MBSs issued against the m ortgage pool. In gener al, these rules require that m ortgage servicers hold unscheduled principal payments in special custodial accounts during the interval betw een receipt from hom eow ners and disbursem ent to MBS investors. GNMA requires that these cus todial accounts be non-interest-bearing demand deposits. FNMA allows funds to be held in interest-bearing accounts as long as they are im mediately available without prior notice of w ith drawal. FHLMC’s rules are similar to FNMA's. A surge in refinancing greatly increases the monthly average amount of funds held in liquid deposits by a m ortgage servicer. In a typical month without refinancing, a servicer holds a hom eow ner's mortgage paym ent for a relatively b rief period of time (up to 15 days) before rem ittance to investors. Following a mortgage refinancing, how ever, the servicer will hold the unpaid principal balance of the extinguished mortgage loan—an amount perhaps 10 to 100 (or more) times as large as the hom eow ner’s regular monthly principal paym ent—in a cus todial account for a m uch longer period, often 10While it is always risky to forecast financial market activity, recent decreases in mortgage rates (through October 1993) are likely to trigger substantial further increases in re financing and MBS activity during late 1993 and early 1994. In addition to older mortgages issued during the 1980s, mortgages that were issued as little as 12 to 18 months ago at 7 to 7-1/2 percent rates now may profitably be refinanced. Rather than the pace of refinancing slowing and related distortions to the monetary aggregates diminish ing as the outstanding stock of seasoned MBSs are rolled over, recent rate decreases have placed nearly the entire outstanding stock of MBSs “ in the money” for rollover. "Homeowners typically make monthly mortgage payments between the 1st and 15th of the month, with the servicer remitting these funds to MBS investors on the 15th. Follow two to six weeks (see the shaded insert).11 Estimates of the size of this effect on monthly grow th rates of demand deposits, M l and M2, are shown in figure 6 .12 W hen MBS liquidations accelerate, the grow th rates of demand deposits and M l after rem oving the MBS effect are smaller than the published grow th rates. Con versely, w hen MBS liquidations slow, the MBSadjusted grow th rates are larger than the pub lished rates. Overall, the estimated differences in grow th rates equal in some months as much as one-half of the change in M l. From D ecem ber 1991 to M arch 1992, for example, inflows to mortgage servicers’ custodial accounts are esti mated to have added betw een 5 to 10 percentage points to the monthly grow th rates of demand deposits. The largest estimated effects w ere in O ctober 1992 and May 1993, when MBS-related inflows likely accounted fo r four-fifths and three-fifths, respectively, o f demand deposit grow th. In both cases, deposit grow th slowed sharply in later m onths w hen deposit levels had increased enough to support the accelerated pace of mortgage activity. Subsequently, during the first q uarter of 1993, runoffs of servicers’ custodial balances likely depressed monthly average deposit grow th by as much as 10 p er centage points. These patterns show through to M l (see the cen ter panel of figure 6) but are muted. C urren cy and OCDs, w hich comprise two-thirds of M l, are unlikely to be affected by MBS activity.13 Nonetheless, the distortions to demand deposits are sufficient that monthly grow th rates of M l since mid-1992 appear to have been distorted by as m uch as 5 to 7 percentage points. Similar estim ates for M2 that include estim ated effects on money m arket demand account (MMDA) balances are shown in the bottom panel of the figure. Overall, fluctuations in mortgage servicers’ ing a refinancing, the funds received by the servicer from the homeowner (at any time within the month) are placed in a custodial account. These funds are remitted by the servicer to MBS investors after the middle of the following month. The exact date, however, depends on the contract specifications of the agency guarantee program under which the MBSs backed by the mortgage pool that con tained the extinguished mortgage were issued. See, for ex ample, Karcher (1989). 1Construction of these estimates is discussed in the ap pendix. 1 3The next section raises the possibility that OCD balances also might have been affected by refinancing since 1991, albeit not through MBS-related transactions. ■ IULY/AUGUST 1993 56 Extra Deposits: Where Do They Come From? Where Do They Go? T he im pact of surges in refinancing-related transactions on the demand for liquid de posits and the m onetary aggregates depends on the m onetary control operating procedure being used by the Federal Reserve. The refinancing of securitized m ortgages gen er ates tem porary increases in the demand for liquid deposits. Since these deposits are sub ject to non-zero reserve requirem ent ratios, the Federal Reserve accom m odates this de mand under its cu rren t m onetary control procedures by furnishing additional reserves as necessary to maintain the federal funds rate near the level expected to be consistent with desired longer-run grow th of M2. W hen final paym ents to MBS investors are com plet ed, both the quantity of liquid deposits de manded and banks’ required reserves fall. To again m aintain the desired level of the fed er al funds rate, the Federal Reserve w ithdraws the now-surplus reserves from the market. Consider, fo r example, a 10 percent m ort gage with an unpaid principal balance of $100,000. Assume that the hom eow ner '$877*15/30. The implied intramonthly pattern of fluctua tions in transaction deposits and reserves is character istic of liquid deposits. Aggregate transaction deposits tend to increase sharply at the beginning of each month custodial deposits likely account for about onehalf of the recen t increase in M l volatility. It is unlikely that these estim ates are too large, since they are based on legal restrictions imposed on m ortgage servicers by federal agencies and realistic but conservative assumptions regarding intra-m onth patterns of mortgage closings and deposit behavior. The estim ates may be biased downward, how ever, for a num ber of reasons. The most im portant perhaps is the omission of any in crease in deposits held by issuers of new MBSs. As some issuers draw on bank warehouse credit lines to fund the purchase of m ortgages to be assembled into new MBS pools, they may offset part of the bank charges for these lines via earnings credits based on their deposit lev FEDERAL RESERVE BANK OF ST. LOUIS deposits some funds (a paycheck, for exam ple) into a demand deposit account on the first day of a typical month. On the fifth day of the month, the hom eow ner mails a check for $877 to his m ortgage servicer. The serv icer receives the check on the 10th and places the funds in a demand deposit ac count. On the 15th, the funds are paid to MBS investors who immediately move the funds out of demand deposits and into new earning assets. The extent to w hich this transaction affects the average daily levels of demand deposits and M l depends on what type of asset/deposit was held by the firm prior to paying the employee and on what type of asset/deposit is held by the MBS in vestor after receipt of the funds. Assuming that the firm held the funds prior to the first of the month in an asset not included in M l, and assuming that investors move funds promptly into assets not included in M l on the 15th, this single transaction contributes about $439 to the average monthly level of demand deposits.1 and run off during the latter weeks of the month. The implied fluctuations in reserve demand are accommo dated by the Federal Reserve, perhaps through an RP by the Open Market Desk. els. Also omitted are any increases in liquid deposits that arise because of the significant volume of additional transactions used to pur chase and sell large quantities of mortgages and MBS. HOUSEHOLD DEPOSITS AND REFINANCING In addition to demand deposits, changes in OCDs since mid-1991 also have reflected the ebbs and flows in the pace of m ortgage refinancing (see the upper panel of figure 7). The apparent increase in the correlation of OCDs with demand deposits contrasts with its behavior b efore 1991 and during 1 9 8 6 -8 7 , the latter shown in the low er panel of figure 7. 57 To illustrate the magnitude of refinancingrelated payments, suppose that the hom e ow ner now refinances the m ortgage on the 25th day of the month, with the servicer receiving funds on the 30th and holding them (as a fiduciary) in a demand deposit custodial account until rem ittance to investors around the middle of the following month. The refinancing, w hen it closes on the 25th, c re ates $100,000 of demand deposits that didn’t previously exist, reflecting the new mortgage loan extended to the hom eow ner. If the transaction is subject to Regulation Z’s rightof-rescission provisions, the $100,000 deposit likely will be held by the settlem ent agent or new lender for the first three days following the mortgage closing. If the mortgage has been securitized via federal-agency-guaranteed MBSs, the funds subsequently will be rem itted to the servicer of the extinguished mortgage. If not, the funds will be paid to the current ow ner of the original mortgage. Since the outstanding MBSs backed by the old mortgage have not yet been extinguished, the new mortgage (and new deposits) rep resent a tem porary net increase in the amount of out standing credit in the economy. Both the new deposit and the MBSs backed by the old m ortgage continue to exist until about the middle of the following m onth.2 The mortgage refinancing of $100,000 con tributes $16,666 to the average level of de mand deposits during the cu rren t month and $50,000 to the average level of the following month, assuming that the servicer rem its funds to investors on the 15th and investors immediately tran sfer the funds from demand deposits.3 W hen investors do so, the ag gregate level of demand deposits drops and the Federal Reserve will drain reserves from the m arket if necessary to maintain discount window borrow ings and the federal funds rate near the desired levels. 2Note that financial market participants (and federal agencies) record an MBS as being liquidated on the last day of the month in which the refinancing occurred, even though investors will not receive the underlying funds until after the middle of the following month. 3 What asset might investors buy with the demand deposit? One possibility is new MBSs backed by the new mortgages. What happens to the demand deposits that they use to purchase these new MBSs? They van ish, in textbook multiple-expansion-of-deposits fashion, accompanied by the Fed’s withdrawal of reserves. Should some portion of the OCD fluctuations during 1 9 9 1 -9 3 be attributed to m ortgage re financing activity? If so, and if the impact of re financing on OCDs w ere similar to its effect on demand deposits, then their com bined effects could account for as much as three-quarters of M l's grow th during a num ber of m onths since 1991. to mortgage activity is necessarily less direct and more circum stantial than that for demand deposits. Tracing direct links betw een house hold deposits and econom ic activity is generally not possible, since the Federal Reserve collects deposit data from the issuers of deposits such as banks and thrifts rath er than from the ow ners of deposits, including households and firm s.14 The recen t parallel monthly m ovements in these two types of liquid accounts is compelling but puzzling. Any evidence linking these deposits W hy might a household increase its OCD balances following a m ortgage refinancing? One possibility could be the conversion of home eq- 14Although the Federal Reserve Board’s flow of funds ac counts present a fairly complete balance sheet for the household sector, few items are directly observed. Most en tries are calculated as residuals, inferred from the double entry nature of the accounts and from balance sheet data for firms and government. See Guide to the Flow of Funds Accounts, p. 120. JULY/AUGUST 1993 58 Figure 6 Published Growth Rates Less Rate Adjusted for MBS Activity Demand Deposits Percent Monthly data 3020- 10- - 20- fl c ____ 1 H 1 noDD _ _ I ^ H I ■ - _ -„ n [l_ w ^m 0- c I E C C 10- J AS ONDJF MAMJJASONDJF MAMJJASONDJFMAMJJAS 1990 1991 1992 1993 M1 Percent Monthly data '6 JASONDJ FMAMJ J A S O N D J 1990 FMAMJ J A S O N D J F M A M J 1991 1992 JAS 1993 M2 Percent Monthly data j l "B- l l m ¥ r NOTE: Includes MBS effects on demand deposits only. JASONDJFMAMJ 1990 JASONDJFMAMJ J ASONDJFMAMJ 1991 http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis 1992 1993 JAS 59 Figure 7 Average Monthly Change in Demand Deposits and OCDs, by Quarter Billions of dollars OCDs Demand Deposits a IV 1 i 1 M r m— r ~iv~ 1990 1991 IV 1992 III 1993 Average Monthly Change in Demand Deposits and OCDs, by Quarter Billions of dollars 1985 1986 1987 1988 JULY/AUGUST 1993 60 uity into cash at the time of refinancing. If operative, this factor should be m uch stronger during the 1990s than during 1 9 8 6 -8 7 , for two reasons. First, many households have been res tructuring their balance sheets, seeking to reduce the levels of debt (and debt service) that they took on during the 1980s. Home equity convert ed to cash at refinancing allows them to repay other outstanding debt and reduce monthly debt service. Second, households generally ex perienced large capital gains on houses during the 1980s. For many, capital gains in housing appeared largely as a windfall, accruing more rapidly than had been anticipated w hen the hom e was purchased and without any overt ef fort by the hom eow ner. As such, these in creases in wealth likely w ere not optimally deployed (from a portfolio standpoint) across all household asset categories. For other hom e ow ners who might have p referred to consume the increased w ealth rath er than save it, the capital gain appears as a type of forced saving in the form of home equity. W hile a home equi ty loan may increase the liquidity of home equi ty, it doesn’t perm it the household to consum e a windfall increase in home equity, since the loan must be repaid. Hence, there may be some pent-up demand by hom eow ners for redirecting part of their home equity tow ard balance sheet restru cturing (reducing other consum er debt), consum ption or perhaps redeploym ent into m ore liquid assets. Although no direct data on cash withdrawals at m ortgage refinancings are available, recent evidence is supportive. Fannie M ae’s 1993 na tional housing survey asked households w hether their prim ary motivation in refinancing was to shorten the m aturity of the loan (thereby build ing equity m ore quickly) or to reduce their monthly payments. W hile a shorter m aturity was the motive m ore frequently stated, in fact 15These provisions do not apply to home purchases, nor to refinancings with the same lender for an amount equal to or less than the unpaid principal balance. The Act exempts from right-of-rescission provisions “ residential mortgage transactions,” which are defined in the Act as extensions of credit to acquire a principal residence. In May 1987, at the request of mortgage market participants, refinancings with the same lender were exempted from Regulation Z. At the time, it was felt that this change likely would significantly reduce the number of refinancings subject to right-ofrescission provisions. 16On the eligibility of lawyers to hold a client’s funds in OCD deposits, see section 2-341 of the Fed’s Regulation D in Federal Reserve Regulatory Service (1993). Client funds also may be placed in MMDA deposits, although the rul ings contained in section 2-341 perhaps suggest a prefer http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis at refinancing m ore households tended to forego a shorter m aturity in favor of low er monthly payments, consistent with reducing the im por tance of home equity in their portfolios. (Unfor tunately, the survey did not ask about the withdrawal of home equity at refinancing.) Home equity lending at banks, shown in figure 8, also has been w eak since mid-1991, with reports suggesting that hom eow ners are indeed repaying outstanding home equity loans with cash w ithdraw n at the time of a mortgage refinancing. W hile the grow th in OCDs likely reflects changes in households' deposits, some profes sionals and small businesses also may account for a portion of the increase. Some real estate payment practices tend to increase the demand for OCDs w hen m ortgage activity increases. The 1969 Tru th in Lending Act, for example, im plem ented through the Federal Reserve’s Regu lation Z, requires a three-day, right-of-rescission period for any new credit transaction secured by the borrow er's principal residen ce.15 During this period, settlem ent agents typically hold funds in a liquid deposit, or perhaps in the form of cashier's and officers' checks. If the funds are held solely for the beneficial interest of the household, they may be placed in an OCD accou nt.16 Cashier's and officers’ checks issued by banks are included as demand deposits in M l, while such checks issued by thrifts typi cally are included in OCDs. This supportive yet largely circum stantial evi dence leaves a num ber of unansw ered questions. If a household extracts funds at refinancing to repay a home equity loan, how long will it keep the funds in a liquid deposit? And isn’t the amount of funds almost surely far smaller than the amounts held by m ortgage servicers, associ ated with MBS refunding activity? If so, can the ence to hold the funds as OCDs. OCDs have no restrictions on the number of third-party withdrawals per month. While both OCDs and MMDA deposits are included in M2, data on MMDAs have not been collected by the Federal Reserve System since September 1990. Banks and thrifts began reporting that month only a combined total for all savings and MMDA deposits. Hence, no separate analysis of MMDA deposits is shown in this article. 61 Figure 8 Refinancing Home Equity Lending Billions of dollars Billions of dollars Monthly data -60 -50 -40 -30 ■20 10 -10 1990 1991 increasingly parallel m ovem ents in OCDs reason ably be attributed to refinancing activity? On balance, while the sharp increase in the correla tion betw een the changes in OCDs and demand deposits since 1991 suggests an underlying rela tionship to mortgage refinancing, the magnitude of any effect on the m onetary aggregates rem ains uncertain and a convincing explanation elusive. 1992 1993 rules governing the custodial accounts of m ort gage servicers. The m echanism generating parallel high-frequency m ovements in OCDs, however, is far less clear. The coincidence of its timing with changes in refinancing activity and the onset of unusual weakness in home equity lending in 1992 suggest that it may be related to the ongoing restru cturing of household balance sheets during the 1990s. SUMMARY Any factors that increase the demand for tran s action deposits can distort the grow th of the m onetary aggregates over significant periods of time. Recent waves of mortgage refinancing ac tivity have caused significant fluctuations in li quid deposits and M l. Under cu rren t Federal Reserve operating procedures for controlling the grow th of M2, such transitory changes in the demand for liquid deposits, like those associated with mortgage refinancing, are automatically ac commodated through changes in bank reserves, leading to increased volatility of M l. A large portion of this increased volatility of demand deposits can be traced to fiduciary REFERENCES Board of Governors of the Federal Reserve System. Federal Reserve Regulatory Service, volume 1 (October 1993). _______ Guide to the Flow of Funds Accounts (Board of Governors of the Federal Reserve System, 1993). _______ “ Improving the Monetary Aggregates,” Report of the Advisory Committee on Monetary Statistics (1976). _______ “ Money Stock Revisions,” supplement to the Federal Reserve statistical release, H.6 (February 4, 1993). Downs, Anthony. The Revolution in Real Estate Finance (The Brookings Institution, 1985). Duca, John V. “ The Impact of Mortgage Activity on Recent Demand Deposit Growth,” Economics Letters (February 1990), pp. 157-61. JULY/AUGUST 1993 62 Federal National Mortgage Association. 1992 Annual Report (Federal National Mortgage Association, 1992). . Fannie Mae National Housing Survey 1993 (Federal National Mortgage Association, 1993). Karcher, Louis J. Processing Mortgage-Backed Securities (New York Institute of Finance, 1989). Pepper, Gordon. “ Controls and Distortions: Monetary Policy in the 1970s,” National Westminster Bank Quarterly Review (February 1992), pp. 58-72. _______ “ Monitoring the Money Supply and Distortions to Monetary Data,” National Westminster Bank Quarterly Review (February 1993), pp. 40-58. Meulendyke, Ann-Marie. “A Review of Federal Reserve Policy Targets and Operating Guides in Recent Decades,” in Inter mediate Targets and Indicators for Monetary Policy (Federal Reserve Bank of New York, 1990), pp. 452-73. Thornton, Daniel. “ The Borrowed-Reserves Operating Proce dure: Theory and Evidence,” this Review (January/February 1988), pp. 30-54. Pavel, Christine. “ Securitization,” Federal Reserve Bank of Chicago Economic Perspectives (July/August 1986), pp. 16-31. Topping, S.L., and S.L. Bishop. “ Breaks in Monetary Series,” Bank of England Discussion Paper, Technical Series No. 23 (February 1989). Appendix Estimates of Mortgage Servicers’ Custodial Account Balances This appendix employs methodology suggested by Duca (1990) to estim ate the im pact of r e financing on the am ount of liquid deposits held by m ortgage servicers.1 At refinancing, the out standing principal of an extinguished securitized mortgage is retu rned to the m ortgage servicer. Following rules established by the federal agency that guaranteed the MBSs issued against the pool containing the mortgage, servicers place incom ing unscheduled payments in custodial accounts (liquid deposits) until rem itted to the holders of the MBSs around the middle of the following month. Since the servicing rules of the three agencies differ, the overall increase in custodial deposits that follows an increase in refinancing depends on the agency composition of MBS liq uidations. D ifferences in this composition during 1 9 9 1 -9 3 relative to earlier periods have attenu ated the deposit im pact of recen t MBS liquida tions from what might have been expected. GNMA-guaranteed issues made up about onehalf of aggregate liquidations during 1 9 8 6 -8 7 , 1The model in this appendix differs from Duca’s in some respects, including assuming a more uniform rate of mort gage closings during each month and that funds remain in liquid deposits somewhat longer during the month follow ing the refinancing before they are withdrawn by investors. 2 The exact monthly scheduled amortization rate is a func tion of the outstanding balances, rates and terms on the mortgages in the pool. Such calculations require extensive databases well beyond the scope of this study. An alterna http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis for example, but only one-quarter in 1 9 9 1 -9 3 . The largest volume of liquidations during 1 9 9 1 -9 3 , on balance, has been FHLMC issues that have a smaller impact, dollar for dollar, on liquid deposits than liquidations of GNMA- or FNMA-guaranteed MBS. The increase in liquid deposits due to MBS liq uidations is estimated from a simple simulation. The param eters are: T he proportion of MBS liquidations during a m onth that result from scheduled am ortization of principal (n o r m __ liq). Separation o f sched uled from unscheduled paym ents m atters fo r reasons explained in the text. Estim ates in this article assum e that scheduled principal pay m ents equal 1 percent (at an annual rate) of the outstanding stock of M BSs.2 T he average nu m ber o f days, expressed as a pro portion of the m onth, th at unscheduled principal paym ents are held in custodial accounts during the m onth in w hich the refinancing occu rred (G N M A __ th is ___ month, F N M A _ _ th is ___ tive set of estimates that assumed scheduled monthly amortization equal to 2 percent of outstanding aggregate principal had a relatively large number of months wherein actual principal payments were less than estimated sched uled payments and, hence, was rejected as implausible. 63 month, FHLMC__th is ___month). Payments received by servicers early in the m onth have larger im pacts on month-average deposit levels than paym ents received late in the month. Herein it is assumed that m ortgages close at a uniform rate during the month. Under this as sumption, the weighted average holding period fo r payments received by GNMA and FNMA _ servicers is 0.50 m onths, i.e., GNMA__th is _ month = FNMA__t h is _ month = 0.50. For _ FHLMC servicers, who generally hold unsche duled principal paym ents fo r five days or less, an average holding period o f 0.15 m onths is assumed. T h e average num ber of days, expressed as a pro portion o f the m onth, th at funds due to MBS investors are held in liquid deposits during the m onth following the refinancing (GNMA__last __month, FNMA_ la s t___month). Under _ GNMA and FNMA servicing rules, unscheduled principal paym ents received by servicers during the preceding month are rem itted to investors on the 15th and 18th of the cu rren t month, re spectively. Funds may b e on deposit longer if investors do not w ithdraw them immediately. Values of 20 days and 23 days, corresponding to GNMA__la s t _ month = 0.67 and FNMA_ _ _ la s t __month = 0.77, are used in th e calcula tions below . T hese som ew hat longer periods w ere suggested by exam ination of daily deposit data reported to the Federal Reserve by several large banks. For FHLMC, this is set equal to zero. For FNMA servicers, the proportion of incom ing funds placed in MMDAs ra th er than demand deposits (MMDA__share). A value o f 0.25 is as sumed below .3 Funds in MMDAs are assumed to rem ain on deposit fo r the same num ber of days as funds placed in demand deposit accounts. Monthly liquidation o f GNMA-guaranteed MBSs equals, by definition, the am ount of GNMAguaranteed MBSs issued during the month minus the change in the amount of GNMA-guaranteed MBSs outstanding as of the end of each month: GNMA _ liq = GNMA _ iss - AGNMA _ stk. In turn, the am ount of unscheduled principal pay m ents received by GNMA servicers during a month is assumed to equal the liquidations of GNMA-guaranteed MBSs minus 1 percent of the amount of GNMA-guaranteed MBSs outstanding at the end of the previous month: G N M A __un = G N M A ___ liq - n o r m ___liq* G N M A __s t k ___lag. Liquidations and unscheduled principal pay ments for FNMA and FHLMC are calculated in the same m anner. The amount of demand deposits that are cus todial account balances due to GNMA mortgage servicers is calculated as: GNMA __dda = G N M A ___ t h i s __ month*GNM A __un + G N M A ___ l a s t ___ m on th *G N M A ___ un — la g. For FNMA servicers, the amount is: FNMA _ dda = (1 -MMDA _ share)*(FNM A 11]is__m on th * FNMA ___ un + F N M A ___ last __m on th * F N M A ___ u n ___lag); and for FHLMC it is: FHLMC _ FHLMC _ d d a = FHLMC _ un this _ m onth * A similar calculation is made fo r the holdings of MMDAs by FNMA servicers. An MBS-adjusted, not seasonally adjusted (n.s.a.) demand deposit series is obtained by subtracting the sum (GN M A__ dda + F N M A ___ dda + F H L M C __dda) from published n.s.a. monthly levels of demand deposits. The resulting demand deposit series is seasonally adjusted us ing the seasonal factors for demand deposits published by the Federal Reserve Board staff in M oney Stock R evisions (1993). (Seasonal factors are recovered from the published data by divid ing the n.s.a. level by the s.a. level.) The differ ences in grow th rates of demand deposits and M l shown in the upper two panels of figure 6 are calculated from published and these adjust ed data. An MBS-adjusted, non-M l com ponent of M2 is obtained by subtracting the estimated amount of MBS-related MMDA deposits from the pub lished, seasonally adjusted, non-M l com ponent of M2. (Since the non-M l com ponent of M2 is seasonally adjusted by the Federal Reserve Board staff as a whole, and separate data on MMDA are not available, the seasonally adjusted series was adjusted by MBS effects.) The grow th rates shown in the low er panel of figure 6 are calculated from published M2 and from the sum of the MBS-adjusted M l and non-M l com po nents of M2. 3The value of 0.25 is from Duca (1990). JULY/AUGUST 1993 F e d e ra l R e s e rv e B a n k o f St. L o u is Post Office Box 442 St. Louis, Missouri 63166 The R e v ie w is published six times per year by the Research and Public Information Department o f the Federal Reserve Rank o f St. Louis. Single-copy subscriptions are available to the public fr e e o f charge. Mail requests f o r subscriptions, back issues, or address changes to: Research and Public Information Department, Federal Reserve Bank o f St. Louis, P.O. Box 442, St. Louis, Missouri 63166. The views expressed are those o f the individual authors and do not necessarily reflect official positions o f the Federal Reserve Rank o f St. Louis or the Federal Reserve System. Articles herein may be reprinted provided the source is credited. Please provide the Bank’s Research and Public Information Department with a copy o f reprinted material.