The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
R Vol. 76, No. 6 W November/December 1994 Symposium on Mutual Funds and Monetary Aggregates An Alternative Monetary Aggregate: M2 Plus Household Holdings of Bond and Equity Mutual Funds The Empirical Properties of a Monetary Aggregate That Adds Bond and Stock Funds to M2 Commentary Replication and Scientific Standards in Applied Economics a Decade After the Journal of Money, Credit and Banking Project THE FEDERAL ^RESERVE Jtk BANK of A r ST. LOUIS R E V I E W President T h o m a s C. M e lze r Review is published six times per year by the Research Department of the Federal Reserve Bank of St. Louis. Single-copy subscriptions are available free of charge. Send requests for subscriptions, back issues or address Director of Research W illia m G. D e w a ld Associate Director of Research C le tu s C . C ou gh lin Research Coordinator and Review Editor W illia m T. G avin changes to: Federal Reserve Bank of St. Louis Public Information Department P.O. Box 442 St. Louis, Missouri 63166 The views expressed are those of the individual authors and do not neces sarily reflect official positions of the Federal Reserve Bank of St. Louis, the Federal Reserve System, or the Board of Governors. Articles may be reprinted Banking R. A lto n G ilb e rt D avid C. W h e e lo c k International C h ris to p h e r J . N e e ly M ic h a e l R. P a k k o P a tric ia S. P o llard or reproduced only if the source is credited. Please provide the Public Information Department with a copy of the reprinted materials. All nonproprietary and nonconfidential data and programs for the articles published in Review are available to our readers. This information can be Macroeconomics D o n ald S. A llen R ich ard G. A n d erson J a m e s B. B u llard M ic h a e l J . D u e k e r Jo se p h A. R itte r Jo h n A. T ato m D a n ie l L. T h o rn to n P e te r Yoo Regional M ic h e lle A. C la rk K e vin L. K lie s e n A d am M . Z a re ts k y obtained from three sources: 1. FR ED (F e d e ra l R e s e rv e E c o n o m ic D a ta ), an e le c tr o n ic b u l le tin bo ard s e rv ic e . You can access FRED by dialing 314-621-1824 through a modem-equipped personal computer. Parameters should be set to: no parity, word length = 8 bits, 1 stop bit, and the fastest baud rate the modem supports (up to 14,400 bps). Information will be in directory 11 under the file name ST. LOUIS REVIEW DATA. For a free brochure on FRED, please call 314-444-8809. 2 . T h e F e d e ra l R e s e rv e B a n k o f S t. L o uis. You can request data and programs on either disk or hard copy by writing to: Research Department, Federal Reserve Bank of St. Louis, Post Office Box 442, Director of Editorial Services D a n iel P. B ren nan Managing Editor C h a rle s B. H e n d erso n Graphic Designer Brian D. E b ert St. Louis, MO 63166. Please include the author, title, issue date and page numbers with your request. 3 . In te r-u n iv e rs ity C o n s o rtiu m fo r P o litic a l an d S o c ia l R e s e a rc h (IC P S R ). Member institutions can request data through the CDNet Order facility. Nonmembers should write to: ICPSR, Institute for Social Research, P.O. Box 1248, Ann Arbor, Michigan 48106, or call 313-763-5010. 1 Federal Reserve Bank of St. Louis Review November/December 1994 In This Issue. . . Symposium on Mutual Funds and Monetary Aggregates During 1990-93, the Federal Reserve’s primary monetary aggregate, M2, grew much more slowly than suggested by historical relationships to economic activity and opportunity costs. This sluggish growth led some to question its usefulness as a policy indicator. In July 1993, Federal Reserve Board Chairman Alan Greenspan told Congress that monetary aggregates, including M2, were being officially de-emphasized as guides to policy. During the same period, households’ holdings of bond and equity mutual fund shares increased at a record pace. Responding to a sharp decrease in the overall level of market interest rates and a record widening of maturity-related yield spreads, households shifted unprecedented amounts of savings away from traditional depository institutions. In March 1994, the Federal Reserve Bank of St. Louis hosted a symposium to examine the implications of the rapid growth of mutual funds for the role of M2 as an indicator of the stance of monetary policy. Speakers included staff from the Division of Monetary Affairs at the Board of Governors, executives from the banking and mutual fund industry, and academic economists. Among the questions addressed were: To what extent do households regard bank deposits and mutual fund shares as close substitutes? Should M2 be replaced by a new monetary aggregate that includes these mutual funds? Or, rather, does M 2’s apparently aberrant behavior only confirm that broad monetary aggregates are generally not suitable as guides for monetary policy? Editor’s Introduction Richard G. Anderson An Alternative Monetary Aggregate: M2 Plus Household Holdings of Bond and Equity Mutual Funds Sean Collins and Cheryl L. Edwards 31 The Empirical Properties of a Monetary Aggregate That Adds Bond and Stock Funds to M2 Athanasios Orphanides, Brian Reid, and David H. Small NOVEMBER/DECEMBER 1994 2 53 Commentary 53 William A. Barnett and Ge Zhou 63 Jacob S. Dreyer 67 John V. Duca 71 Josh Feinman 75 George G. Pennacchi 79 Replication and Scientific Standards in Applied Economics a Decade After the Journal of Money, Credit and Banking Project Richard G. Anderson and William G. Dewald Since early 1993, the Research Department of the Federal Reserve Bank of St. Louis has made data and programs for articles published in its R eview available to the public. The files, developed during a pre-publication repli cation that assures the accuracy of an article’s empirical results, allow the interested reader to delve into the details of the author’s research. The R eview is one of only a few journals in economics that provides access to the data used by its authors, and the only one to include the underlying programming. A decade ago, the Jou rn al o f M oney, Credit a n d B an kin g project was the first attempt by an economics journal to furnish readers with the data and programs used by its authors. The project affirmed that readers could not obtain data and programs directly from authors. Further, the project sug gested a positive correlation between the journal’s furnishing data to read ers and the care with which authors completed their research. Finally, the project demonstrated that professional journals are a low-cost way to dis tribute data to readers. The more recent experience at the Federal Reserve Bank of St. Louis reaffirms these conclusions. FEDERAL RESERVE BANK OF ST. LOUIS 3 Editor’s Introduction On March 29, 1994, the Federal Reserve Bank of St. Louis hosted a symposium on the im plica tions of rapid mutual fund growth for monetary policy. From 1990-93, household holdings of shares in bond and equity mutual funds increased at a record pace. During the same period, the Federal Reserve’s primary monetary aggregate, M2, grew much more slowly than suggested by its historical relationships to economic activity and opportunity costs. Does the confluence of these events suggest that M2 has become less useful as an indicator of the stance of monetary policy? Should it be replaced with a new aggre gate that includes these mutual funds? Financial innovation and advances in technology change the structure of financial markets, alter the indicator properties of monetary aggregates, and give rise to pressures for their redefinition. When Regulation Q capped deposit offering rates during the 1970s, for example, many households learned that money market mutual funds provided an attractive alternative to holding bank and thrift deposits. As a result, money market funds were included in M2 when it was redefined in 1980. During the 1980s, households became increasingly familiar with financial institutions other than banks and thrifts. Mortgage loans were increasingly originated by mortgage brokers, auto loans extended by finance compa nies, and retirement funds held in self-managed IRA and Keogh accounts. At the same time, the mutual fund industry benefited from technical progress that reduced the cost of servicing large customer lists and managing portfolios of mar ketable securities. Thus, participants on both sides of the financial markets seemed poised to react swiftly during the 1990s to the combination of a sharp decrease in the overall level of market interest rates and a record widening of maturity-related yield spreads. Available survey and anecdotal evidence, as well as negative statistical correlations in the aggregate data, suggest that households shifted savings away from traditional depository institutions toward bond and equity mutual funds. With offering rates on deposits decreasing, prospective returns on bond and equity funds often appeared to be three- or four-fold greater. The subsequent slow growth of M2 during a period when many analysts perceived monetary policy as becoming increasingly expansionary led to doubts about its usefulness as a policy indicator. The monetary aggregates were offi cially de-emphasized as policy guides in Federal Reserve Board Chairman Greenspan’s July 1993 Humphrey-Hawkins Act testimony before Congress. At about the same time, a group of economists at the Board of Governors completed two studies evaluating whether M2 might usefully be replaced by a redefined aggregate that included bond and equity mutual fund shares. These studies, and the five commentaries that appear here, were presented at the St. Louis symposium. Written by economists closely involved in policy analysis, the studies provide a unique perspective on the range of issues that arise whenever it is suggested that a monetary aggregate be redefined. Before a monetary aggregate may be used, it must be measured. In the first article, Sean Collins and Cheryl Edwards discuss the measurement of a monetary aggregate M2+ that includes both M2 and shares in bond and equity mutual funds. They first describe how interest in redefining an aggregate arises when its growth differs from that suggested by its historical behavior. Although necessary, such deviant behavior may not be sufficient unless there also has been significant innovation or technical progress in financial markets since the previous redefinition of the aggregates. The latter imparts an a priori reason ableness to suspicions that the array of money substitutes available to households and firms has expanded, or that the transaction costs of substituting among various alternatives has decreased. The current M2 monetary aggregate is designed to measure household and firm holdings of liquid assets that are either available for spending now NOVEMBER/DECEMBER 1994 4 or will become so in the near future. Retaining this focus in a new monetary aggregate that includes bond and equity mutual funds requires separating institutional holdings of mutual fund shares from those held by firms and households, as Collins and Edwards discuss in the latter half of their article. Further, data on several items that are not included in the new aggregate, such as the M2-type assets owned by the bond and equity funds and the amount of mutual fund shares held by households as illiquid retirement balances, are also required. These amounts are subtracted from the new aggregate. Collins and Edwards discuss several unique problems that arise while building M2+ that have no direct parallel in the current monetary aggregates. One is the inclusion of assets denominated in foreign currencies. All compo nents of the current official monetary aggregates (M l, M2, M3 and L) are denominated in U.S. dollars. Mutual funds that invest in foreign currency-denominated assets, however, have been among the most rapidly growing type of funds in recent years. Second, current M2 includes only assets that are capital-certain or, in other words, only assets whose value does not vary with the level of market interest rates. Capital gains and losses are a significant factor in changes in the value of bond and equity funds, and present a thorny problem for both the construction and interpretation of the M2+ aggregate. In the policy arena, monetary aggregates may be valuable as either targets or indicators, with requirements for the former generally more stringent than for the latter. In both cases, the aggregate must have a reliable empirical relationship to future economic activity. In addition, to be useful as a policy target, the demand for the aggregate must be a stable func tion of a relatively small number of variables and the Federal Reserve must be able to control the aggregate’s growth. In the second article, Athanasios Orphanides, Brian Reid and David Small judge the M2+ monetary aggregate relative to each of these criteria. Modeling the demand for a monetary aggre gate requires identifying its close substitutes and, in turn, the opportunity cost of holding the components of the aggregate rather than alterna tive assets. The inclusion of capital-uncertain assets in M2+ complicates calculation of its own rate of return and identification of an appropriate Digitized forFEDERAL FRASER RESERVE BANK OF ST. LOUIS opportunity cost. The authors conclude that the nonbank public’s holdings of bond and equity fund shares seem to respond to both the spread between the return on the mutual fund and the Treasury bill yield, and the change during the previous period in the overall level of market rates or equity prices. In the latter part of their article, Orphanides, Reid and Small examine the stability of the demand for M2+ and its value as a leading indi cator for nominal GDP. Working within the linear error-correction framework developed by Board staff in previous money demand studies, they find a reasonably good overall fit to the data. Yet, the estimated semi-elasticities of M2+ demand with respect to market yields and various measures of its opportunity cost appear highly sensitive to the form in which these variables enter the regression. Perhaps more disappointing, however, is their conclusion that M2+ has not been a generally better indicator than M2 of movements in nominal GDP growth during the 1990s. In their commentary, William Barnett and Ge Zhou interpret the definition of M2+ as a dynamic index number problem. When the financial assets in an economy can be partitioned into two non-overlapping groups—those that provide monetary (transaction) services and those that do not—the economy’s money stock is correctly measured by summing the quantities of the assets in the former group. It is the essence of financial innovation, however, to blur the distinction between these groups and allow assets in the latter category to provide monetary as well as non-monetary services. The authors show, however, that sequentially redefining a broad monetary aggregate may cause the aggregate to move further away from, rather than closer to, the economy’s true money stock. They laud the authors of the symposium’s two major papers for grappling with the difficult issue of including capital-uncertain assets in a monetary aggregate. Jacob Dreyer doubts, however, that a new M2+ monetary aggregate would be useful to policymakers. The necessity of estimating expected holding period yields likely precludes obtaining useful estimates of a demand equation for the aggregate. He also argues that turnover rates suggest that bond and equity mutual funds lack the necessary degree of “moneyness” for them to reasonably be included in a monetary aggregate along with the current components of M2. Although acknowledging that the transaction 5 costs of buying and selling bond and equity funds have fallen sharply, he regards this change as creating only the illusion, rather than the substance, of moneyness. In his commentary, John Duca attributes the slower-than-anticipated growth of M2 during the 1990s to a combination of changes in banking regulation and the pattern of market interest rates. Duca notes that significant shifts in financial intermediation also occurred in the 1970s and 1980s when government regulation interacted with large movements in market interest rates. Nevertheless, he concludes that the unprece dented steep slope of the yield curve during a period when offering rates on bank and thrift deposits fell to their lowest levels in decades likely was the primary motivation for households to substitute holdings of bond and equity funds for M2-type assets during the 1990s. This substitution does not, by itself, call for a redefinition of M2, however. Josh Feinman suggests that the dramatic shrinkage of the federal subsidy to the depository sector since 1990 seems to have played the larger role in depressing M2 growth. He cites higher deposit insurance premiums, new capital stan dards and stricter supervision as examples. These changes reduced the incentive for deposi tories to pursue their traditional lending and deposit-taking activities, resulting in an increasing proportion of intermediation being conducted through market instruments such as corporate equity and commercial paper, popular investments of bond and equity mutual funds. He doubts that M2+ could ever be useful as a policy target or indicator. In his view, the inverse correlation between movements in market interest rates and the value of the capital-uncertain assets included in M2+ likely precludes formulating any policy feedback rules based on M2+. George Pennacchi suggests that households’ increased holdings of mutual fund shares might be interpreted as a reaction to the narrower rate spreads between liquid and time deposits at banks during the 1990s. Reductions in transaction costs and improvements in computer technology have increased the liquidity of bond and equity mutual funds, making them more competitive with liquid deposits in such an environment. He suggests that the lower turnover rates of bond and equity fund shares should not be taken as evidence that households have not responded to, and do not value, the increased liquidity of the funds. Moreover, although the funds are subject to liquidity (interest rate) risk, economic theory suggests that the expected holding period yields of these mutual funds should not differ system atically, on balance, from other market returns. Finally, I want to recognize the economic analysts in the Research Department of the Federal Reserve Bank of St. Louis who provided invaluable help in reviewing references and data for the sympo sium papers: Heidi L. Beyer, Heather Deaton, Kelly M. Morris and Richard D. Taylor. Richard G. Anderson St. Louis, Missouri November 1 ,1 9 9 4 NOVEMBER/DECEMBER 1994 7 Sean Collins and Cheryl L. Edwards Sean Collins and Cheryl L. Edwards are staff economists with the Board of Governors of the Federal Reserve System, Division of Monetary Affairs. An Alternative Monetary Aggregate: M2 Plus Household Holdings of Bond and Equity Mutual Funds St a n d a r d m o n e y -d e m a n d m o d e l s began to go off track in 1990, attesting to an apparent shift in the velocity of M2. This shift sparked debate about whether M2 velocity is stable, an important property for an indicator of monetary policy.1 It also raised questions about the usefulness of money-demand models for predicting the effects of Federal Reserve policy. If velocity shifts in some unforeseeable way, how is it possible for policymakers to exploit the statistical relationship between M2 and nominal income to attain their policy goals? In mid-1993, the Federal Reserve responded to the velocity shift by formally downgrading M2 as an indicator of the state of the economy. Meanwhile, interest has been rekindled in defining “money” and searching for alternative monetary aggregates. One explanation for the velocity shift is the increased importance of bond and equity mutual funds, also called long-term funds. Over the four-year period from 1990 to 1993, net purchases of bond and equity mutual funds by investors totaled about $655 billion, compared with about $400 billion over the 1980s. The surge largely reflected the yield-seeking behavior of retail investors. Yields on M2-type assets fell to his torically low levels in the early 1990s, while long-term market interest rates were unusually high relative to short-term rates, and equity prices were rising sharply. In this environment, investors sought higher returns by investing in bond and equity mutual funds. They may also have been attracted by the enhanced liquidity of long-term mutual fund shares. Most large mutual fund complexes upgraded their share holder services during the late 1980s to permit investors to write checks against their bond fund balances. At the same time, banks entered the mutual fund business as regulations that once prevented banks from selling mutual funds virtually evaporated. As a result, investors could buy and sell mutual fund shares in a familiar environment: the hank lobby. Taken together, these two developments—the recent case of missing M2 and the ascendant 1 In this context, stable is usually taken to mean that velocity is a stationary stochastic process. For some recent evi dence on the stability of velocity, see Hallmann and Anderson (1993). NOVEMBER/DECEMBER 1994 8 Figure 1 M2 Velocity and its Opportunity Cost Standardized value Standardized by subtracting mean and dividing by standard deviation mutual fund industry—raise the issue of whether bond and equity mutual funds ought to be added to M2. This paper proposes just that, but offers some caveats. The first section reviews the recent behavior of M2 and its deterioration as a predictor of economic activity. The second section provides historical background on the mutual fund indus try. The third section asks whether bond and equity mutual funds meet the tests of “moneyness” usually applied to assets considered for inclusion in a monetary aggregate. The fourth section discusses the data needed to construct a theoret ically sound monetary aggregate that includes bond and equity mutual funds. The fifth section describes how we constructed such an aggregate, which we will refer to as M2+. The final section concludes by pointing out some of the drawbacks of this aggregate. RECENT BEHAVIOR OF M2 The relationship between M2 and nominal income was fairly stable for many years before 1990. M2 velocity, which is the ratio of gross domestic product to the level of M2, fluctuated Digitized for FEDERAL FRASER RESERVE BANK OF ST. LOUIS around a constant level. Moreover, movements away from this level were strongly positively correlated with the opportunity cost of holding M2, measured as the spread between the yield on a short-term Treasury security and the weighted-average return on M2. When short-term Treasury rates rose, the opportunity cost of holding M2 increased because the rates on the components of M2 did not climb as fast as market rates. As a result, M2 growth would slow relative to the growth in income, and velocity would rise. As the rates paid on M2 completed their adjust ment to the higher level of market rates, M 2’s opportunity cost would narrow, M2 growth would rise relative to income growth, and velocity would return toward its trend level. This relationship meant that M2 growth could serve as a guidepost to current (but as yet unknown) income growth. In m id-1990, however, the velocity of M2 rose substantially above its long-run average, despite a very considerable drop in the opportunity cost of holding M2 (Figure 1). At the same time, conventional demand equations for M2, which are statistical representations of the relationship between money and variables such as interest 9 Figure 2 Actual and Predicted Growth Rates of M2 Percent rates and income, began to go off track (Figure 2). Between the first quarter of 1990 and the fourth quarter of 1993, the Board staffs quarterly model of M2 demand overpredicted growth by an average of 2.5 percentage points per quarter—m isesti mates that cumulated to nearly $380 billion by the end of 1993.2 Several alternative hypotheses have been advanced to explain the missing M2. Duca (1993a) postulated that the resolution of thrifts by the Resolution Trust Corporation (RTC) may have been a factor. In short, he argued that deposit rates were reset by depository institu tions that acquired the deposits of resolved thrifts. More often than not, the new deposit rates would be lower—typically much lower— than the rates that thrift investors had enjoyed earlier. This “sticker shock” led thrift deposi tors to reassess their portfolios in ways not cap tured by conventional money-demand models. Duca’s explanation is an important one for 1990 and 1991, but cannot explain the subsequent weakness in M2 because RTC funding and, thus, resolution activity dried up in 1992. Wenninger and Partlan (1992) focused on the weakness in small time deposits. They noted that the phasing out of Regulation Q (which limited the rates banks could pay on deposits) encouraged banks to think of small time deposits as managed liabilities. Consequently, banks, which faced weak credit demand, were rather unaggressive bidders for small CDs. The authors also noted that, on the demand side, consumers may have been surprised by the substantial decline in deposit rates, and they therefore sought the higher returns available on mutual fund investments. Other explanations advanced in the press and elsewhere attributed the weakness in M2 to a number of sources: the credit crunch; rising deposit insurance premiums; the imposition of new, higher capital standards for depositories; the downsizing of consumer balance sheets (which was accomplished by using M2 balances to pay off debt); the unusual 2 This model is described in Moore, Porter and Small (1990). NOVEMBER/DECEMBER 1994 10 steepness of the yield curve; and, finally, the especially strong flows into bond and equity mutual funds over the 1991-93 period. A common thread binds these stories: They highlight some facet of household demand for money not captured in conventional moneydemand models. For instance, if the demand for money by households is influenced by returns on capital market instruments, then this effect would be reflected as an error in conventional money-demand models, because such models usually depend only on the spread between the yield on a short-term Treasury security and the weighted-average rate paid on M2 balances. Similarly, to the extent that rising deposit insur ance premiums were not accurately reflected in reported deposit rates, conventional models could experience significant forecast errors. The incompleteness of conventional money-demand models sparked attempts to revamp such models. Feinman and Porter (1992) augmented the Board staffs model of the demand for M2. Rather than defining the opportunity cost of M2 as the spread between a short-term Treasury rate and the rate of return on M2 balances, they estimated the opportunity cost of holding M2 balances. Their model chose the opportunity cost by selecting among rates of return on M2-type balances and competing assets.3 In contrast to conventional models, Feinman and Porter found that yields on longer-term Treasury instruments and consumer debt were significant factors in determining money demand. A steep yield curve tended to dampen money growth and helped to explain weak M2 growth. Although the Feinman-Porter model achieved some success in predicting M2 growth out-of-sample, the model had difficulty beginning in m id-1993, when long-term interest rates fell sharply. In part, this problem may have stemmed from an asymmetric response of investors to changes in the slope of the yield curve. If the yield curve flattens because long-term interest rates have declined, investors in mutual funds may enjoy temporary capital gains, thus depressing their appetites for M2 balances. In contrast, if the yield curve flattens 3 The rates of return included on M2-type balances were for other checkable deposits, savings accounts (including MMDAs), small time deposits with original maturities of six months, small time deposits with original maturities of two-and-a-half years or over, and the yield on money market mutual funds. Yields on competing instruments included those for three-month Treasury bills, five-year Treasury notes, 30-year Treasury bonds and the 48-month auto loan rate. FEDERAL RESERVE BANK OF ST. LOUIS because short-term rates have risen, investors would garner no capital gains (and might even confront capital losses on short-term securities), and they would see an erosion of the yield advantage of mutual funds over M2-type balances. However, the Feinman-Porter model treats the two kinds of flattening the same. A different approach advocated by Hendry and Ericsson (1990) for the United Kingdom, and recently employed by the Board staff, is to introduce an “error-learning” term into the conventional M2 demand equation. The errorlearning term attempts to capture changes in preferences as investors “learn” about mutual funds and their potentially higher yields. Nonetheless, as suggested by Higgins (1992), if the slowdown in M2 was to some extent a permanent phenomenon related to restructuring (both regulatory and technical) in the financial industry, then error-learning models might even tually go off track as well. Indeed, this appears to be the case, as the Board’s error-learning model has recently been overpredicting money growth. The standard model has been modified by adding other variables as well. Carlson and Byrne (1992) and Duca (1993a) both included variables that accounted for the impact of thrift closings. Duca (1993b) further modified the model by changing the dependent variable to be M2 plus various measures of households’ holdings of bond mutual funds. He found that both the assets of bond mutual funds and RTC activity helped explain the missing M2. Instead of reworking conventional moneydemand models, many economists have suggested abandoning M2 as an aggregate and replacing it with another, more predictable (they hope) aggregate. The search for a replacement to M2 has given rise to a cottage industry of constructing and testing alternative aggregates. Among the proposed successors to M2 are M l, M l A, liquid M2 (M2 less small time deposits), MZM (M2 less small time deposits plus assets of institution-only money market mutual funds), M2E (M2 plus assets of institution-only money funds), household M2 11 (M2 less demand deposits, and overnight RPs and Eurodollars), M2BF (M2 plus bond mutual funds) and, most recently, M2+ (M2 plus household holdings of bond and equity mutual funds).4 To date, none of these proposed aggregates has been particularly well-received because all are plagued by theoretical or empirical difficulties. With respect to M l and M lA , recent history clearly demonstrates that they are too highly interest elastic to serve as a useful indicator of income growth. Liquid M2, considered by Wenninger and Partlan (1992), among others, seems appealing on the theoretical grounds that small CDs are neither very liquid nor trans action balances; liquid M2, however, suffers from the same interest-elasticity problem as M l. Moreover, the velocity of liquid M2 has been less predictable than that of M2 itself. Poole’s proposed aggregate, MZM, is subject to the same criticism and would additionally include a com ponent (institution-only money funds) that is extremely sensitive to money market pressures and thus is highly volatile. Institutional investors will make large adjustments in their holdings of money funds in response to very small differen tials between market rates and those on money funds. This sensitivity was demonstrated in February 1994, when nearly $16 billion flowed out of institution-only money funds following a 0.25 percentage point increase in the federal funds rate. M2BF and M2+ are not trouble-free, either on empirical or theoretical grounds. Each attempts to internalize some of the observed substitution between M2 and long-term mutual funds. These aggregates therefore should have a more stable relationship to nominal income than M2 alone. The empirical evidence, however, suggests that these aggregates may not be much more stable than M2. For instance, Orphanides, Reid and Small (1994) point out that the velocity of M2+, although perhaps more predictable than that of M2, would have led to substantial overpredictions 4 Monetary economists in other industrialized countries have grappled with similar problems as their official monetary aggregates have succumbed to financial innovation and deregulation. In Canada, McPhail (1993) proposed an M2+ aggregate consisting of M2 balances plus savings bonds and short-term Treasury securities. Arestis and others (1993) discussed the difficulties faced by the Bank of England in defining and controlling monetary aggregates during the 1980s. They concluded that “trying to target the growth rate of broad monetary aggregates in the UK...has always been problematic because of the weak and some of GDP growth during the past few years, just like M2 (indeed, the velocity of M2+ grew quite rapidly in early 1994). In part, these overpredic tions may not stem from the definition of the augmented aggregate but rather from an inappro priate measuring of the opportunity cost variable. The authors use the slope of the yield curve as a proxy for the yield advantage of holding bond and equity funds; however, there is no necessary reason why the yield on long-term bonds should be a good proxy for the return on equities. Moreover, one can argue that if the expectations hypothesis is true, the yield on short-term Treasury securities should adequately measure the opportunity cost.5 Difficult issues arise on the theoretical side as well. For instance, should we add just bond funds to M2, or should equity funds be included as well? Duca (1993) focused mainly on bond funds on the grounds that equity funds carry substantial principal risk and therefore are less substitutable for M2 balances. The liquidity of equity funds, however, suggests that there may be some benefit to including these funds in an augmented aggregate. A thornier issue is the treatment of capital gains. Orphanides, Reid and Small (1994) note that excluding capital gains from net assets could lead to substantial misestimates of potential balances and introduce an element of arbitrariness into measuring the aggregates, but including capital gains may permit changes in interest rates or equity prices to intro duce excessive volatility into the aggregate. As a consequence, the M2+ aggregate, although it has the advantage of internalizing portfolio shifts between mutual funds and M2, will be quite sensitive to movements in bond and equity prices. This problem poses difficulties, but the difficulties may be somewhat less severe than the problems affecting the alternative aggregates discussed earlier. Nonetheless, in order to better interpret the movements in M2+, one must track capital gains and losses. In this paper, we focus on issues related to the construction of M2+. These issues include times perverse relationships between the level of absolute rates and the relative rates which form the key to [money targeting].” To our knowledge, though, economists in the United States were the first to propose incorporating bond and equity mutual funds into a monetary aggregate. 5 This point is a matter of debate. For a contrary view, see Feinman and Porter (1992). NOVEMBER/DECEMBER 1994 12 Figure 3 Net Assets of Bond and Equity Mutual Funds Billions of dollars 1600 ' 1400 ' 1200 ' 1000 - 800 600 400 200 - 0 1980 81 “ I-----------i---------- r — 82 83 84 85 i---------- i---------- i---------- i---------- i---------- i-----------i---------- i---------- r - 1 86 the following: Should we exclude from M2+ the assets of mutual funds devoted to retirement accounts? Should we exclude the liquid assets held by mutual funds on the grounds that such assets are “money” and have already been counted in M2? Should we exclude the assets of international funds, whose underlying invest ments are denominated in currencies other than dollars and thus may not reflect purchasing power within the United States? More pragmatically, do the data exist to make such adjustments? RECENT HISTORY OF THE M U T U A L FUND INDUSTRY Net assets of stock and bond mutual funds were nearly $1.5 trillion at the end of 1993, about 24 times higher than in 1980 (Figure 3). Most of this dramatic growth reflected heavy purchases of fund shares by investors, as opposed to revaluations of fund investments. 6 A mutual fund is a type of investment company. It sells shares representing an interest in a pool of securities. The minimum initial investment for many long-term funds is around $2,500, in contrast to, say, the $10,000 minimum investment needed to purchase a Treasury bill. For a more FEDERAL RESERVE BANK OF ST. LOUIS 87 88 89 90 91 92 93 1994 During the 1980s, net purchases of bond and equity mutual funds averaged $54 billion per year. The upswing during the 1980s was prompted, in part, by rising stock and bond prices. With incomes and wealth rising, investors were inter ested in taking advantage of potential gains in equity and bond markets, and mutual funds permitted small investors to invest in a diversified portfolio at low cost.6 Investor interest may also have been spurred between 1982 and 1986 by the incentives to invest in individual retirement accounts (IRA) and Keogh accounts.7 The popularity of bond and equity mutual funds soared in the early 1990s. Two record years were reported in 1992 and 1993, when investors made net purchases of $202 billion and $266 billion, respectively (Figure 4). The increased pace of purchases stemmed, in large part, from the low-interest rate environment and the steepness of the yield curve. In early 1989, complete discussion of recent trends in the mutual fund industry, see Mack (1993). 7 IRAs and Keogh accounts are two types of tax-sheltered accounts that are used to save for retirement. 13 Figure 4 Net Flows into Long-Term Bond and Equity Mutual Funds Billions of dollars, 12-month sum Net flow equals total sales less redemptions. Figure 5 Yields on Treasury Bills and Bonds Yield in percent NOVEMBER/DECEMBER 1994 14 Table 1 Bond and Equity Bank-Related Mutual Funds 1989 1990 1991 Total assets of bank-related funds 9.8 13.1 26.4 Percent of industry assets 1.7 2.3 3.3 Number of funds 1989 1990 1991 Total bank-related mutual funds 213 271 359 Percent of all mutual funds 9.5 11.5 13.8 Dollar Value of Assets1 1992 1993 19942 46.7 85.5 96.4 4.4 5.7 6.3 1992 1993 1994 502 954 1211 16.8 26.2 31.1 Source: Upper Analytical Services and Investment Company Institute. 1 Billions of dollars 2 Observation for 1994 is June. the yield curve was essentially flat. Between March 1989 and December 1993, the yield on three-month Treasury bills fell by about 6 per centage points to around 3 percent, while the yield on 30-year bonds fell by 3 percentage points to about 6.25 percent (Figure 5). At the same time, equity prices were rising. Over the 24 months from January 1992 to December 1993, the stock market advanced just over 20 percent. In this rate environment, investors sought the higher returns available in long-term mutual funds. Equity funds were also apparently boosted by households substituting out of direct purchases of equities. The flow-of-funds accounts show that households’ direct holdings of equities fell $21 billion in 1993, in contrast with the $140 billion of inflows into equity mutual funds. Faced with a loss of deposits to mutual funds, many banks began entering the mutual fund business in the late 1980s. Since 1989, the assets of bank-related mutual funds have increased tremendously, relative to the total assets of the mutual fund industry. As shown in the top panel of Table 1, long-term bank-related mutual funds accounted for only about 2 percent of total assets of long-term funds in 1989, but this figure had more than tripled by mid-1994. As the bottom panel of Table 1 shows, the number of long-term funds offered by banks climbed even faster, experiencing more than a fivefold increase from 1989 to mid-1994. In contrast, for the industry as a whole, the number of long term funds less than doubled over the same FEDERAL RESERVE BANK OF ST. LOUIS period (from 2,242 funds to 3,894 funds). As a result, the banking sector created nearly 60 per cent of all new long-term funds over this period. The watershed year for bank-related long-term funds was 1992, when the average number of funds sold per bank jumped sharply. In that year, Nationsbank started 55 new long-term funds, and many other banks plunged into the business as well. ARE MUTUAL FUND SHARES MONEY? Economists have typically asked two questions when designing monetary aggregates: Do the assets in question serve as a transaction balance or a medium of exchange, and is the asset readily convertible into a transaction balance? For our purposes, we consider to what extent shares held in long-term mutual funds may be used as payment for goods, services or other assets. We also explore whether individuals consider fund shares to be readily convertible into transaction balances. Investments in bond and equity mutual funds cannot generally be used as a medium of exchange or as a transaction balance. Some bond funds are exceptions. Bond funds some times offer a check-writing option that permits investors to make purchases by writing a check directly against their bond fund assets. Thus, there is good reason to consider some portion of the assets of bond funds to be transaction balances. Indeed, check writing is nearly uni versal among money market mutual funds, whose assets are included in M2. On the other 15 hand, although both money market funds and bond funds typically impose minimum dollar amounts on checks written against them (often $500), the check-writing feature of bond funds is usually less flexible, because they sometimes impose maximum dollar amounts (often 50 per cent of an investor’s total bond fund assets).8 Irrespective of the check-writing features they offer, mutual funds can be quite liquid. Investors can usually redeem assets by telephone and have the proceeds wired to their checking or money market fund the same day.9 Moreover, the vast majority of fund complexes (a group of funds managed by the same advisor) routinely offer exchange privileges at nominal cost or no cost. Exchange privileges allow investors to shift assets between long-term funds and money market funds (or other long-term funds) within their complex. Thus, a telephone call again allows fund investors to shift in and out of M2-type accounts (that is, retail money market mutual funds). Figure 6 shows the cumulative sum of net exchanges out of money market funds into long-term funds for the years 1989 to 1993. This sum rose during the 1990-93 period, when the M2 forecasting equation went off track. Although the cumulative sum is not particularly large, relative to the $380 billion of missing M2, it does suggest that there is a quantitatively important, and technologically direct, substitution between a component of M2 and long-term mutual funds. One factor that tempers this argument is that some mutual funds charge exit fees (back-end loads). By raising transaction costs, back-end loads reduce substi tutability between long-term mutual funds and M2 balances. Another way to decide whether an asset is “money” is to look at its turnover rate, measured as total withdrawals divided by outstanding balances. Spindt (1985) argued that transaction accounts have high turnover rates; therefore, the higher the transaction balance, the greater the degree of “moneyness.” Figure 7 shows turnover rates for some of the components of M2, along with turnover rates for equity and bond funds. Turnover rates for checkable deposits are quite high, reflecting their use as the primary transaction account for most households. Turnover rates on savings accounts and money market mutual funds are somewhat lower. Turnover ratios on 8 Maximums are imposed to prevent an investor from incurring an overdraft if bond prices fell sharply on the day that the check clears. long-term funds are quite low, an indication that individuals regard these accounts mainly as sav ings vehicles, rather than transaction balances. Figure 7 lacks one important series: the turnover ratio for small time deposits. Unfortunately, the Federal Reserve does not collect this information. Staff at the Federal Reserve Board, however, have estimated that the turnover ratio for small time deposits is on the order to 1 percent to 1.5 percent at an annual rate, which is reasonably close to the estimated turnover ratio for long-term funds. In summary, there are some reasons to think of the assets of bond and equity mutual funds as “money,” or at least close substitutes for money. Although the reasons for believing mutual fund assets to be money are not overwhelmingly strong, they are about as favorable as the case for calling small time deposits money. Both small time deposits and long-term mutual fund assets are mainly savings balances, as opposed to transaction balances. Both have a high degree of substi tutability (or potential substitutability) with other kinds of M2 balances. In addition, investors in these instruments may face some penalty for withdrawing their balances. DESIRED DATA The current monetary aggregates measure the public’s holdings of money first by summing the total outstanding amounts of instruments deemed to be money and then subtracting money holdings of the U.S. government, foreign governments, depository institutions, and money market mutual funds. (Table 2 shows the construction of M2.) The current aggregates also attempt to distinguish between individual (retail) and institutional holdings of money market mutual funds. This distinction is based on the belief that individuals’ holdings of money market mutual funds are more closely related to consumption and income than are institutions’ holdings, which are more tightly linked to financial market conditions. In practice, however, making such a distinction is difficult because data on money funds are available only by type of fund, not by type of holder. Conse quently, the distinction between retail and institutional money funds is not clear cut: Funds deemed to be institution-only funds accept investments from individuals as long as those individuals meet the sizable minimum investment 9 Proceeds can also be mailed by check, but this option would reduce liquidity relative to the wire transfer option, NOVEMBER/DECEMBER 1994 16 Figure 6 Net Exchanges Into Long-Term Funds from Money Market Mutual Funds Billions of dollars Figure 7 Turnover Ratios of Selected Financial Assets Turnover ratio (annualized)/log scale FEDERAL RESERVE BANK OF ST. LOUIS 17 requirements, and institutions can invest in retail funds.10 The guiding principle we followed when constructing M2+ was to parallel the current construction of M2. The data needed to add bond and equity mutual funds to M2 are in many ways similar to those that were needed to incorporate money market mutual funds into M2. We must be able to measure the total net assets of long-term mutual funds and, for netting purposes, the monetary investments of such funds, as well as balances held by institutions or invested in retirement accounts. Table 3 summarizes the series necessary for the construction of M2+. The first requirement for building M2+ is an accurate measure of total net assets of bond and equity mutual funds. Total net assets of a mutual fund are simply its total assets— essentially the market value of its securities portfolio—less its total liabilities, which include such items as accounts payable (for investments purchased or shares redeemed), accrued management fees, and other accrued expenses." We included reinvested dividends and capital gains in order to treat these items like interest credited on deposits, which is included in M2. The second requirement is to distinguish between mutual fund holdings of individual and institutional investors. In order to parallel the current treatment of money market mutual funds, we would split bond and equity mutual funds into retail and institution-only funds; we cannot do so, however, because long-term funds typically accept investments from both individual and institutional investors. Therefore, we need data on holdings of long-term mutual funds by type of investor, with individual holdings appearing in M2+. Bond and equity mutual funds invest in instruments included in the monetary aggregates. If the total net assets of these funds were simply added to M2, the funds’ holdings of monetary instruments would be counted twice because their holdings already would be included in the outstanding amounts of monetary instruments, such as overnight RPs with banks. To avoid this Table 2 Current Definitions of the Monetary Aggregates_______________________ M1 = currency (public holdings) + travelers checks of nonbank issuers + demand deposits at all commercial banks (less cash items in the process of collection and Federal Reserve float) - demand deposits due to depository institutions, the U.S. government, foreign banks, and foreign official institutions + other checkable deposits M2 = M1 + overnight and continuing contract RPs issued by all depository institutions + overnight Eurodollars issued to U.S. residents by foreign branches of U.S. banks worldwide + savings deposits (including money market deposit accounts) + small-denomination time deposits + balances at retail money market mutual funds - U.S. commercial bank, U.S. government, foreign government, foreign commercial bank, and retail money market mutual fund holdings of all non-M1 components - IRA/Keogh balances at depository institutions and money market mutual funds double counting, monetary investments of long term funds should be excluded from M2+, just as holdings of monetary instruments by money funds and depository institutions are excluded from M2. Therefore, we need data on bond and equity fund holdings of overnight Eurodollars and overnight RPs with depository institutions. Moreover, the data ideally would allow us to apportion these so-called netting items between those due to retail investors and those due to institutional investors. Finally, paralleling the current aggregates requires us to exclude from the M2+ aggregate IRA and Keogh balances held as bond and equity 10 Institution-only money funds are funds that impose high min imum balances on shareholders. These minimums— usually $50,000— are prohibitively high for the great majority of retail customers. 11 The value of a share in a long-term fund is calculated each day by dividing the total net assets of the fund by the num ber of shares outstanding. NOVEMBER/DECEMBER 1994 18 mutual fund shares. IRA/Keogh accounts are savings vehicles and typically are not used for transaction purposes. Indeed, these accounts are extremely illiquid because federal law imposes stiff penalties for withdrawals before retirement. Of course, for individuals of retirement age (59and-a-half or older), IRA/Keogh balances are liquid. As a practical matter, though, it would be nearly impossible to estimate the percentage of IRA/Keogh balances that might be considered liquid. As a result, instead of making an arbitrary assumption about the proportion of IRA/Keogh balances that are liquid, M2 simply excludes all IRA/Keogh balances held in deposit accounts and money market funds, and we follow that approach for M2+ by removing balances held in IRA/Keogh accounts from the measure of total net assets of long-term funds. ACTUAL CONSTRUCTION OF M2+ Not all of the desired data set forth in the previous section are available. This section therefore discusses data availability and gives the technical details on how the augmented aggregate M2+ was constructed. (Table 4 provides a summary.) The Investment Company Institute (ICI) is a prominent source of mutual fund data. It is funded by contributions from its members, which comprise the vast majority of investment compa nies. On behalf of the Federal Reserve, ICI has been collecting data on money funds for over 12 years, including data on their assets, their invest ments and their IRA/Keogh balances. These s e r ie s a re u s e d in th e c o n s tr u c tio n o f M 2 , M 3 and L. We have used ICI’s data on long-term funds to construct M2+. Net Asset Data Monthly data on the total net assets of bond and equity mutual funds are drawn from ICI’s Trends in M utual F u n d Activity. This release publishes (with a lag of about one month) data on total net assets held by long-term funds measured as of month-end. In 1993, ICI collected data from over 3,000 long-term funds. From the Trends releases, we compiled a monthly series on the total net assets of bond and equity mutual funds back to 1959. We deviated from our guiding principle of constructing M2+ like M2 in only one area: net assets of global and international mutual funds. These funds invest in debt and equity instruments FEDERAL RESERVE BANK OF ST. LOUIS Table 3 Theoretical Definition of M2+ M2+ = current M2 + net assets of bond funds held by individual investors + net assets of equity funds held by individual investors - bond and equity fund holdings of overnight RPs attributable to individuals - bond and equity fund holdings of overnight Eurodollars attributable to individuals - IRA and Keogh holdings at bond and equity mutual funds denominated in foreign currencies, as well as those denominated in dollars. Although M2 excludes deposits denominated in foreign cur rencies, we incorporated the assets of these types of funds in M2+. This treatment seems justified by the following considerations. First, doing so permits M2+ to internalize substitutions between non-dollar-denominated and dollar-denominated funds. Second, the foreign currency deposits netted from M2 have been quite small, amounting to only a few billion dollars. Moreover, these assets are thought to be held by institutions, mainly for clearing purposes. Thus, the netting of these assets has little impact on the monetary aggregates, and doing so skirts a difficult theo retical issue: Are non-dollar-denominated assets money in the United States? This issue cannot be swept aside when considering long term mutual funds. The assets of international mutual funds are thought to be held mainly by retail investors and are growing rapidly. Although the underlying investments of these funds are denominated in foreign currencies, these funds’ net asset values (share prices) are reported in dollars and thus may be viewed by retail investors as liquid investments that are highly substitutable for assets in M2 or for other kinds of mutual funds whose underlying investments are denominated in dollars. Finally, because of data limitations, it would be very difficult to extract only the dollar-denominated holdings of these funds for inclusion in M2+. Institutional Holdings In order to apportion the total net assets of 19 Table 4 Actual Construction of M2+ holdings on a monthly basis. Nonetheless, analyses of the long-run behavior of M2+ likely would not be impaired much because, by con struction, the interpolated series is tied to ICI’s year-end observations. M2+ = current M2 + net assets of bond funds Liquid Investments + net assets of equity funds The current monetary aggregates, M2 and M3, are constructed to avoid double-counting of assets. For example, the RP and Eurodollar investments of money market mutual funds are excluded from the monetary aggregates to avoid counting them twice: once in the money fund component of M2, and again in the RP and Eurodollar components of M2. To avoid double-counting when con structing M2+, we needed data on long-term mutual fund holdings of M2 instruments, such as overnight RPs and Eurodollars. Most bond and equity mutual funds hold a portion of their port folios in liquid investments to meet investors’ demands for share redemption. ICI has collected monthly data on the total liquid investments of long-term funds since 1960, with a detailed break down of such investments into short-term Treasury securities, short-term municipal secu rities, and “cash and other receivables” beginning in 1991. Nonetheless, this breakdown is still too aggregated for our purposes. Rather than arbitrarily create a series on long-term mutual funds holdings of M2-type assets, which would consist principally of overnight RPs and Eurodollars, we chose to make no adjustment at all. Our view was that these holdings would be a relatively small portion of M2+ and that any assumption we might make to remove them could well introduce greater error into M2+ than that caused by double-counting them. - institutional holdings of bond and equity funds - IRA and Keogh assets at bond and equity funds long-term mutual funds between retail and insti tutional holdings, we made use of the ICI’s survey of institutional holdings of long-term mutual funds. ICI first surveyed such holdings in 1954 and did so biennially until 1980. Since that time, its survey has been conducted annually. In the survey, mutual funds report the percent of their assets held by various institutions on the last day of the calendar year. Among the largest institutional holders are fiduciaries (bank trust departments) and retirement plans. The data are disaggregated enough to permit us to apportion institutional holdings of mutual funds between long-term funds and money market funds for survey dates back to 1974. For earlier years, we have assumed that the assets of money market mutual funds were zero; thus, any institutional holdings in these years are allocated to long term funds. We had to make several adjustments to the raw data in order to derive a monthly series on insti tutional holdings. First, we adjusted for the changes in the format of ICI’s survey by making several ad hoc adjustments to remove the breaks in the series. Second, we converted the breakadjusted series into monthly observations. We did so by linearly interpolating between surveys the end-of-year ratio of institutional assets to total net assets of long-term funds. Third, we multi plied each of the resulting monthly ratios by total net assets in the corresponding month to derive the series on institutional holdings. We could then construct a series on retail holdings of long-term mutual funds by subtracting our institutional series from total net assets. The resulting series therefore may be subject to con siderable month-to-month error because of the assumptions required to estimate institutional 12 Only year-end figures on these assets are available for 1981 and 1982. Quarter-end figures are available for IRA accounts from 1975 to 1980 and from 1964 to 1980 for IRA/Keogh Assets Retirement account balances are thought to be too illiquid to be used as transaction balances. Consequently, the monetary aggregates exclude balances in IRA and Keogh accounts. To parallel this treatment, we used ICI’s data on IRA/Keogh balances in long-term mutual funds to exclude such holdings from M2+. Month-end data on the assets of bond and equity mutual funds that are held in IRA and Keogh accounts balances begin in January 1983. Data for earlier periods are avail able either quarterly or annually.12 For years prior to 1983, we constructed monthly observations on Keogh accounts. Earlier data for IRA accounts are not avail able, but balances in these accounts were essentially nil before 1977. NOVEMBER/DECEMBER 1994 20 IRA/Keogh balances of long-term funds by linearly interpolating the ratio of IRA/Keogh assets to total net assets of long-term funds between the quarterly or annual observations. We then multiplied this interpolated ratio by the total net assets of long term funds for the corresponding month.13 Deriving Monthly Averages ICI’s long-term mutual fund data are monthend observations. M2 is a monthly average derived from either daily or weekly data, depending on the component. In order to add the mutual fund data to M2, items related to mutual funds had to be converted to a month-average basis. We approximated monthly averages for ICI’s data by taking two-month moving averages of the month-end figures for total net assets, institutional holdings and IRA/Keogh balances. Seasonal Adjustment As Figure 8 shows, some of the components of the monetary aggregates have large seasonal regularities. Seasonality in the aggregates arises primarily from their transaction nature; for instance, currency demand is seasonally high in December because of Christmas shopping. Tax payments, other holidays and interest cred iting also occur seasonally. If the monetary aggregates were not adjusted for seasonal varia tion, it would be hard to discern changes in money demand related to movements in interest rates and income. Because it is these latter effects that are of primary interest to policymakers, seasonal adjustment of the monetary aggregates is imperative. As a rule, seasonality is strongest within the components of M l and less so for those in nonM l M2 because the components of non-M l M2 are not used as extensively as M l for transaction purposes (Figure 9). With respect to long-term mutual fund assets, which we have suggested may be driven less by transaction motives than by savings motives, one might expect to find weak, or nonexistent, seasonality. If so, it would obviate 13 There are many other types of retirement vehicles, such as employer-sponsored retirement plans and annuities; however, only IRA/Keogh balances are subtracted from M2. The measure of M2+ that we constructed did not include balances in retirement accounts other than IRA and Keogh accounts because these accounts are included in ICI’s measure of institutional holdings. Thus, subtracting institutional holdings from total net assets removes these balances from M2+. Strictly speaking, we have again deviated from our practice of paralleling the construction of M2, but we do not feel that the deviation impairs the usefulness of the M2+ aggregate. FEDERAL RESERVE BANK OF ST. LOUIS the need for us to seasonally adjust such assets before adding them to M2. Figure 10 indicates that there is indeed little apparent seasonality in the assets of mutual funds. Nevertheless, our guiding principle of paral leling the current construction of M2 dictates that mutual fund assets should be seasonally adjusted before adding them to M2. Accordingly, we used the following seasonal adjustment pro cedure. We constructed a not-seasonally adjusted (NSA) measure of household holdings of mutual fund assets, called the “plus” portion of M2+, as shown in Table 4. This component is the assets of bond and equity funds less institutional hold ings and IRA/Keogh balances. The “plus” portion is then seasonally adjusted using Census X - l l , assuming multiplicative seasonality.14 Seasonally adjusted M2+ is constructed by adding the sea sonally adjusted “plus” portion to seasonally adjusted M2. A comparison of M2 and M2+ is shown in Figure 11, and Appendix 1 provides estimates of M2+ for recent years.15 CONCLUDING REMARKS This paper has detailed construction of an alternative monetary aggregate that adds house hold holdings of bond and equity mutual funds to M2. This aggregate has the advantage of internalizing some of the observed substitution between long-term mutual funds and M2 bal ances. The empirical evidence discussed by Orphanides, Reid and Small (1994) suggests that M2+ does have some advantages over M2, in that its velocity appears to have been more “sensible” over the past few years. Consequently, the predictability of GDP is improved. A major drawback of this aggregate, however, is that it is very sensitive to movements in bond and equity prices. For example, in early 1994, the velocity of M2+ rose sharply, in large part reflecting declines in bond and equity prices, and it remains to be seen whether the velocity of M2+ will return to its trend level. In fact, the only sure test of a monetary aggregate is the test of time. If financial 14 This algorithm is essentially the same one used to seasonally adjust the current monetary aggregates. For a complete description of that algorithm, see Farley and O’Brien (1987). 15 The reader is referred to Orphanides, Reid and Small (1994) for a discussion of the empirical properties of M2+. Figure 8 Seasonality in Selected Components of M2 (billions of dollars) First Differences in Currency 6 First Differences in Demand Deposits 15 to First Differences in Savings (including MMDAs) 25 NOVEMBER/DECEMBER 1994 1990 1991 1992 1993 First Differences in Small Time Deposits 10 FEDERAL RESERVE BANK OF ST. LOUIS Figure 9 Seasonality in Selected Monetary Aggregates (billions of dollars) First Differences in M1 First Differences in M2 30 30 20 20 10 10 0 0 -10 -10 -20 -20 -30 hJ N> First Differences in Non-M1 M2 20 15 10- 5- o-5 - 10- -15 - 20- -25 - 1990 1991 First Differences in Non-M2 M3 15 - 1992 1993 23 Figure 10 Seasonality in Assets of Long-Term Mutual Funds Growth rate of assets of bond and equity mutual funds (annualized) Figure 11 M2, and M2 Plus Household Holdings of Bond and Equity Mutual Funds Billions of dollars NOVEMBER/DECEMBER 1994 24 innovation continues at a rapid pace in the 1990s and depositories continue to lose their share of credit intermediation, then the usefulness of M2+ as an indicator of economic activity may grow as intermediation continues to shift to mutual funds. REFERENCES Arestis, Philip, I. Mariscal Biefang-Frisancho, and P.G.A. Howells. “UK Monetary Aggregates— Definition and Control,” in Philip Arestis, ed., Money and Banking, Issues for the Twenty-First Century, Essays in Honour o f Stephen F. Frowen. St. Martin’s Press, pp. 163-91. Carlson, John B., and Susan M. Byrne. “Recent Behavior of Velocity: Alternative Measures of Money” , Federal Reserve Bank of Cleveland Quarterly Review (second quarter 1992), pp. 2-10. Duca, John. “Should Bond Funds be Included in M2” , Journal o f Banking and Finance (forthcoming). _____ . “RTC Activity and the ‘Missing M2,’ Economics Letters (No. 1, 1993), pp. 67-71. Farley, Dennis, and Yuch-Yun C. O’Brien. “Seasonal Adjustment of the Money Stock in the United States,” Journal of Official Statistics, vol. 3 (No. 3, 1987), pp. 223-33. Feinman, Joshua, and Richard Porter. ‘T he Continuing Weakness in M2” , Federal Reserve Board FEDS Paper #209 (September 1992). Hallman, Jeffrey J., and Richard G. Anderson. “Has the Longrun Velocity of M2 Shifted? Evidence from the P* Model,” Federal Reserve Bank of Cleveland Economic Review (first quarter 1993), pp. 14-26. Hendry, David F., and Neil R. Ericsson. “Modeling the Demand for Narrow Money in the U.K. and the United States,” FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Board International Finance Discussion Papers #383 (1990). Higgins, Bryon. “Policy Implications of Recent M2 Behavior,” Federal Reserve Bank of Kansas City Economic Review (third quarter 1992), pp. 21-36. Investment Company Institute. Trends in Mutual Fund Activity. October 1994. Mack, Philip R. “Recent Trends in the Mutual Fund Industry,” Federal Reserve Bulletin (November 1993), pp. 1001-12. McPhail, Kim. “The Demand for M2+, Canada Savings Bonds, and Treasury Bills,” Bank of Canada Working Paper 93-8 (September 1993). Moore, George, Richard Porter, and David H. Small. “Modeling the Disaggregated Demands for M2 and M1: The U.S. Experience in the 1980s”, in Peter Hooper and others, eds., Financial Sectors in Open Economies: Empirical Analysis and Policy Issues. Federal Reserve Board of Governors, 1990, pp. 21-105. Orphanides, Athanasios, Brian Reid, and David H. Small “The Empirical Properties of a Monetary Aggregate That Adds Bond and Stock Funds to M2”, this Review (November/December 1994), pp. 31-51. Poole, William. Statement Before the Subcommittee on Domestic Monetary Policy of the Committee on Banking, Finance and Urban Affairs, U.S. House of Representatives. November 6, 1991. Spindt, Paul A. “Money is What Money Does: Monetary Aggregation and the Equation of Exchange,” Journal of Political Economy (No. 1, 1985), pp. 175-204. Wenninger, John, and John Partlan. “Small Time Deposits and the Recent Weakness in M2”, Federal Reserve Bank of New York Quarterly Review (spring 1992), pp. 21-35. 25 Appendix 1 M2 and M2 Plus Bond and Stock Mutual Funds (M2+) Levels (billions of $) M2 (1) Growth rates M2+ (2) M2 (3) M2+ (4) Difference (4)-(3) Monthly 1993 1993 1993 1993 1993 1993 1993 1993 1993 1993 1993 1993 1994 1994 1994 1994 1994 1994 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 $ 3502.8 3494.2 3494.8 3498 3521.9 3528.7 3533.7 3536 3544.2 3547.8 3560.1 3567.6 3572.8 3568.9 3582.9 3589.9 3590.9 3581.3 $ 4046.7 4063.6 4066.2 4072.8 4108.1 4128.5 4148.3 4169.2 4201.8 4225.6 4252.3 4265.1 4278.9 4280 4276.7 4277.1 4278 4263.8 -2.1 -2.9 .2 1.0 8.1 2.3 1.6 .7 2.7 1.2 4.1 2.5 1.7 -1.2 4.6 2.3 .3 -3.1 3.5 5.0 .7 1.9 10.4 5.9 5.7 6.0 9.3 6.8 7.5 3.6 3.8 .3 -.9 .1 .2 -3.9 -5.6 -7.9 -.5 -.8 -2.2 -3.6 -4.0 -5.2 -6.6 -5.5 -3.4 -1.0 -2.1 -1.6 5.6 2.2 .0 .7 1991 1991 1991 1991 1992 1992 1992 1992 1993 1993 1993 1993 1994 1994 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3380.6 3418.2 3427.1 3444.3 3478 3480.6 3488.9 3509 3497.3 3516.2 3538 3558.5 3574.9 3587.4 3702.5 3765.8 3796.3 3851.2 3912.6 3935.7 3974 4022 4058.8 4103.1 4173.1 4247.7 4278.5 4273 3.8 4.4 1.0 2.0 3.9 .2 .9 2.3 1.3 2.1 2.4 2.3 1.8 1.3 5.5 6.8 3.2 5.7 6.3 2.3 3.8 4.8 3.6 4.3 6.8 7.1 2.9 -.5 -1.7 -2.3 -2.1 -3.7 -2.4 -2.0 -2.9 -2.5 -5.0 -2.2 -4.3 -4.8 -1.0 1.9 2916.6 3070.6 3219.2 3348.5 3444.3 3509 3558.5 3197.4 3353.9 3535.3 3651.6 3851.2 4022 4247.7 4.3 5.2 4.8 4.0 2.8 1.8 1.4 4.9 4.8 5.4 3.2 5.4 4.4 5.6 -.6 .3 -.5 .7 -2.6 -2.5 -4.2 Quarterly Annual (Q4/Q4) 1987 1988 1989 1990 1991 1992 1993 1 The market value of mutual funds balances added to M2 excludes balances in IRA/Keogh accounts and institutional holdings of long-term mutual funds. Data are seasonally adjusted. NOVEMBER/DECEMBER 1994 26 Appendix 2 Levels of the Augmented Aggregate M2+ T h e ta b le p r e s e n ts le v e ls f o r t h e a u g m e n te d a g g re g a te , M 2 + , as w e l l as M 2 . M 2 + w a s c o n s t r u c t e d u s in g t h e f o r m u la s i n T a b le 4 a n d s e a s o n a lly a d ju s te d as d e s c r ib e d i n t h e t e x t. Month M2+ M2 Month M2+ M2 Month M2+ M2 1959 1 299.3 286.6 1964 1 419.2 395.4 1969 1 615.5 569.3 1959 2 300.5 287.7 1964 2 421.6 397.6 1969 2 617.2 571.5 1959 3 302.0 289.0 1964 3 424.0 399.5 1969 3 618.9 573.9 1959 4 303.4 290.1 1964 4 426.3 401.7 1969 4 621.0 575.9 1959 5 305.9 292.2 1964 5 429.0 404.2 1969 5 622.2 576.5 1959 6 307.8 293.9 1964 6 432.2 406.8 1969 6 623.3 578.4 1959 7 309.6 295.3 1964 7 436.4 410.1 1969 7 623.2 579.8 1959 8 310.8 296.4 1964 8 439.7 413.3 1969 8 623.5 580.5 1959 9 310.7 296.4 1964 9 443.1 416.5 1969 9 625.5 582.2 1959 10 310.9 296.5 1964 10 446.1 419.2 1969 10 627.8 583.8 1959 11 311.5 297.1 1964 11 448.9 422.2 1969 11 630.8 586.9 1959 12 312.1 297.7 1964 12 451.4 424.7 1969 12 631.9 589.5 1960 1 312.4 298.2 1965 1 454.7 427.8 1970 1 631.1 591.1 1960 2 312.4 298.5 1965 2 457.8 430.4 1970 2 628.8 588.5 1960 3 313.2 299.2 1965 3 460.5 433.0 1970 3 630.9 589.3 1960 4 314.1 300.0 1965 4 463.1 435.5 1970 4 629.0 590.4 1960 5 315.2 300.9 1965 5 465.0 437.0 1970 5 628.8 593.5 1960 6 316.9 302.1 1965 6 467.5 439.8 1970 6 631.0 597.1 1960 7 319.2 304.2 1965 7 470.7 442.8 1970 7 635.8 600.4 1960 8 321.9 306.8 1965 8 474.3 445.6 1970 8 643.0 606.1 1960 9 323.3 308.2 1965 9 478.6 449.1 1970 9 650.2 612.4 1960 10 324.6 309.5 1965 10 483.1 452.7 1970 10 656.0 617.6 1960 11 326.1 311.0 1965 11 486.8 455.9 1970 11 660.8 622.3 1960 12 327.7 312.3 1965 12 490.7 459.3 1970 12 668.1 628.1 1961 1 330.3 314.1 1966 1 494.2 462.3 1971 1 675.5 634.0 1961 2 333.5 316.6 1966 2 496.6 464.5 1971 2 685.8 642.6 1961 3 335.6 318.1 1966 3 499.1 466.9 1971 3 695.9 651.2 1961 4 337.9 319.9 1966 4 501.5 469.3 1971 4 706.5 660.5 1961 5 340.8 322.2 1966 5 501.7 469.9 1971 5 715.1 668.7 1961 6 342.9 324.1 1966 6 502.4 470.9 1971 6 720.8 674.8 1961 7 344.7 325.6 1966 7 502.8 470.8 1971 7 727.6 681.3 1961 8 347.1 327.5 1966 8 503.2 472.6 1971 8 733.5 687.4 1961 9 349.1 329.3 1966 9 504.4 475.2 1971 9 740.5 694.4 1961 10 351.5 331.2 1966 10 505.4 476.0 1971 10 745.6 700.4 1961 11 354.3 333.4 1966 11 507.6 477.4 1971 11 750.7 706.9 1961 12 356.4 335.4 1966 12 510.6 479.9 1971 12 757.9 712.6 FEDERAL RESERVE BANK OF ST. LOUIS 27 M2+ Month M2 Month M2+ M2 Month M2+ M2 1972 1 766.3 719.6 1975 1 940.1 912.2 1978 1 1328 1296.2 1972 2 776.1 728.1 1975 2 949.6 920.1 1978 2 1332.2 1301.3 1972 3 784.6 735.8 1975 3 962.0 931.3 1978 3 1341.4 1308.6 1972 4 789.8 741.2 1975 4 973.4 941.5 1978 4 1350.7 1317.1 1972 5 795.3 745.9 1975 5 987.4 954.1 1978 5 1360.7 1326 1972 6 801.9 752.6 1975 6 1003.5 969.2 1978 6 1367.6 1333.3 1972 7 811.3 762.3 1975 7 1015.1 981.0 1978 7 1376.3 1341.8 1972 8 820.9 771.8 1975 8 1023 989.7 1978 8 1385.5 1350 1972 9 829.4 780.9 1975 9 1030.6 998.5 1978 9 1399 1363.3 1972 10 838.3 789.5 1975 10 1037.5 1004.7 1978 10 1407.2 1372.6 1972 11 846.2 796.6 1975 11 1048.3 1014.8 1978 11 1412.7 1379.5 1056.5 1023.2 1972 12 855.7 805.1 1975 12 1978 12 1421.8 1388.6 1973 1 861.7 813.5 1976 1 1068.2 1033.9 1979 1 1429 1395.2 1973 2 863.1 817.8 1976 2 1083.2 1047.8 1979 2 1435.7 1401.8 1412.2 1973 3 862.3 818.7 1976 3 1092.6 1057 1979 3 1446.3 1973 4 865.1 823.2 1976 4 1103.6 1068.4 1979 4 1460.6 1426.5 1973 5 870.6 830.3 1976 5 1116.3 1081.8 1979 5 1467.8 1434 1973 6 876.7 837.2 1976 6 1120.9 1086.2 1979 6 1482.2 1448.4 1973 7 882.0 841.0 1976 7 1129.4 1095.2 1979 7 1495.1 1460.9 1973 8 885.5 843.6 1976 8 1143 1109 1979 8 1505.9 1471 1973 9 886.6 844.6 1976 9 1155.6 1120.9 1979 9 1517.7 1482.8 1973 10 892.1 848.5 1976 10 1171.4 1136 1979 10 1522.4 1488.4 1973 11 894.9 854.4 1976 11 1183.8 1148.7 1979 11 1523.9 1490.5 1973 12 899.0 860.9 1976 12 1199.4 1163.6 1979 12 1530.7 1497 1974 1 902.7 865.4 1977 1 1212.6 1177.1 1980 1 1542.3 1507.8 1974 2 906.6 870.3 1977 2 1222.5 1188.3 1980 1555.8 1520.7 1974 3 911.9 876.5 1977 3 1233.3 1199.4 1980 3 1559.8 1527.3 1974 4 913.2 879.1 1977 4 1244.7 1211.2 1980 4 1554.9 1524.1 1974 5 914.1 881.2 1977 5 1254.9 1221.5 1980 5 1565.2 1532.7 1974 6 916.2 884.6 1977 6 1263.9 1230.1 1980 6 1586.6 1552.5 1974 7 918.5 888.0 1977 7 1275 1240.9 1980 7 1609.2 1573.6 1974 8 919.8 891.1 1977 8 1284.1 1250.3 1980 8 1627.9 1590.3 1974 9 921.2 895.1 1977 9 1293.9 1260.3 1980 9 1642.4 1604.5 1974 10 927.3 900.2 1977 10 1302.8 1269.9 1980 10 1654.9 1616.8 1974 11 933.7 905.4 1977 11 1311.2 1278.2 1980 11 1667 1628.9 1974 12 935.9 908.4 1977 12 1320.8 1286.5 1980 12 1667.6 1629.3 NOVEMBER/DECEMBER 1994 28 Month M2+ M2 Month M2+ M2 Month M2+ M2 1981 1 1677.6 1640 1984 1 2278.5 2202.4 1987 1 3105.7 2836.8 1981 2 1689.8 1652 1984 2 2296.6 2221.2 1987 2 3121.6 2837 1981 3 1709.7 1670.8 1984 3 2311.4 2236.8 1987 3 3136.3 2841.4 1981 4 1730.5 1691.7 1984 4 2327.9 2252.8 1987 4 3152.7 2855.8 1981 5 1736.3 1697 1984 5 2341.4 2267.7 1987 5 3155.2 2860.7 1981 6 1745.4 1705.5 1984 6 2352.9 2279.8 1987 6 3159.6 2862.6 1981 7 1757.9 1718.7 1984 7 2364.7 2290.3 1987 7 3173.8 2868.8 1981 8 1773.1 1734.9 1984 8 2378.8 2299.9 1987 8 3195.2 2882.7 1981 9 1782 1745.8 1984 9 2399.7 2316.3 1987 9 3211.9 2898 1981 10 1797.1 1760.5 1984 10 2413.6 2328.7 1987 10 3207.4 2913.8 1981 11 1814.7 1776.8 1984 11 2438.1 2353.1 1987 11 3189.5 2915.8 1981 12 1831.4 1793.3 1984 12 2464.1 2377.8 1987 12 3195.4 2920.1 1982 1 1850.3 1812.6 1985 1 2495.5 2403.5 1988 1 3223.4 2944.6 1982 2 1853.2 1815.9 1985 2 2524.4 2427.2 1988 2 3245 2963.3 1982 3 1865.4 1828.5 1985 3 2536.6 2438.7 1988 3 3262.3 2981.1 1982 4 1879.8 1842.6 1985 4 2541.8 2442.2 1988 4 3282.5 3003.7 1985 5 2564.4 3021.5 1982 5 1892.2 1854.4 2457.8 1988 5 3299.4 1982 6 1902.6 1865.3 1985 6 2597.9 2484.2 1988 6 3312.3 3033.9 1982 7 1914.4 1877.1 1985 7 2619.8 2500.4 1988 7 3319.9 3041.2 1982 8 1936.2 1895.8 1985 8 2643.1 2518 1988 8 3319.2 3044.6 1982 9 1954.2 1910.7 1985 9 2661.9 2533.3 1988 9 3325.7 3047.6 1982 10 1972.8 1926 1985 10 2678.9 2543.8 1988 10 3341.5 3056.9 1982 11 1988.7 1939.2 1985 11 2702.6 2557.2 1988 11 3356.7 3073.6 1982 12 2004.3 1953.2 1985 12 2731.3 2575 1988 12 3363.5 3081.4 1983 1 2062.6 2009 1986 1 2744.3 2580 1989 1 3368.4 3085.5 1983 2 2103.9 2046.8 1986 2 2764.1 2590 1989 2 3365.6 3085.4 1983 3 2126.2 2066.1 1986 3 2798.2 2611.2 1989 3 3371.7 3093 1983 4 2146 2082.4 1986 4 2833.7 2637.6 1989 4 3380.3 3097.1 1983 5 2166.9 2099.6 1986 5 2868 2663.7 1989 5 3388.7 3099 1983 6 2181.4 2111.2 1986 6 2895.4 2684.5 1989 6 3409.5 3116.3 1983 7 2194.8 2123.6 1986 7 2926.6 2711.7 1989 7 3441.9 3143.4 1983 8 2203.1 2132.1 1986 8 2959 2734 1989 8 3466 3161.4 1983 9 2216.7 2144.2 1986 9 2987.4 2753.5 1989 9 3488.7 3178.9 1983 10 2238.9 2165.2 1986 10 3017.6 2777 1989 10 3509.5 3199 1983 11 2250.8 2177.1 1986 11 3043.6 2793 1989 11 3534.5 3218.9 1983 12 2263 2187.6 1986 12 3075.5 2818.2 1989 12 3561.9 3239.8 FEDERAL RESERVE BANK OF ST. LOUIS 29 Month M2+ M2 Month M2+ M2 1990 1 3563.4 3252.9 1993 1 4046.7 3502.8 1990 2 3567.3 3267.3 1993 2 4063.6 3494.2 1990 3 3582.7 3279.5 1993 3 4066.2 3494.8 1990 4 3594.4 3291.1 1993 4 4072.8 3498 1990 5 3600.4 3294 1993 5 4108.1 3521.9 1990 6 3619.8 3305.5 1993 6 4128.5 3528.7 1990 7 3631.5 3315.3 1993 7 4148.3 3533.7 1990 8 3640 3330.9 1993 8 4169.2 3536 1990 9 3644.6 3345 1993 9 4201.8 3544.2 1990 10 3644.1 3347.3 1993 10 4225.6 3547.8 1990 11 3648 3345.1 1993 11 4252.3 3560.1 1990 12 3662.8 3353 1993 12 4265.1 3567.6 1991 1 3676.6 3363.5 1994 1 4278.9 3572.8 1991 2 3700.7 3380 1994 2 4280 3568.9 1991 3 3730.3 3398.1 1994 3 4276.7 3582.9 1991 4 3750.9 3409 1994 4 4277.1 3589.9 1991 5 3767.6 3418.9 1994 5 4278 3590.9 1991 6 3778.8 3426.6 1994 6 4263.8 3581.3 1991 7 3783.3 3426.4 1991 8 3794.8 3427.4 1991 9 3810.7 3427.5 1991 10 3828.9 3432.3 1991 11 3850.8 3445.4 1991 12 3873.9 3455.3 1992 1 3892.4 3464.1 1992 2 3918 3483.6 1992 3 3927.4 3486.3 1992 4 3929.8 3481.9 1992 5 3936.3 3482.1 1992 6 3941.1 3477.8 1992 7 3956.8 3480.7 1992 8 3973.7 3489.4 1992 9 3991.7 3496.6 1992 10 4008.7 3507.5 1992 11 4022.5 3510.5 1992 12 4034.8 3509 NOVEMBER/DECEMBER 1994 31 Athanasios Orphanides, Brian Reid and David H. Small Athanasios Orphanides, Brian Reid and David H. Small are staff economists with the Board of Governors o f the Federal Reserve System, Division o f Monetary Affairs. The Empirical Properties of a Monetary Aggregate That Adds Bond and Stock Funds to M2 R THE PAST FEW YEARS, the unexpected slowdown of M2 growth and strength in its velocity have coincided with a surge in out standing balances in bond and stock mutual funds. Although this surge partly reflects capital gains from rising prices in stock and bond markets, it also is due to record inflows of new balances. Indeed, anecdotal and statistical evidence suggests that a significant portion of these inflows have come from M2, contributing to the unexpected weakness in this aggregate. Consequently, aug menting M2 with bond and stock funds in order to internalize these flows offers the prospect of an aggregate that is more stably related to income and prices than M2 has been of late. This article examines empirical issues associ ated with whether such an augmented aggregate, called M2+, would be useful in the conduct of monetary policy. The three main issues examined are the stability of its demand function, its infor mation content as an indicator of spending, and its controllability. To assess these characteristics of M2+, demand functions and reduced-form relationships between M2+ and income have been estimated. The M2+ series used in this empirical work includes capital gains and losses on bond and stock funds. M2+ was defined in this manner on the presumption that meaningful behavioral distinctions cannot be made between balances generated in the past by capital gains and balances originally brought into stock and bond funds from outside sources. Specifying a demand function for M2+ requires identifying its close substitutes. One broad class of substitutes includes real assets such as com modities and durable goods, while alternative financial assets may include direct holdings of short-term market instruments, bonds and stocks. The estimated demand functions posit direct holdings of Treasury bonds and bills as the sole substitutes for M2+ because of difficulties in finding plausible measures of expected ex ante returns on real assets and equities. Using rough proxies for ex ante financial returns, including those on stock and bond funds themselves, the estimated demand functions have reasonable behavioral properties. But perhaps because of the problems with measuring ex ante returns, as well as the difficulty in distinguishing between returns on direct and mutual-fund holdings of the same assets, the estimated demand functions are not very stable. Hence, the usefulness of these equa tions in interpreting and forecasting movements in M2+ may prove to be limited. Moreover, these relationships have been estimated over a period of major innovation and growth of the bond and stock fund industry, raising further questions about the stability of the specification. NOVEMBER/DECEMBER 1994 32 In terms of information content, M2+ and M2 do not appear to have differed significantly. The velocity of M2+ may have been more “sensible” than that of M2 over the past three years, given the anomalous behavior of V2, which was rising while the opportunity cost of M2 was declining. As a consequence, in reduced-form relations, forecasts of nominal GDP growth for 1992 are somewhat stronger and more accurate when based on M2+ rather than M2. But leading into the past recession and continuing to the end of 1991, both aggregates yielded substantial overpredictions of nominal GDP growth. Within M2+, the volatile monthly swings in M2 are modestly offset by changes in net inflows to stock and bond funds, but capital gains and losses cause growth of M2+ to be more volatile than that of M2. Moreover, the capital gains and losses in M2+ may cause movements in the aggregate that neither reflect shifts in the stance of monetary policy nor provide appropriate signals for changes in policy. The remainder of the paper addresses these issues in more detail. Section two explains why the inclusion of capital gains and losses in M2+ seems necessary and highlights some of the problems that result from their inclusion. The section also briefly describes the data used to create a bond and stock mutual fund series that is comparable to M 2.1 Section three provides a broad empirical overview of the growth of bond and stock mutual funds over the past decade. The section then goes on to describe the recent behavior of M2+ and compares its behavior with that of M2. In the final section, we conduct an econometric investigation of M2+ demand, noting several of the difficulties associated with speci fying a demand function for M2+ and with the stability and controlability of the aggregate. Also in this section, the indicator properties of M2+ are formally examined and contrasted with those of M2. DEFINING M2+ AND ASSOCIATED DATA LIMITATIONS Because stock and bond funds are revalued daily to reflect realized capital gains and losses (unlike the deposit components of the current 1 For a more complete discussion of the issues associated with constructing the bond and stock fund component of M2+, see Collins and Edwards (1994). 2 Defining an aggregate by adding only bond funds to M2 would mitigate some of the problems associated with capital gains because such gains are more volatile for stock funds than for bond funds. However, such an aggregate suffers FEDERAL RESERVE BANK OF ST. LOUIS monetary aggregates whose values are fixed at par), questions arise as to the proper treatment of these changes in value. One option is to exclude capital gains by selecting a value of outstanding balances on a specific historical date to which subsequent inflows (net of capital gains) are added. However, both the selection of this date and the assumption that net capital gains before this date, but not after, constitute money are arbitrary. Presumably after some period of time, mutual fund shareholders cease to distinguish between balances generated by capital gains and those that stem from new investments of funds. A second problem with excluding net capital gains is that, at least conceptually, M2+ balances (or at least the mutual fund component) could become negative. For example, if an individual starts with no M2+ and then adds $100 to a bond fund, the M2+ holdings would then equal $100. If capital gains add $50 to the value of the bond fund account, M2+ will still equal $100 if capital gains are excluded from M2+. If the individual then withdraws all $150 from the bond fund, net inflows would equal minus $150, and M2+ would equal minus $50 for this individual. To avoid such problems, M2+ is defined to include capital gains.2 But the inclusion of capital gains and losses raises issues about the signals that M2+ can convey about the stance and proper course of monetary policy. For example, a decline in the stock market that reflected lower profit expecta tions and slower business activity would lower M2+, ceteris paribus, and may appropriately be calling for a more stimulative monetary policy. But, if capital losses and lower M2+ result from an increase in long-term interest rates in response to rising inflation, a monetary ease is not likely to be the appropriate response. With bond and stock funds currently amounting to about 15 percent of M2+, and likely to increase further, capital gains and losses can have noticeable short-run effects on the growth of the expanded aggregate. Data used to measure bond and stock fund balances are provided to the Federal Reserve by from the data limitation that a large number of mutual funds invest in both bonds and stocks, and therefore separating mutual funds into these two categories is problematic. An aggregate consisting of M2 plus bond funds has been exam ined by Duca (forthcoming). 33 the Investment Company Institute (ICI). As received from ICI, the bond and stock fund balances include IRA/Keogh and institutional holdings.3 Because IRA/Keogh and institutional holdings of M2-type accounts are netted from M2, such holdings of bond and stock funds need to be netted out in constructing the bond and stock fund component of M2+. Such netted bond and stock fund series will be referred to as M 2-type series. For M2-type bond and stock funds, it would be useful to have data on inflows excluding capital gains as a direct measure of portfolio shifts, but this series is not reported by ICI and cannot be constructed from available data.4 To proxy for this missing series, we rely on total inflows excluding capital gains.5 This growth may be attributed, in part, to declining transaction costs when investing in mutual funds. Between 1970 and 1992, load fees on mutual funds fell from an average of 8.5 percent to 4.5 percent. In addition, during the past decade, no-load mutual funds have become more widely available, further reducing the transaction costs involved in shifting in and out of mutual funds.8 Historical Trends o f Bond and Stock Mutual Funds In reducing transaction costs associated with moving into mutual funds, banks have also been playing a role. Large domestic banks sampled for the Federal Reserve System’s March 1993 Senior Financial Officer Survey reported significant increases during the past few years in their overall sales staffs for mutual funds, as well as in the share of branch offices with sales representatives located at them.9 These developments suggest that investment balances that previously had been held in M2 can now be moved more readily into bond and stock funds, and may be more likely to be switched back and forth between these two lodgings as relative yields shift. Although bond and stock mutual funds have existed in the United States since 1924, most of their growth has been quite recent. In the early 1970s, there were about 400 such funds with total assets of about $40 billion. Today, there are more than 3,000 funds with total assets of about $1.4 trillion (see Figure 1, top panel).6 The bottom panel of Figure 1 shows that holdings of aggregate bond and stock mutual funds have grown roughly in parallel and are about equal.7 Another innovation that may have contributed to the growth of the industry is that the balances held in such funds can be used more readily as a means of payment in the purchase of goods and services, in part because many funds now allow withdrawals to be made by checks. However, the writing of a check means that mutual fund shares must be sold, so there is a potential capital gain or loss associated with each check. Anecdotal evidence suggests that the inconvenience of GROWTH OF THE MUTUAL FUND INDUSTRY AND THE BROAD EMPIRI CAL PROPERTIES OF M2+ 3 See ICI (1994, Appendix A) for the type of funds classified as stock and bond funds. 4 Inflows to M2-type bond and stock funds can be estimated by starting with the change in the market value of M2-type bond and stock funds and subtracting estimated capital gains. These capital gains and losses can be estimated from capital gains and losses on total bond and stock funds, which are available. We have made these estimates by tak ing total capital gains and losses and multiplying them by the lagged ratio of M2-type outstandings to total outstandings of bond and stock funds. Although we show such estimates of M2-type capital gains and losses in Figure 10, these estimates and the associated estimates of M2-type inflows do not seem reli able enough to use throughout the analysis. 5 Reinvested earnings are included in both the M2-type and total inflows. This treatment is conceptually similar to the interest crediting of M2 balances. According to ICI data, about 75 percent of shareholders automatically reinvest their earnings in the mutual fund. 6 For an overview of the growth of the mutual fund industry and how mutual funds operate and compete, see Sirri and Tufano (1993). types of funds offered, and much of the growth has occurred in specialized funds. For example, municipal bond funds, which were introduced in 1976, had total assets of $50 bil lion by 1985, and had grown to $225 billion by 1993. Government income and Ginnie Mae funds, which had a total of about $2.5 billion under management at the start of 1984, had $175 billion by 1993. 8 See Mack (1993) for a more extended discussion of the decline in load fees. 9 The percentage of sampled banks with more than 50 repre sentatives selling retail mutual funds increased from less than 10 percent three years ago to more than 40 percent currently; and nearly all sampled banks currently have some retail sales force. Also, banks have made mutual funds more accessible through their branches. Over half the sam pled banks have personnel selling retail mutual funds on a part-time basis or by appointment at 90 percent or more of their branches, a significant increase from three years ago, when only 20 percent of the banks had sales representa tives available part-time at 90 percent or more of their branches. The results of the March 1993 Senior Financial Officer Survey and the growth of bank-related mutual funds more generally is discussed by Reid and Small (1993). 7 This growth of the bond and stock fund industry has been associated with increased diversity and specialization of the NOVEMBER/DECEMBER 1994 34 Figure 1A Number and Market Value of Bond and Stock Mutual Funds Number of funds Billions of dollars - 600 - 400 - 200 - 100 - 50 Figure 1B Market Value of Aggregate Bond and Stock Mutual Funds Billions of dollars Aggregate bond and stock mutual fund balances include IRA/Keogh balances and institutional holdings. These aggregate mutual fund data are not seasonally adjusted. FEDERAL RESERVE BANK OF ST. LOUIS 35 tabulating capital gains and losses for the purpose of income taxes significantly limits the frequency of the writing of checks on stock and bond fund accounts.10 The transaction activity in bond and stock funds is presented in Figure 2, which plots gross outflows from total bond and stock funds measured as a share of outstanding balances.11 Contrary to what would be expected if transaction activity had increased, there is no discernible secular trend in the ratio of gross outflows to the value of shares, and the most recent value of the ratio seems rather modest.12 Consequently, the rationale that we explore for adding such funds to M2 is based on the substitutability of bond and stock funds for small time deposits or other M2 balances as savings vehicles, and not on the transactability of bond and stock funds. R ecent Behavior ofM 2+ Figure 3 displays the levels of M2 and M2+, with M2+ defined to include capital gains and losses, but excluding IRA/Keogh accounts and institutional holdings. The upper panel of Figure 4 shows the GDP velocities of the two aggregates and the bottom panel shows the growth rates of M2 and M2+. Three distinct episodes are evident: (1) From 1984 through 1986, the gap between the two velocity measures increased as growth in M2+ outpaced that of M2 (lower panel); (2) From 1986 through 1990, the velocities moved roughly parallel to each other; (3) From 1990 to the present, the velocity of M2 rose sharply while that of M2+ is about unchanged. changes in the value of bond and stock funds and changes in M2— such as from 1990:Q4-1991:Q3. As a consequence, M2+ is slightly more variable than M2, including the most recent years. In terms of deviations from trend, and as shown in the top panel of Table 1, the mean absolute deviation of M2+ growth over the period from March 1984 through September 1993 is 3.07 percentage points — somewhat greater than that for M2 growth at 2.46 percentage points. However, during the recent sub-sample of January 1989 to September 1993, the difference in variability is much smaller. The middle panel of the table indicates that these differences persist using quarterly data. The bottom panel of the table examines the variability of velocity growth about its mean. The variability of V2+ has been close to that of V2 in recent years, but V2+ was con siderably more variable in the mid-1980s. MODELING BOND AND STOCK FUNDS AND M2+ Modeling the demand for any monetary asset requires identifying alternative uses for the balances held in the aggregate. Bond and stock funds compete with M2 for balances, and a potential advantage of considering M2+ is that this competition need not be addressed in mod eling M2+. But bond and stock funds compete with many more assets than just M2— such as commodities or other real assets. Because rea sonable proxies for the expected rates of return on real assets are not available, these assets will not be included in the models developed below. Figure 5 shows the monthly growth rates of M2 and M2+. As shown in the top panel, much of the monthly variability of M2 growth shows through to M2+ growth. Monthly net inflows to stock and bond funds apparently are negatively correlated with changes in M2, but offset only a small portion of the change in M2, as shown in the middle panel.13 However, as shown in the bottom panel, capital gains and losses can induce positive co-movements between the The only alternatives to M2+ that are included in the models below are direct holdings of Treasury bonds and bills. As a result, analyzing M2+ entails the measurement of own-rates that determine the flows among the following five assets: M2, bond funds, stock funds, and direct holdings both of bonds and bills. The own-rate for M2 can be calculated from posted rates on M2 deposits. Proxies for ex ante returns on bond and stock funds are examined immediately below. 10 Although mutual funds could compute the capital gains on the checks written by each shareholder, this does not seem to be a widespread practice as yet. See Clements (1993). savings accounts is 4.7. See Collins and Edwards (1994) for a further discussion. 11 Gross outflows are the value of all balances withdrawn, whether or not the balances are reinvested in another fund. 13 The modest offset is only apparent because the inflows shown are aggregate inflows that include flows to IRA/Keogh and institutional balances. 12 As shown in the figure, recently this ratio has averaged about 2.5 percent, which at an annual rate yields a turnover rate of 0.3. In comparison, the turnover ratio for traditional NOVEMBER/DECEMBER 1994 36 Figure 2 Aggregate Bond and Stock Mutual Fund Outflows (as a percentage of total market value) Percent Aggregate bond and stock mutual fund outflows include flows from IRA/Keogh accounts and institutional holdings. Figure 3 M2 and M2+ Billions of Dollars FEDERAL RESERVE BANK OF ST. LOUIS 37 Figure 4A Velocities of M2 and M2+ Figure 4B Growth Rates of M2 and M2+ Percent NOVEMBER/DECEMBER 1994 38 Figure 5 Variability of M2 and M2+ Billions of dollars Billions of dollars Billions of dollars * Aggregate bond and stock mutual fund net inflows include flows to IRA/Keogh accounts and institutional holdings. FEDERAL RESERVE BANK OF ST. LOUIS 39 Table 1 Variability of Money and Velocity About Trend* (percent annual rates) Monthly money growth Mean absolute deviation Sample period Standard deviation M2 M2+ M2 M2+ 2.46 3.07 3.04 3.81 1984:M3 - 1988:M12 2.72 3.37 3.25 4.15 1989:M1 - 1993:M9 2.23 2.38 2.82 3.05 1984:M3 - 1993:M9 Quarterly money growth Mean absolute deviation Standard deviation Sample period M2 M2+ M2 1984:Q3 - 1993:Q3 1.80 2.43 2.10 M2+ 2.91 1984:Q3 - 1988:Q4 1.99 2.46 2.33 3.09 1989:Q1 - 1993:Q3 1.56 1.80 1.88 2.15 Quarterly velocity growth Mean absolute Standard deviation deviation Sample period M2 M2+ M2 M2+ 1984:Q3 - 1993:Q3 2.89 3.40 3.60 4.25 1984:Q3 - 1988:Q4 3.04 4.06 3.91 5.12 1989:Q1 - 1993:Q3 2.45 2.61 2.98 3.10 * For money growth rates, regressions were used to esti mate separate time trends for each sub-period. The time trends pick up the general slowing of money growth since 1984. For velocity growth, for each sub-period the summary statistics refer to the growth of velocity about its mean for that sub-period. No trend in the growth of velocity (that is, acceleration of velocity) was removed before calculating the summary statistics. Some of the additional variability in M2+ may be because the monthly figures for mutual funds represent averages of end-of-month figures, while M2 data represent averages of daily data. Proxies fo r E x Ante Returns and their Effects on Changes in M2-Type Bond and Stock Funds This sub-section attempts to identify determi nants of the demands for bond and stock mutual 14 Increases in these proxies would be expected to increase demands for the relevant type of mutual funds, but may also raise demands for direct holdings. 15 Decreases in aggregate bond and stock fund inflows in 1987 were related in part to changes in the tax exemption for IRA contributions, which were liberalized in 1981 and tightened in 1987. The latter change in the tax law does not seem to be the dominant reason for the dropoff in aggregate inflows in 1987. Net inflows went from $144 billion in 1986 to $49 billion in 1987, but data on IRAs indicate that a drop in their flows account for no more than $7 billion of this slowing. funds. In particular, proxies are needed for the ex ante returns on mutual funds. For bond funds, two alternative proxies are available. One approach relies on contempora neously observable market rates and the other employs lags of ex post realized returns.14 The first approach uses the slope of the Treasury yield curve and recent changes in the Treasury bond rate. The yield curve captures the difference between quoted yields to maturity on long-term bonds and posted returns on short-term assets including small time deposits or money market mutual funds (MMMFs) in M2 as well as money market instruments. As such, increases in the spread will make bond mutual funds (as well as direct holdings) look more attractive. Recent bond rate changes may affect investors’ expectations of prospective capital gains or losses if investors have extrapolative expectations. The top panel of Figure 6 shows that both net aggregate inflows to bond funds and the change in market value of M2-type bond funds surged from 1984 through the first quarter of 1987. This surge may well have reflected the relatively steep yield curve going into that period (middle panel) and capital gains caused by falling bond rates throughout most of the period (lower panel).15 But then with the upturn in bond rates in the second quarter of 1987, bond funds initially experienced modest average outflows during the last half of 1987, and then grew moderately over the next two years as long rates leveled off and the slope of the yield curve fell because of rising short-term rates. More recently, aggregate bond inflows surged as bond rates started to drift down again (although more slowly than during the 1984-1986 period) and as the slope of the yield curve rose to record heights. A quarterly regression that examines the effects of the slope of the yield curve and recent capital gains and losses is given in equation 1. The change in market value of M2-type bond funds is scaled by the lagged value of M 2+.16 16 Bond funds are scaled by M2+ to avoid placing undue weight on the early years of the regression. Such overem phasis could result if bond fund changes were scaled by lagged bond funds (that is, if we modeled the growth rate of bond funds) because during the early years when bond funds were modest, a small shift from M2 would represent a large percentage change in bond funds. NOVEMBER/DECEMBER 1994 40 Figure 6 Changes in Bond Fund Balances Billions of dollars Spread Between the 30-Year Treasury Bond and the 3-Month Treasury Bill Rates Percentage points 30-Year Treasury Bond Rate Percent * Aggregate bond fund net inflows include flows to IRA/Keogh accounts and institutional holdings. FEDERAL RESERVE BANK OF ST. LOUIS 41 (1) ABFUND = .00019 + .000025 TIME M 2+ , (.25) (.75) + .00047 (RT30Y - RTBE)_, (1.97) - .0022 ART5Y .0014 ART5Y, (3.71) (2.28) four-quarter moving average of the realized returns is given in equation 2.17 (2) ABFUND = .0012 M 2+4 (2.03) - .00036 (RTBE - RETBND4)_1 (7.47) - .0017 ART5Y, - .0018 ART5Y,,. (3.04) (3.14) R2 = .735 R 2 = .676 D.W. statistic = 1.22 Estimation period: 1985:Q2 - 1993:Q3, where: ABFUND is the change in the market value of M2-type bond funds; RT30Y is the 30-year Treasury bond rate; RT5Y is the five-year Treasury note rate; RTBE is the three-month Treasury bill rate; and the absolute values of t-statistics are in parentheses. The positive coefficient on the spread vari able indicates that a steep yield curve draws balances into bond mutual funds. The negative coefficients on the changes in the five-year Treasury note show that recent capital losses reduce bond fund inflows. The contemporaneous change in the note rate also captures the direct effect of changes in the market values. While the top panel of Figure 7 repeats that of Figure 6, the lower panel of 7 uses the realized returns on bond funds to proxy for expected returns. Comparisons across the two panels show a rough relation between the changes in bond fund balances and the spread of realized returns over Treasury bill rates. A regression relating the change in market value of M2-type bond funds to the opportunity cost that uses the .000098 TIME (.34) - .0017 ART5Y. (3.14) R2 = .758 R 2 = .733 D.W. statistic = 1.33 Estimation period = 1985:Q2 —1993:Q3, where: ABFUND is the change in market value of M2-type bond funds; RETBND4 is a four-quarter moving average of the ex post realized return on bond funds; RTBE is the three-month Treasury rate; and the absolute values of t-statistics are in parentheses. These results are similar to those reported in equation 1, with both equations having R2s of about .75. For stock funds, the upper panel of Figure 8 shows two measures of changes in balances and the lower shows the spread of the realized returns over the Treasury bill rate. Comparing the two panels, the surge in balances prior to the stock market crash of 1987, the subsequent falloff, and then the resurgence in balances in 1989 are all tracked by the movements in realized returns minus the Treasury bill rate. The latest spurt in balances seems anomalous, but some of this rise is lessened when the changes in market values are scaled by M2+ as in the following regression, which explains the change in the market value of M2-type stock funds: 17 The four-quarter moving average is used for realized returns to save on the number of freely estimated parameters, in comparison to using four lags of realized returns. When the spread between the quarterly realized return and the Treasury bill rate is entered with four lags, the four individual coefficients are roughly of the same magnitude and the sum of the coefficients is statistically insignificantly different from the estimated coefficient on the four-quarter moving average reported in equation 2. Also, the realized return is entered in the form of the spread of the Treasury bill rate over the realized return in keeping with standard usage of entering opportunity costs. NOVEMBER/DECEMBER 1994 42 Figure 7A Changes in Bond Fund Balances Billions of dollars Figure 7B Realized Returns on Bond Funds Minus the 3-Month T-Bill Rate Percent * Aggregate bond fund net inflows include flows to IRA/Keogh accounts and institutional holdings. FEDERAL RESERVE BANK OF ST. LOUIS 43 Figure 8A Changes in Stock Fund Balances Billions of dollars Figure 8B Realized Returns on Stock Funds Minus the 3-Month T-Bill Rate Percent * Aggregate bond fund net inflows include flows to IRA/Keogh accounts and institutional holdings. NOVEMBER/DECEMBER 1994 44 (3) ASFUND = - .00093 + .000072 TIME M2+.J (1.68) (2.96) - .000054 (RTBE - RETEQ4).! (2.59) + .00017 Alog(NYSE). (6.98) R2 = .661 R 2 = .628 D.W. statistic = 2.15 Estimation period = 1985:Q2 - 1993:Q3, where: ASFUND is the change in market value of M2-type stock funds; RETEQ4 is a four-quarter moving average of the ex post realized return on stock funds; NYSE is the New York Stock Exchange price index; and the absolute values of t-statistics are in parentheses. As with the bond inflow equations, the spread of the three-month Treasury bill rate over the realized return has a significant, negative impact. Moreover, the overall fit of the equation is quite high, as the capital gains reflected in the con temporaneous change of stock prices explain most of the variation in the market value of stock fund balances. An unattractive aspect of equations 1-3 is the absence of a scale variable such as income or wealth and, consequently, the absence of well-defined, long-run equilibrium levels of bond and stock funds relative to income or wealth, or even to M2+. Specifications of such relations were unsuccessful, apparently because innovation and growth of the bond and stock fund industry have caused bond and stock fund balances to grow relative to other nominal quantities. As a result, in what follows we do not attempt to provide a model of the components of M2+ but only of M2+ as a whole, which can be modeled with such a scale variable. 18 These are error-correction models. To derive the long-run properties of the model, let nominal GDP and all interest rates be constant. Then M2+ will also be constant and the left-hand side of the regression model can be set to zero. The logarithm of velocity can then be solved for as a linear function of the opportunity costs and the Treasury yield curve spread. 19 Recent models that estimate significant yield-curve effects on M2 demand were developed by Hess(1990). Subsequently, Feinman and Porter (1992) found the same effect, and Mehra (1992) obtained similar results. FEDERAL RESERVE BANK OF ST. LOUIS M odeling the D em and fo r M2+ This section presents and evaluates simple models of the demand for M2+. The models are similar to earlier specifications for the demand for M2 that posit a long-run demand in which velocity is a linear function of opportunity costs. The regression models are of the general form:18 Alog(M2+) = cO + c l Time + c2 log(M2+/GDP).1 + c3 (opportunity cost variables).! + c4 (RT30Y - RTBE).! + error. Opportunity costs are defined as the difference between the yield on the three-month Treasury bill and the own-rate of return on monetary assets. For own-rates we use the own-rate on M2 (RM2E), and as proxies for the own-rates of return on bond and stock mutual funds, we use the four-quarter moving averages of realized returns on bond and stock mutual funds (RETBND4 and RETEQ4), as in equations 2 and 3. Because we assume direct holdings of bonds and bills are the primary com peting asset for M2+ balances, the opportunity cost variables all incorporate Treasury rates as the competing rate. For this, the three-month Treasury bill rate is used. Because longer-maturity Treasury securities may also be competing assets for M2+, especially for the M2 balances in M2+, all specifications considered below include the slope of the term structure. This variable has been found to have a significant, negative impact on M2 in recent years and above we found that it had a positive impact on bond-fund inflows.19 If the substitution away from M2 that this variable captures is towards bond and stock funds only, there would be no net impact on M2+.20 Indeed, it is the internalization of just such substitutions that makes M2+ a potentially useful aggregate. However, if the slope of the term structure cap tures some substitution of M2 into direct hold ings of stocks and bonds, then it would have a negative impact on M2+. 20 Indeed, such substitution is consistent with the findings of equation 1, which showed a positive impact of the yield curve on bond fund balances. 45 Table 2 Estimated Demand Equations for M2+ (2) -0.031 (2.85) (3) -0.026 (2.61) Constant (1) -0.0038 (.30) Time -0.00066 (6.67) log(M2+/GDP2)_, -0.069 (2.24) -0.232 (6.36) -0.215 (6.65) S30Y , -0.000079 (.10) -0.0047 (5.76) -0.005 (6.86) OCM2+., -0.0000043 (4.31) OCM2., -0.014 (5.89) -0.013 (6.35) OCMB., -0.00020 (1.18) OCMS., -0.000025 (.35) -0.00069 (10.38) -0.00088 (13.07) W *OCMB., -0.0041 (2.23) W 'O C M S., -0.00037 (.50) R2 0.614 0.857 0.880 R2 0.566 0.828 0.856 D.W. statistic 1.56 1.69 1.88 Estimation period = 1984:Q3 through 1993:03, where: the dependent variable = log(M2+t) -log(M2+,)., S30Y = RT30Y -R T B E slope of the yield curve OCM2+ = RTBE -R M 2 + opportunity cost of M2+ OCM2 = R TBE--RM2 opportunity cost of M2 OCMB = RTBE-- RETBND4 opportunity cost of bond funds OCMS = RTBE-- RETEQ4 opportunity cost of stock funds w a proxy for the availability of mutual funds : NETMF/(M2 + NETMF) NETMF = Market value of M2-type bond and stock funds GDP2 = Two-quarter moving average of nominal GDP a n d a b s o lu te v a lu e s o f t-s ta tis tic s are in p a re n th e s e s . Within this general specification, we fit three alternative M2+ demand functions. In the first alternative (shown in column one of Table 2), we used the slope of the yield curve and one oppor tunity cost that incorporates a weighted-average own-rate on M2+.21 The regression in column one im plicitly contains the restriction that the responses of M2+ to the opportunity costs of the three components (M2, bond funds and stock funds) after being weighted by dollar shares are equal. To allow for different responses to changes in the individual opportunity costs, column two of Table 2 shows a regression in which the un weighted opportunity cost for each of the three components of M2+ (OCM2, OCMB and OCMS) is entered separately.22 As can be seen, the adjusted R2 rises significantly when the constraint is relaxed. However, the opportunity costs of bond and stock funds are statistically insignificant. As a final specification, an attempt is made to incorporate the effects of the increasing access of retail customers to bond and stock funds. In column three of Table 2, the bond and stock fund opportunity costs are weighted by the ratio of bond and stock fund balances to the value of M2+.23 As can be seen, the fit of the equation improves modestly with the incorporation of these weights. In addition, the statistical significance of the opportunity cost of bond funds improves noticeably. The opportunity cost of stock funds remains insignificant, which may not be surprising because the highly volatile ex post returns on stock funds may be little used by investors in forming expectations of future movements in the market. Stability ofM 2+ D em and and the Controllability o f M2+ The stability of the above estimated demand functions for M2+ is generally suspect for a 21 In this setting, the M2+ own-rate, RM2+, is constructed as the weighted average of RM2E, RETBND4 and RETEQ4, weighted by the quantities of M2, bond funds and stock funds held in the previous quarter relative to M2+. The opportunity cost of M2+ is defined as OCM2+ = RTBE - RM2+, where RTBE is the yield on three-month Treasury bills. 22 There are several reasons to believe that separating the two opportunity costs will produce better estimates. First, given the likely error in measuring expected ex ante own-rates for bond and stock mutual funds, the aggregation of those two rates with the better-measured M2 own-rate will contaminate the estimated response to all three own-rates, in general lowering the estimated coefficient from its true value. Second, and equally important, since the relevant alternative assets for stock and bond mutual funds outside M2+ are likely to include assets other than three-month Treasury bills, the opportunity cost relative to the three-month Treasury bill may be more important for M2 than for stock and bond mutual funds. 23 As a proxy for the availability of mutual fund accounts, these weights suffer from being determined in part by the other factors driving demand for M2+. A better, but unavailable, proxy would be one based on transaction costs. The weight grows from about 3 percent in 1984 to around 8 percent in 1988 and then to 15 percent currently. NOVEMBER/DECEMBER 1994 46 Figure 9 Growth of Actual and Simulated M2+ Percent The estimation period of 1986:Q3 through 1991:Q2 is marked off by the vertical lines. The simulation begins in 1985:Q1. number of reasons. First, 10 years is a very short sample period. Second, within that period there was considerable innovation in the bond and stock fund industry. The weights used in the model of column three of Table 2 are likely at best to capture only broadly the effects on M2+ of such innova tions. Third, there are the obvious limitations of the opportunity cost variables in terms of identifying competing rates of return when the aggregate is as broad as M2+ and in terms of proxying ex ante expected rates of return on bond and stock funds. Indeed, all three equations in Table 2 fail Chow tests, strongly suggesting a lack of stability. Under these tests, the estimation period is split into two sub-periods, and the estimation results over the two sub-periods are compared statistically. In a less formal check, the model of column three in Table 2 was estimated over a sub-period that starts in 1986:Q3 and ends in 1991:Q2, allowing for an examination of the model’s forecasting performance both before and after this period. The simulation shown in Figure 9 shows fairly FEDERAL RESERVE BANK OF ST. LOUIS substantial errors over the pre-estimation period, but reasonable accuracy starting by 1988 and carrying through nearly to the present. Figure 10 provides some evidence on the his torical contribution of capital gains and losses to the growth of M2+. These can be a guide to the size of the shocks to the growth of M2+ that might need to be offset if fairly close control of M2+ were taken seriously. Although these esti mates likely are somewhat imprecise, they do show the effects of the stock market crash of 1987 and major market swings. However, since stock and bond funds are becoming a larger proportion of M2+, the aggregate is probably becoming increasingly vulnerable to swings in capital gains and losses. THE INDICATOR PROPERTIES OF M2 AND M2+ As a final set of statistical comparisons of the behavior of M2 and M2+, the ability of these aggregates to predict changes in nominal GDP 47 Figure 10A Quarterly Growth of M2+ Accounted for by Capital Gains* Percentage Points Figure 10B Annual (Q4/Q4) Growth of M2+ Accounted for by Capital Gains* Capital gains over the year divided by M2+ of Q4 of previous year Percentage Points * Capital gains and losses for M2-type bond and stock funds are estimated taking capital gains and losses for all bond and stock funds and multiplying that series by the ratio of M2-type to total bond and stock outstandings — la g g e d o n e p e rio d — a n d ta k in g a tw o -m o n th m o v in g a v e ra g e . is examined. Each test regresses one-quarter nominal GDP growth on lagged nominal GDP growth, lagged growth in one of the monetary aggregates, and lagged changes in the three-month Treasury bill rate.24 The regression models are of the general form: Alog(GDPNt) = B0 + T n=1 fin Alog(GDPNtJ + r , 1=1 ?„ Alog(M,n) + r n=] An A(RTBE,n) + e t , where GDPN = nominal GDP, M = monetary aggregate, and RTBE = three-month Treasury bill. The top panel of Figure 11 presents in-sample measures of the statistical significance of the lagged money measures for a rolling 15-year estimation window. The measure of significance at an indicated date is the significance level when the estimation period includes 15 years of data ending at that date. By this measure, the two aggregates performed equally well in predicting 24 The observation for 1980:Q2, during which credit controls were imposed, was omitted from the data. NOVEMBER/DECEMBER 1994 48 Figure 11A Significance Levels for the Predictive Impact of M2 and M2+ on Nominal GDP Growth* (15-year window) Significance level Figure 11B Significance Levels for the Predictive Impact of M2 and M2+ on Nominal GDP Growth* (71/2-year window) Significance level * Likelihood of falsely rejecting the monetary variable from the regression. FEDERAL RESERVE BANK OF ST. LOUIS 49 Figure 12A One-Quarter-Ahead Forecast Errors for Nominal GDP Growth* (15-year window) Percentage points Figure 12B One-Quarter-Ahead Forecast Errors for Nominal GDP Growth* (71/2-year window) Percentage points * Forecast errors are calculated as actual minus predicted GDP growth. NOVEMBER/DECEMBER 1994 50 Figure 13A Four-Quarter-Ahead Forecast Errors for Nominal GDP Growth* (15-year window) Percentage points * Forecast errors are calculated as actual minus predicted GDP growth. Figure 13B Four-Quarter-Ahead Forecast Errors for Nominal GDP Growth* (71/2-year window) Percentage points * Forecast errors are calculated as actual minus predicted GDP growth. FEDERAL RESERVE BANK OF ST. LOUIS 51 nominal GDP growth until the mid-1980s. In 1986, inflows to bond and stock funds were heavy in response to falling interest rates, and in 1987 there were modest net outflows when interest rates reversed course early in the year and the stock market crashed late in the year. The resulting swings in M2+ growth were not subsequently reflected in nominal GDP growth and, thus, the rise in significance levels. In the past couple of years, the significance levels of the two aggregates moved closer together. The lower panel of the figure presents the statistical significance levels for the same speci fication using a shorter, seven-and-a-half-year, rolling estimation window. With a shorter estimation horizon, the ill behavior of M2 over the past few years becomes clear with the large significance level since 1991. This increase is not accompanied by the corresponding increase in the significance level of M2+.25 Figure 12 shows the results of out-of-sample experiments that compute one-period-ahead forecast errors based on the regressions used in the previous exercise. Again, the time horizons over which the forecasting equations are estimated are 15 and seven-and-a-half years prior to the forecast date. Figure 13 presents results for four-period-ahead forecasts. The forecast errors in both figures are calculated as actual minus forecasted growth— a positive error indicates an underprediction. Comparing the results for M2 and M2+, these figures indicate that the forecast errors for the two aggregates are about the same over the whole period shown, as would be expected given that both aggregates are marginally useful in explaining future nominal GDP growth in-sample. Both aggregates tended to overpredict GDP growth in late 1990. Reasonably strong money growth in the second half of 1989 was not consistent with the subsequent recession at the end of 1990. Both aggregates generally underpredicted nominal GDP growth in 1992, with M2+ forecast errors being smaller than those of M2. REFERENCES Clements, Jonathan. “Wait! Don’t Write That Mutual Fund Check,” Wall Street Journal (November 1993), p. C14. Collins, Sean, and Cheryl Edwards. “An Alternative Monetary Aggregate: M2 Plus Household Holdings of Bond and Equity Funds,” this Review (November/December 1994), pp. 7-29. Duca, John V. “Should Bond Funds Be Included in M2?” Journal o f Banking and Finance (forthcoming). Feinman, Joshua, and Richard D. Porter. “The Continuing Weakness in M2," Board of Governors of the Federal Reserve System, Finance and Economics Discussion Series Paper #209 (September 1992). Investment Company Institute. Trends in Mutual Fund Activity. October 1994. Mack, Phillip R. “Recent Trends in the Mutual Fund Industry,” Board of Governors of the Federal Reserve System, Federal Reserve Bulletin (November 1993), pp. 1001-12. Mehra, Yash. “Has M2 Demand Become Unstable?” Federal Reserve Bank of Richmond Economic Review (September/October 1992), pp. 27-35. Reid, Brian, and David H. Small. “Bank Involvement in the Mutual Fund Industry,” Board of Governors of the Federal Resen/e System, mimeo (September 1993). Sirri, Eirk K., and Peter Tufano. “Competition and Change in the Mutual Funds Industry,” in Samuel L. Hayes, ed., Financial Services: Perspectives and Challenges. Harvard Business School Press, 1993, pp. 181-214. 25 The results are little changed when interest rates are omitted from the equations, although significance levels are lower (money is more important) since money picks up some of the variability of nominal GDP growth explained by interest rates in the broader model. NOVEMBER/DECEMBER 1994 53 William A. Barnett and Ge Zhou William A. Barnett is professor of economics at Washington University, St. Louis. Ge Zhou recently received a doctorate in economics from Washington University, St. Louis. The authors benefitted from useful discussions with Athanasios Orphanides. Research on the project was partially supported by NSF grant SES 9223557. Commentary W, E ARE VERY PLEASED to be invited to comment on the new M2+ index, recently proposed in interesting papers by some Federal Reserve Board staff members (Collins and Edwards, 1994; and Orphanides, Reid and Small, 1994). The two papers presented at this conference by those Board staff members raise important and challenging questions that we believe should motivate much research in future years. In addition, we wish to commend those Board staff economists for their courage and integrity in pushing past barriers that have intimidated prior researchers and thereby pre cluded prior research on these difficult matters. The basic issue is whether riskiness of the investment rate of return on an asset is a charac teristic that rules out the possibility of an asset’s contribution to the economy’s liquidity. Oddly, that issue has largely precluded prior considera tion of risky assets as components of central bank monetary aggregates. Yet clearly the position is groundless. While it is clear that risky assets are not good candidates for legal means of payment, monetary aggregates now contain many assets that are not legal means of payment. It has long been recognized that currency and demand deposits provide much, but by no means all, of the economy’s monetary service flow. No one has suggested that bond or stock mutual funds should be made legal means of payment. In addition, stock and bond mutual funds currently are bundled by companies into packages of funds that include money market funds within the bundle. Hence, it often is as easy as a telephone call to transfer funds from stock and bond funds into checkable money market funds. Although stock and bond funds certainly should not be made legal means of payment, it simply makes no sense to exclude bond and stock mutual funds from consideration as assets contributing monetary liquidity to the economy. There is no necessary conflict between the existence of r i s k y r e tu r n and the c o n t r i b u t i o n of liquidity to the economy. The two are not mutually exclusive.1 Yet prior researchers have excluded assets having substantial principle risk from consideration as components of monetary aggregates. It indeed is odd that such an obviously groundless prejudice has precluded research by the entire economics profession on an important 1 Formally, the correct method used to determine the cluster ing of components within an aggregation-theoretic monetary aggregate is testing for blockwise weak separability. An innovative new approach to testing for weak separability was recently proposed by Swofford and Whitney (1994). Although risky return complicates testing for weak separability, risk in no way precludes acceptance of that hypothesis. In fact, a successful test of weak separability with random rates of return is included in Barnett and Zhou (1994). NOVEMBER/DECEMBER 1994 54 topic. The authors of the two Board staff papers are right. The authors have done a service to the profession by exploring the topic for the first time. CHALLENGES PRESENTED TO ECONOMIC THEORY Riskiness of the rate of return simply does not preclude the production of monetary services by an asset. Riskiness of the rate of return, how ever, certainly does make life more difficult for index number theorists and aggregation theorists. Most of the literature in those fields is produced under the assumption of perfect certainty or risk neutrality. Extensions of that literature to risk aversion were begun recently by Poterba and Rotemberg (1987), Barnett and Yue (1991), Barnett, Hinich and Yue (1989) and Barnett and Zhou (1994). We believe that the important issues raised by Collins and Edwards (1994) and Orphanides, Reid and Small (1994) at this conference should serve as motivation for further research on index number theory and aggregation theory under risk aversion. We do indeed welcome the increased motivation in that area provided by the work of those Board staff researchers. But there is an even more fundamental problem. The existence of an investment rate of return, even a perfectly certain one, raises questions about how an asset should be incorpo rated into an aggregate. While the existence of such a rate of return does not prevent an asset from producing monetary services, the share of the asset’s services that can be viewed as “monetary” is strongly affected. This comment is directed towards an investigation of that share. HISTORICAL BACKGROUND At one time, money was cash plus demand deposits. No controversy existed on that topic. But a sequence of technological changes and innovations occurred, and continued to occur, such that an increasingly large number of substi tutes for money produced an increasingly large share of the economy’s monetary services. The result was the Bach Commission, the Gurley and Shaw (1960) book, the Pesek and Saving (1967) book, and many other important contributions that influenced the growing movement towards the construction of increasingly broad monetary aggregates. The response of most central banks, however, has been to accept one aspect of that research while conveniently overlooking another closely related aspect. FEDERAL RESERVE BANK OF ST. LOUIS In particular, the researchers who first worked in that area were very clear on one simple, ele mentary fact: Investment yield is not a m onetary service. There was a reason that all monetary economists once agreed that money included only cash and non-interest-bearing demand deposits. If investment yield were a monetary service, then coal mines would be money. Land would be money. The entire capital stock of the United States would be money. This is not to deny that assets that produce an investment rate of return, whether risky or not, can produce monetary services. Interest yielding monetary assets, however, are joint products. Some of their services are monetary. Some are not. This fact seems to have escaped many of the world’s central banks. To the degree that the economy equates marginal utilities per dollar across assets, the marginal utility of mon etary services produced by an asset must decrease as its marginal non-monetary services increase— and investment return is very clearly not a monetary service. To underscore our point, we bring up the famous diamonds-versus-water paradox. The total utility of water exceeds that of diamonds, even though the marginal utility of diamonds exceeds that of water. In fact, as one moves along a concave utility function, marginal utility varies inversely with total utility. Hence, the statements made above about marginal utilities should not be confused with the total or average monetary service flow produced by an asset. But it is the marginal utilities that are relevant to measuring the prices in index numbers, such as the Divisia, Fisher ideal, Paasche or Laspeyres quantity indexes. We hope that we also do not have to remind this audience that the prices (user costs) in such indexes are not the weights. We nevertheless find that this literature contains many misunderstandings of monetary index number theory, and most of those misunder standings are produced by confusing prices with weights and marginal utilities with total or average utilities. FERRARI SPORTS CARS It has been asked at this conference whether stock funds or bond funds are “m oney,” or are they not money. We would like to ask a differ ent question. Are Ferrari sports cars transporta tion machines or recreational machines? We can imagine a Ferrari owner responding that a 55 Ferrari is strictly a transportation machine, and that the high price is produced by the Ferrari’s superior performance on highways and on winding roads. Hence, the price of a Ferrari is the discounted present value solely of the transportation services. But I expect that most of the rest of us would view the price of a Ferrari as being the sum of the discounted present values of two different flows: transportation services and recreational services. Ferraris are joint products, in terms of the services produced. Interest-bearing monetary assets similarly are joint products. Such assets produce both monetary and non-monetary ser vices, whereby the interest yield unquestionably is in the latter category. Hence, the correct answer to the question asked by the Board’s staff econo mists at this conference is that such assets, including stock and bond mutual funds, are partially money and partially not money. W HAT TO DO NEXT We see that we are presented with a paradox. More and more assets are contributing to the economy’s monetary service flow. As made clear by the authors of the two Board staff papers, stock and bond funds now are among those assets. The investment yields of that growing collection of assets, however, are not monetary services. If we do not add such assets into the monetary aggre gates, we overlook some of the economy’s mone tary service flow. If we do add those assets into the aggregates, we contaminate the aggregates with non-monetary services. The answer should be obvious. We must untangle the two discounted present values: the discounted present value of the monetary service flow and the discounted present value of the investment yield. Indeed, it can be done. THE THEORY Barnett (1987) defined the economic stock of money to be the discounted present value of expenditure on the services of monetary assets. Barnett (1991) derived that discounted present value in the form that we display below. During period s let p * = the true cost of living index, let Mis be nominal balances of monetary asset i, let rjs be the nominal expected holding period yield on monetary asset i, and define mis = M J p * to be real balances of monetary asset i. The current period is defined to be period t so that s > t. Define the discount rate for period s to be 1 fo r s = t (1) p s =<! s-i n (i + R u) f o r s > t . By letting the planning horizon, 7, go to infinity in the second term of Barnett (1978, eq. 2; 1980, eq. 3.3; 1981, eq. 7.3), we immediately acquire the following definition for the Economic Stock of Money, first derived as definition 1 and equation 2.2 in Barnett (1991): Definition 1: Under risk neutrality, the economic stock of money during period t is (2K=XX P's Ps P l i l + fj, m, Ps+1 The concept of economic stock used to produce Definition 1 is the user-cost-evaluated expenditure on the services of the n monetary assets that are its components. It should be observed that the pro cedure used in Barnett (1978, eq. 2; 1980, eq. 3.3; 1981, eq. 7.3) to acquire that discounted present value for finite T was just back substitution and algebraic manipulation of the sequence of flowof-funds identities. Hence, our conclusion is produced entirely from accounting identities. If we now substitute equation 1 and mis = M J p * into equation 2, we acquire the following result: M. Unfortunately, this equation includes expected future values of interest rates and of monetary asset holdings. While it may be reasonable to assume that interest rates are stationary, no such easy simplification is available for the stochastic process of future monetary asset holdings. We believe that a useful way to proceed would be to use VAR forecasts of the monetary asset hold ings and rates of return. We plan to produce results using that approach. But considering the time constraint that we faced with this confer ence, we had no choice but to make strongly simplifying assumptions. In particular, we make the assumption which causes equation 3 to collapse into the Rotemberg, Driscoll and Poterba (1994) CE index, first interpreted to be a stock index in Barnett (1991). NOVEMBER/DECEMBER 1994 56 Definition 2: The CE index is (4) R, -M , We seek to find conditions under which equa tion 4 will equal equation 3. To that end, sup pose that expectations are stationary in the sense that ris = r„ and Ris = Rit for all s > t, and consider the static portfolio, (Mjs, M2s, ..., Mns) = Mnf), for all s > t . Equation 3 reduces to viewed the interest yield on monetary assets to be a monetary service. In fact, that possibility was considered carefully and rejected unequivo cally in Pesek and Saving (1967). In the discussion that follows, we shall use this result to decompose the simple-sum aggregates into their investment share and their monetary services share. In each case, the share is produced by discounting the flow to the present value, the interest yield in one and the service flow in the other. The decomposition then is into Vt and V * , which partition Rt ( 5) I M„ ( i+Rt : i= 1 into its two parts, in accordance with equation 9. Observe, however, that (6)S R, - ri, £li +Rty -,+1 Rt R, since the left side of equation 6 is a convergent geometric series (minus the first term in the series). Substituting equation 6 into equation 5, we acquire our result: “AN CIEN T” HISTORY There was a time— long, long ago—when money was currency and demand deposits, and demand deposits did not yield interest. In those days, we see that n V* = 0, so that Vt = ^ Mit. 1= 1 Theorem 1: Under stationary expectations, the CE index is equal to the Economic Money Stock. Under the stationary expectations assump tion, we easily can discount to present value the expected investment yield flows, risMis = ritMit for s > t to get the following capitalized value: , (7) (i+r rm Again, we have a convergent geometric series in the summation over s at any given i, so that we find F“ M„ (8) Vt* 1= 1 llt Adding equation 8 to equation 4, we find that (9) Vt + Vt' = f^ M ir 1=1 The conclusion is clear. The simple-sum monetary aggregates measure the stock of money only if the investment (interest) yield of the monetary components is treated as a monetary service. Yet it is difficult to think of any macroeconomic school of thought which has ever FEDERAL RESERVE BANK OF ST. LOUIS Those were the days when the simple-sum monetary aggregates were created, and we see that the people who created them knew what they were doing. But that simpler world is long gone. Many assets that contribute to the economy’s monetary services also yield an investment rate of return. THE DATA We computed the decomposition into Vt and V* of the official simple-sum M l and M2 indexes along with the corresponding decomposition into Vt and V* of the newly proposed simple-sum M2+ index. We also computed the decomposition into V, and V* of bond mutual funds and stock mutual funds as a means of further investigating the source of the difference in behavior of M2 versus M2+. The attached figures provide the results. The decomposition depends upon the measurement of the benchmark rate of return Rt. Clearly, Vt increases as Rt increases, and V* decreases as Rt increases. Hence, the monetary service share (versus the investment share) of the simple-sum aggregate “joint product” increases as R, increases. The results can be biased in the direction favoring the inclusion of stock and bond funds by choosing an artificially high setting for Rt. For the purpose of biasing the 57 Figure 1 M1 Joint Product and Economic Capital Stock M1= simple sum joint product CEM1= economic capital stock part of the joint product results in that direction intentionally, we chose the highest possible setting for Rt that could be connected in any way with the available data. As shown by Rotemberg, Driscoll and Poterba (1994), Vt has a very volatile growth rate and, hence, they advocate smoothing the interest rates to produce smoother growth of the aggregates. This is not surprising, since Vt and V * are stock aggregates which tend to have volatile growth rates. We use the same smoothing method advocated by Rotemberg, Driscoll and Poterba (1994). In particular, we replaced all of the interest rates in the index by 13-quarter centered moving averages. Since the moving averages are centered, they are not defined for the first six quarters or the last few six quarters. We used the method advocated by Rotemberg, Driscoll and Poterba and phased in the centered moving average from asymmetric averages computed during the first six and last six observations. Once the smoothed interest data had been constructed, we searched over those series for the highest smoothed interest rate ever attained by any component asset during our sample. That ex post rate of return was 24.2 percent, which we selected to be the value of Rt for all t. In general, there is no reason for the benchmark rate to be a constant or to equal any ex post rate of return, since ex post rates of return tend to be much more volatile than ex ante expected rates of return. Our selection for the benchmark rate, however, produces the largest value that we could connect with the data, and we wanted to produce results that would be biased in favor of the Board staff members’ proposal. In interpreting our results, the division of the simple sum into the components Vt and V,* should be understood to be biased very strongly towards Vt and away from V *. Hence, the monetary services share in the joint product should be viewed as inten tionally exaggerated. THE RESULTS Figure 1 contains the partition of simple-sum M l into its investment share and its monetary services share. The solid line is the monetary service share produced from the computed value of Vj. The vertical gap between the solid line and the dotted line is the investment-motivated NOVEMBER/DECEMBER 1994 58 Figure 2 M2 Joint Product and Economic Capital Stock M2= simple sum joint product CEM2= economic capital stock part of the joint product share, V,*, which could be interpreted as the “error-in-the-variable” embedded in the simplesum index, M l. The height of the dotted line from the horizontal axis is the simple-sum index, equaling the sum of V, and V * . As is evident from Figure 1, the error-in-the-variable gap is relatively small and does not vary much over the sample. With a relatively constant vertical gap, the rate of growth of M l is not greatly affected by the error-in-the-variable gap. For most statistical inferences and for policy, the growth rate of money is what matters. Hence, we see that the existence of the V * error gap produces little difficulty for M l. Figure 2 contains the analogous plot and decomposition for the official simple-sum M2 aggregate. Observe that the error-in-the-variable gap is large (and would be much larger for a more realistic choice of Rt). In addition, that gap is variable and trends downward, especially recently. Hence, the existence of the error gap not only effects the short-run growth rate dynamics of the aggregate, but also biases downward the FEDERAL RESERVE BANK OF ST. LOUIS long-term growth rate. Inferences and policy are not invariant to the existence of this gap. Figure 3 contains the decomposition for the M2+ aggregate that has been proposed at this conference by Collins and Edwards (1994) and Orphanides, Reid and Small (1994). Observe that while the error gap is even larger than for M2, the size of the gap is less variable and no longer trends downward. Hence, the growth rate of the error-shifted dotted line approximately tracks the growth rate of the “correct” solid line. To see why M2+ stabilizes the size of the gap and thereby improves on the aggregate’s growth rate performance, see Figures 4 and 5, which display the decomposition of the stock mutual funds data and the bond mutual funds data into their economic capital stock share and their error-in-the-variable shift. Observe that in each of those two cases, the size of the gap grew rapidly during the past two years. This growing error gap offsets the declining error gap in M2, when the stock and bond fund data are added into M2. 59 Figure 3 M2+ Joint Product and Economic Capital Stock M2+= simple sum joint product CEM2+= economic capital stock part of the joint product In short, we conclude that the authors of the two papers presented at this conference are correct in concluding that the growth rate behavior of M2 is improved by incorporating the stock and bond mutual fund data into M2. Indeed, it does appear that substitution from M2 components into stock and bond mutual funds has become important. W HERE IS A L L OF THIS GOING? There is an underlying dynamic to this trend in monetary theory. Stabilizing the size of the error gap requires continually incorporating more assets into the monetary aggregates. The size of the gap keeps growing. The share of the monetary aggregate representing discounted monetary ser vices continues to decrease, and the monetary aggregates look increasingly like pure investment capital rather than money. Even if stabilizing the size of the gap offsets long-run errors in growth rate paths, the short-run dynamics of the aggre gates are likely to become increasingly disjoint from monetary services growth. In this paper we use the CE index, equation 4, to permit easy decomposition of the simple-sum aggregate “joint product” into its monetary service and investment shares. Using the formula in equation 3 with forecasted variables, perhaps by a VAR, would be better. But generating data that depends upon forecasts is unpleasant for data-producing governmental agencies. Smoothing interest rates to decrease the volatility of the resulting aggregate is also unpleasant for governmental agencies. For this conference, decomposition of the stocks in that manner was revealing. But as a means to produce data for a central bank, there is a better way. It is the Divisia monetary aggregates long advocated by Barnett (1980). See Barnett, Fisher and Serletis (1992) and Belongia and Chalfant (1989) for an overview and some of the relevant empirical results. The Divisia monetary aggregates directly mea sure monetary service flows, not the discounted stock levels. The Divisia monetary aggregates do not require smoothing of interest rates to smooth the index’s growth rate, and the Divisia monetary NOVEMBER/DECEMBER 1994 60 Figure 4 Common Stock Mutual Funds Joint Product and Economic Capital Stock StockQ= simple sum joint product CEstock= economic capital stock part of the joint product Figure 5 Bond Mutual Funds Joint Product and Economic Capital Stock BondQ= simple sum joint product CEbond= economic capital stock part of the joint product FEDERAL RESERVE BANK OF ST. LOUIS 61 aggregates contain no variables that need fore casting.2 In addition, Barnett (1991) proved that if we could do the forecasting needed to compute the monetary capital stock (equation 3), the result would be identical to that produced by discounting to present value the future stochastic process of the Divisia monetary aggregate. _____ , and Piyu Yue. “Exact Monetary Aggregation Under Risk,” Washington University (Department of Economics), St. Louis, Working Paper #163 (December 1991). REFERENCES Belongia, Michael T., and James Chalfant. “The Changing Empirical Definition of Money: Some Estimates From a Model of the Demand for Money Substitutes,” Journal of Political Economy (April 1989), pp. 387-97. Barnett, William A. “Reply to Julio J. Rotemberg,” in Michael T. Belongia, ed., Monetary Policy on the 75th Anniversary of the Federal Reserve System: Proceedings of the Fourteenth Annual Economic Policy Conference of the Federal Reserve Bank o f St. Louis. Kluwer Academic Publishers, 1991, pp. 232-43. _____ . “The Microeconomic Theory of Monetary Aggregation,” in William A. Barnett and Kenneth J. Singleton, eds., New Approaches to Monetary Economics, Proceedings of the Second International Symposium in Economic Theory and Econometrics. Cambridge University Press, 1987, pp. 115-68. _____ . “The Optimal Level of Monetary Aggregation,” Journal of Money, Credit and Banking (February 1982), pp. 687-710. _____ . Consumer Demand and Labor Supply: Goods, Monetary Assets, and Time. North-Holland, 1981. _____ . “Economic Monetary Aggregates: An Application of Index Number and Aggregate Theory,” Journal of Econometrics (summer 1980), pp. 11-48. _____ . “The User Cost of Money," Economics Letters, vol. 1 (1978), pp. 145-9. _____ , and Ge Zhou (1993), “Financial Firm’s Production and Supply-Side Monetary Aggregation Under Dynamic Uncertainty,” this Review (March/April 1994), pp. 133-65. _____ , Douglas Fisher, and Apostolos Serletis. “Consumer Theory and the Demand for Money.” Journal o f Economic Literature (December 1992), pp. 2086-119. 2 It is necessary to measure the benchmark rate, R, to construct the Divisia monetary aggregates, and we advocate the use of the upper envelope of the yield-curve-adjusted, holdingperiod yields on all of the components in the broadest aggregate. Obviously, we do not advocate the use of the extreme, constant setting of 24.4 percent, chosen for an illustrative purpose in this paper. However, it should be observed that the behavior of the Divisia monetary aggre gate is much more robust to variations in the method of measuring the benchmark rate than is the CE index, which is very sensitive to that rate’s selection. The reason is that _____ , Melvin Hinich, and Piyu Yue. “Monitoring Monetary Aggregates Under Risk Aversion,” in Michael T. Belongia, ed., Monetary Policy on the 75th Anniversary of the Federal Resen/e System: Proceedings of the Fourteenth Annual Economic Policy Conference o f the Federal Reserve Bank of St. Louis. Kluwer Academic Publishers, 1989, pp. 189-222. Collins, Sean, and Cheryl L. Edwards. “An Alternative Monetary Aggregate: M2 Plus Household Holdings of Bond and Equity Mutual Funds,” this Review (November/December 1994), pp. 7-29. Gurley, John, and Edward S. Shaw. Money in a Theory of Finance. Brookings Institution, 1960. Orphanides, Athanasios, Brian Reid, and David H. Small. “The Empirical Properties of a Monetary Aggregate That Adds Bond and Stock Funds to M2,” this Review (November/December 1994), pp. 31-51. Pesek, B.P., and T.R. Saving. Money, Wealth and Economic Theory. Macmillan, 1967. Poterba, James M., and Julio J. Rotemberg. “Money in the Utility Function: an Empirical Implementation,” in William Barnett and Kenneth Singleton, eds., New Approaches to Monetary Economics, Proceedings of the Second International Symposium in Economic Theory and Econometrics. Cambridge University Press, 1987, pp. 219-40. Rotemberg, Julio J., John C. Driscoll, and James M. Poterba. “Money, Output, and Prices: Evidence From a New Monetary Aggregate,” Journal o f Business and Economic Statistics (forthcoming). Swofford, James L., and Gerald A. Whitney. “A Revealed Preference Test For Weakly Separable Utility Maximization With Incomplete Adjustment,” Journal o f Econometrics (January/February 1994), pp. 235-49. the benchmark rate appears symmetrically in both the numerator and denominator of the share weights of the Divisia index, which in turn is a growth-rate index. Since the CE index is a level index, variations in any interest rate, including the benchmark rate, produce jumps in the level of the unsmoothed index. Jumps in levels produce spikes in growth rates. NOVEMBER/DECEMBER 1994 63 Jacob S. D reyer Jacob S. Dreyer, former chief economist at the Investment Company Institute, is president of JDA International, a consult ing firm. Commentary I PRESUME THAT I WAS ASKED to comment on the papers by Collins and Edwards, and Orphanides, Reid and Small because I spent a good portion of the last six years looking at the “+,” that is, on the assets, the desirability of whose inclusion into a broadened monetary aggregate is the subject of these deliberations. It is true that I not only watched stock and bond fund assets, and money market fund assets, but also had some influence on their definition, methods of collec tion and aggregation, and similar mundane matters. So, when the yield curve steepened a few years ago and, consequently, the flow of retail savings into longer-term debt instruments intensified, those who were uncomfortable with the apparent unresponsiveness of M2 to consec utive reductions in the Fed funds rate frequently used me as a sounding board for all sorts of p r o p o s a l s to r e d e f in e th i s m o n e t a r y aggregate. In other words, I have been exposed to the alleged problems and proposed solutions before and I am pleased to discuss them once again. However, since I was asked to specifically comment on the presented papers, and duty comes before pleasure, let me start with the papers and then move on to the broader issue of re-definition of monetary aggregates. I will begin with the article examining empirical properties of an expanded M2, M2+. The authors of this article, Orphanides, Reid and Small, do an excellent job of examining the stability, indicator properties and, to some extent, controlability of M2+ (that is, M2 plus assets of bond and equity mutual funds) by trying to determine the demand for M2+, including reduced-form relationships that presumably would capture the determinants of supply as well. To specify the demand for M2+, they correctly recognize the importance of specifying the ex ante return on its close substitutes—real assets such as commodities and durable goods, as well as financial assets such as direct holdings of short term instruments, bonds and stocks. However, they note the difficulty of measuring such returns. They also note that M2+ would include the capital gains and losses on stock and bond mutual funds, because of the difficulty—I would say impossi bility— of removing such ex ante gains and losses from the mutual fund component. Not inciden tally, they do not believe such gains and losses should be removed, a subject to which I shall return later. It is important for me to summarize their conclusions because I will be using them in commenting on the other article. The authors conclude from their analysis, and I agree with their conclusions, that: 1. On stability: “...the estimated demand functions are not very stable. Hence, the usefulness of these equations in interpreting and forecasting movements in M2+ may prove to be limited.” 2. On indicator properties: “In terms of their information content, M2+ and M2 do not appear to have differed NOVEMBER/DECEMBER 1994 64 significantly.” Both overpredicted nominal GDP before and during the last recession. “... capital gains and losses cause growth of M2+ to be more volatile than that of M2. Moreover, the capital gains and losses in M2+ may cause movements in the aggregate that neither reflect shifts in the stance of monetary policy nor provide appropriate signals for changes in policy.” 3. On controllability: The authors do not explicitly examine this aspect but, based on their previous statement just quoted, presumably would agree that M2+ is far less controllable than M2, which itself has not proven to be con trollable at all during the period when it was the major monetary aggregate targeted by the Federal Reserve. On the basis of these conclusions, they doubt that anything would be gained by expanding M2 to include assets of stock and bond mutual funds. I agree completely. By contrast, the authors of the second paper, Collins and Edwards, still consider M2+ (and its companion, M3+) a worthwhile replacement for M2. They argue that M2+ is less worse than any other proposed replacement, first because the added components, that is, assets of long-term mutual funds, have enough moneyness—their medium-of-exchange and liquidity attributes, and second, because, as intermediation continues to shift wealth from present components of M2 to long-term mutual funds, M2+ should, in fact, grow in value as an indicator of economic activity. The authors of the paper recognize the poten tial pitfalls of using M2+. For example, some of the components of M2+ will have to be inter polated from annual data to monthly series in order to net out the institutional assets that would otherwise lead to double counting. Such interpolation drastically reduces the indicator properties of an aggregate for short-term purposes. Even if we leave the interpolation problem aside, the usefulness of M2+ as a long-run indicator also would be limited: For the demand for M2+ to be properly specified, one would n eed to specify ex ante returns—a task bordering on the im possible, as the authors o f both papers seem to acknow ledge. FEDERAL RESERVE BANK OF ST. LOUIS The technical issue then appears to be, at least in part, whether the moneyness attributes of the “+” make its components so indistin guishable from the components of M2 proper as to warrant the expansion of M2 despite the deficiencies of the broader measure established by Orphanides, Reid and Small, and pitfalls of using it as recognized by Collins and Edwards. A broader issue is, of course, what would be gained, if anything, by including assets of stock and bond funds into an expanded M2. So now, as my principal assigned duty is fulfilled, I feel I am entitled to express my other thoughts on this matter. In order not to keep the audience breathless, I shall state at the outset that I disagree with the conclusions of the paper by Collins and Edwards and think that nothing would be gained by introducing yet another, broader monetary aggregate. First, if the demand for M2+ is neither stable nor more informative than that for M2, as Orphanides, Reid and Small show, then deciding to adopt it because it is less worse than other proposed aggregates is a most unusual criterion for selection. I shall elaborate on this in a moment. Second, it is fairly easy to show that the added components making up M2+ lack moneyness, contrary to what the authors of the second paper claim. Table 1 gives the turnover of some of the major components of M2 (those in columns one through four, and six and seven of the table; small time deposits, overnight repurchase agreements and overnight Eurodollar deposits were not available) and the turnover of bond and equity funds (the last two columns). The striking feature of the table is that the turnover rates for the bond and equity funds are so much lower than that for any other financial assets. They are less than 10 percent of that of general-purpose money market funds, and less than 0.07 percent of that of demand deposits outside of New York City hanks. Moreover, in contradistinction to demand deposits and even money market funds whose turnover ratios have tended to drift upwards over the nine years por trayed in the table, turnover ratios on bond and equity funds have declined or remained constant. Thus, their relative moneyness, at least as mea sured by turnover ratios has, in fact, decreased. I may add that the turnover ratio has in its numerator the sum of debits, that is, it includes exchanges out of these funds. These in most 65 Table 1 Turnover of Deposits, MMMFs and Bond & Equity Funds, 1984-92 Deposits at commercial banks Demand deposits Demand deposits NYC banks other banks MMMFs OCDs Savings Institution Bond & equity funds Broker/ General Bond/ dealer purpose income Equity 1984 1,843 269 15.8 5.0 4.2 3.0 2.2 0.4 0.3 1985 2,169 302 16.7 4.5 5.0 3.5 2.5 0.3 0.4 0.5 1986 2,461 327 16.8 3.0 4.6 3.4 2.9 0.3 1987 2,671 357 13.8 3.1 4.6 3.7 3.0 0.5 0.6 1988 2,897 333 13.2 2.9 4.6 3.5 2.4 0.3 0.5 1989 3,421 408 15.2 3.0 4.5 3.6 2.4 0.3 0.4 1990 3,820 465 16.5 6.2* 4.3 3.2 2.2 0.3 0.4 1991 4,271 448 16.2 5.3* 4.8 3.2 2.3 0.3 0.4 1992 4,798 436 14.4 4.7* 6.9 3.6 2.7 0.3 0.3 'Includes money market deposit accounts at banks for these years. SOURCES: Federal Reserve Bulletin (various issues) and Investment Company Institute cases represent changes in asset composition of a shareholder’s overall portfolio rather than debit to a checking account or a money market fund which has for its counterpart acquisition of, say, a new suit. Excluding exchanges out (that is, redemption exchanges) from debits would make the already low turnover ratios for long-term funds considerably lower. In any event, the low turnover of bond and equity funds indicates that they probably have far less to do with the types of transactions that might influence GDP over the span of time normally adopted in framing monetary policy. In fact, it seems plausible that a switch of, say $1,000, from my money market fund or bank account to a bond or stock fund, suggests a decision to move a potential near-term $1,000 purchase into the future, quite possibly beyond the relevant time frame. Consequently, it seems to me that, contrary to the claims of Collins and Edwards, very little, if any, moneyness resides in the bond and equity components of M2+. I think that the extremely low transaction cost of converting these assets into cash creates this illusion of moneyness. Third, adopting M2+ because it might eventu ally have better indicator properties and would, at least, in the meantime, envelop close substitutes for M2 assets, leads me to ask where will the broadening end? Milton Friedman recognized many years ago that the definition of money was as much an empirical as a theoretical matter, but deciding where to draw the line is not easy and, under current law, the Federal Reserve must decide where to draw it. Now, I don’t want to get into a largely doctrinal argument over the wisdom of the requirement to set targets for monetary aggregates. However, because financial institu tions are able and willing to grant credit lines collateralized by all types of assets, it does seem to me that broadening the definition to envelop close substitutes h as as its fea sib le upper limit all o f hou seh old wealth i f not all o f national wealth. I can get liquidity out of my home equity by simply opening my desk drawer, grasping my pen and writing out the amount I wish to spend. Moreover, I don’t have to fear the tax consequences of my decision because I don’t have to compute and report any of the capital gain on my home at the time I wrote the check. By contrast, if I were to write a check on my bond fund this would be a capital transaction requiring relatively complex computations of my cost basis for reporting short- or long-term capital gains. In this sense, a home equity line of credit has greater moneyness than a stock or bond fund. Thus, rather than stopping with including bond and equity mutual funds, shouldn’t we go all the way and include home equity, margin credit, other lines of credit, NOVEMBER/DECEMBER 1994 66 including signature loans and, eventually, as our experience with financial innovation suggests, any assets that can be sold or borrowed against?1 This third comment brings me to a question that has been too easily dismissed by Collins and Edwards. Has the definition of money gone in the wrong direction? It seems to me that is a conclusion a reasonable person could draw from the experience of the past several decades as the Federal Reserve tended to broaden the definition of money. Despite the view of the authors that the Federal Reserve now would be reluctant to backtrack towards narrow measures of money, it is my humble opinion that the movement has been indeed in the wrong direction. Although we all understand that monetary aggregates are not the ultimate objectives of policy— sustained growth and low inflation are really the measures by which the Federal Reserve policies will be judged—I have always been struck by the fact that the Federal Open Market Committee (FOMC) has typically used various components of bank reserves as internal guides for its open market operations even as it was proceeding in evolutionary steps from narrow to broad definitions of money when communicating its objectives to the public. 1 I can understand the tendency to resort to broader defini tions if the desire is to incorporate wealth effects into the for mation of monetary policy. However, such effects tend to have little explanatory power from quarter to quarter, or from year to year, in explaining movements in aggregate demand, production, employment and especially inflation. Furthermore, if this were the case, all assets that can easily be liquified would have to be included into such a wealtheffect-capturing aggregate. FEDERAL RESERVE BANK OF ST. LOUIS Of course, I don’t want to give the impression that the links between narrow aggregates like reserves or the monetary base and the ultimate policy objectives are any more precise than the links between broader aggregates and those objectives. Nevertheless, the place of such narrow aggregates in the interplay of policy, ulti mate objectives and information variables (such as commodity prices, interest rates, yield curves and exchange rates) has been more “enduring” in the conduct of monetary policy than any of the publically targeted monetary aggregates. To summarize both papers, I agree with Orphanides and others and disagree for the same reasons with Collins and Edwards that adding M2+ to the monetary zoo would not make monetary policy more effective in achieving its ultimate objectives or make the Fed’s stance more trans parent to the Congress and the public. I have tried to put the work of the two papers in the larger context of either being prepared to draw the line at total wealth rather than repeat edly drawing the line, erasing it, and redrawing it again and again at arbitrary segments of the wealth spectrum or, on the other hand, of moving back towards narrower aggregates which the Federal Reserve uses and can control. To me the choice is obvious. 67 John V. Duca John V. Duca is a research officer at the Federal Reserve Bank o f Dallas. The author thanks Anne King and Chih-Ping Chang for their excellent research. Commentary SOME BACKGROUND ON MONEYDEMAND INSTABILITY PAST EPISODES OF MISSING money can shed light on whether bond and/or equity funds should be added to M2. My perspective on how to analyze monetary aggregates can be char acterized as a dynamic market-share approach. If financial aggregates have a stable relationship to nominal GDP and if banks have a stable share of the financial market, then bank-based mone tary aggregates like M2 will be helpful indica tors. The two most pronounced episodes of missing money, M l in the mid-1970s and M2 in the early 1990s, occurred when the competi tiveness of the banking system declined. In the mid-1970s, firms shifted away from bank loans toward commercial paper at a time when Regulation Q induced banks to ration credit and banks were passing along the heightened cost of reserve requirements when interest rates were high. On the liability side, binding deposit rate ceilings and high interest rates led firms and households to adopt cash management and to use money market funds which purchased commercial paper and Treasury bills. In terms of flows, firms used the proceeds from issuing paper to pay off bank loans while banks used these funds to pay off depositors who were shifting assets into money funds. In Figure 1, the development of money funds allows part of the flow of short-term finance to bypass the banking system. By comparison, the bypassing of the banking system in the early 1990s occurred in the flow of medium- to long-term finance (Figure 2). Higher deposit insurance premiums and more costly risk-based capital standards led banks to boost the spread of prime over short-term rates, which helped induce firms to shift toward bond and equity financing. At the same time, wider net interest margins stemming from regulatory changes, coupled with a steep yield curve, encouraged households to shift out of small time deposits into bond and equity funds. In terms of flows, firms paid off bank loans with proceeds from issuing bonds and stocks bought by mutual funds whose purchases, in turn, were financed by assets that households shifted out of bank deposits. Both episodes show how the banking system is not a closed loop, because agents innovate to circumvent banks when banks become relatively more costly to use. THE CEN TRAL EM PIRICAL ISSUE If one could model the shocks to money demand, then modified money-demand models would work. However, if households have fun damentally changed their asset behavior, then it may be better to broaden an aggregate. In assessing the impact of Resolution Trust Corporation (RTC) activity and the yield curve on M 2 , 1 have found that M2 plus non-IRA/Keogh household bond funds is more explainable than M2 using Federal Reserve Board-style (circa 1990) M2 models (Duca, forthcoming). This suggests that the behavioral relationships have changed. However, given that bond funds were negligible prior to the mid-1980s, the analysis was effectively conducted over a period when bond fund assets did not suffer sizable capital losses. T he issu e o f NOVEMBER/DECEMBER 1994 68 Figure 1 Short-Term Finance w h eth er to a d d b o n d or b o n d an d s to ck fu n d s to M2 is an em p irical on e an d b oils dow n to w h eth er we lo s e m ore from m akin g M2 m ore vu ln erable to c a p ita l gain s an d losses than we gain from internalizin g p ortfo lio shifts betw een M2, an d b o n d an d equ ity fu n ds. A CO M M ENT ON COLLINS AND EDWARDS This is a good paper. The authors are very careful in how they construct and describe the data used in building M2+. This study will be a helpful resource for many analysts. SOME COMM ENTS ON ORPHANIDES, REID AND S M A LL (ORS) Overall, this is a very nice and careful study. The only suggestion I have regards how the authors assess the indicator properties of M2 and M2+. I have some reservations about using only Granger regressions of GDP and money growth rates to assess indicator properties in the ORS study. This approach has the problem of letting bygones-be-bygones. That is, variability in money and GDP growth may obscure any FEDERAL RESERVE BANK OF ST. LOUIS information in long-run relationships between money and nominal output, if such relationships still exist. On this point, ORS could look into a simple error-correction model of nominal GDP that imposes a long-run velocity relationship. They then could compare results using M2 versus M2+ as ad d itio n a l evidence about indicator properties. To shed some light on this point, consider some forecasts of inflation using a framework that imposes a long-run relationship between money and nominal output. Namely, the (in)famous P-star model. Figure 3 shows out-of-sample forecasts of inflation, as measured by the implicit GDP deflator. These extend recent research with Zsolt Becsi (Becsi and Duca, forthcoming). As we can see, the P-star model using M2 severely under-predicts inflation to the point of forecasting deflation in 1993. By contrast, M2 plus bond funds (M2B) and M2+ do a good job of tracking this inflation measure since 1991, with a slight edge to M2B. Interestingly, M2 does a better job in predicting inflation during the m id-1980s’ surge in bond and equity funds, whereas M2+ and M2B do better during the early-1990s’ surge in mutual funds. Why? 69 Figure 2 Long-Term Finance W H Y THE MID-1980S’ SURGE IN M U T U A L FUNDS DIFFERS FRO M THE 1990s’ SURGE I believe that the answer reflects differences in the sources of inflows during these episodes. The surge of the mid-1980s came shortly after IRA, 401K and Keogh regulations were liberalized. Given the incentives to use these retirement vehicles, many households learned more about mutual funds and likely applied this knowledge to other asset holdings. This is consistent with the fact that household holdings of IRA/Keogh and non-IRA/Keogh bond and equity fund bal ances grew rapidly in the mid-1980s. It is also consistent with flow-of-funds data, which sug gest that the assets that households shifted into bond and equity funds came more from direct holdings of bonds and equities than from M2 deposits. This finding is also consistent with the relatively good fit of M2 demand models in the mid-1980s. By contrast, flow-of-funds data suggest that more of the inflows into bond and equity funds during the early-1990s reflected shifts out of M2 deposits rather than out of direct bond and equity holdings. This is consistent with the missing M2 phenomenon of recent years. Four factors may explain why the inflows into bond and equity funds came more from M2 in the early-1990s relative to the mid-1980s. First, compared to the m id-1980s, the yield curve was steeper for a longer period of time in the early1990s. Thus, households had a greater incentive to shift out of M2 deposits in recent years. Second, because short-term rates fell much more in the early 1990s than in the mid-1980s, there were negative income effects on retirees holding small time deposits that encouraged them to shift out of bank CDs into higher-earning bond and equity funds. Third, declines in loads and fees on mutual funds (as shown by ORS) reduced the cost of shifting into mutual funds. Milbourne’s (1986) modified Miller-Orr model implies that smaller loads will induce shifts from M2 into bond and stock funds. The fourth factor reflects the realization during the early-1990s that jobs are less secure—especially for professionals. As the world becomes more Schumpeterian, house holds will increasingly rely on portable, defined contribution pensions. Such plans typically require that households make investment NOVEMBER/DECEMBER 1994 70 Figure 3 Actual and Forecasted Inflation from the P* Model (Implicit GDP Deflator, SAAR) Percent decisions. As a result of being more active in managing their retirement assets, households are becoming increasingly aware of alternatives to M2 and are becoming better managers of their assets. Differences between the mid-1980s and early 1990s imply that future research should examine the substitutability of bond and equity funds not only for M2, but also for direct holdings of bonds and equity. In addition, future work that applies learning models to bond and equity funds may prove fruitful. W HAT SHOULD THE FED DO? I favor an eclectic approach to conducting monetary policy because innovation by the private sector at times causes breakdowns in the relationship between financial variables and the economy. That said, part of our job at the Fed is to update financial indicators in light of those innovations. As for using monetary aggregates as indica tors, I have two positions. First, since recent innovations are mainly affecting the non-M l component of M2, narrow money measures, net of currency, could be used as information FEDERAL RESERVE BANK OF ST. LOUIS variables within models that control for the high sensitivity of narrow money to interest rates and mortgage refinancing activity. Nevertheless, the high rate sensitivity of narrow aggregates limits their usefulness as monetary targets under the Humphrey-Hawkins Act. Second, I would also monitor M2 and M2 broadened to include bond and/or equity mutual funds, keeping in mind that capital gains and losses w ill have direct price effects on M2+ and M2B balances and will induce portfolio substitution between M2 and these broader aggregates. REFERENCES Becsi, Zsolt, and John V. Duca. “Adding Bond Funds to M2 in the P-Star Model of Inflation,” Economics Letters (forthcoming). Collins, Sean and Cheryl L. Edwards. “An Alternative Monetary Aggregate: M2 Plus Household Holdings of Bond and Equity Mutual Funds,” this Review (November/December 1994), pp. 7-29. Duca, John V. “Should Bond Funds Be Included in M2?,” Journal o f Banking and Finance (forthcoming). Milbourne, Ross. “Financial Innovation and the Demand for Liquid Assets,” Journal o f Money, Credit and Banking (November 1986), pp. 506-11. Orphanides, Athanasios, Brian Reid, and David H. Small. ‘The Empirical Properties of a Monetary Aggregate that Adds Bond and Stock Funds to M2,” this Review, (November/December 1994), pp. 31-51. 71 Joshua Feinm an Joshua Feinman is vice president o f Banker’s Trust. Commentary L SYMPATHIZE WITH THE DIFFICULTIES that the authors have had in trying to explain and model the financial innovation intermediation that has been occurring in recent years, and the ways that these changes have affected the relationships between monetary aggregates and GDP. In fact, I grappled with some of these issues m yself when I was at the Board of Governors. I certainly appre ciate the difficulties and potential pitfalls. At the Board, I had it a bit easier than these authors, since I was mainly working on modeling M2. These authors have taken it a step further by constructing and modeling a new aggregate, M2+, that really goes in a very different direction and takes us in ways that the traditional monetary aggregates have not had to deal with. A starting point for discussing the behavior of any monetary aggregate—M2, M3 or M2+, or any other—has to be the changes that have occurred in our financial system in recent years. Particularly important have been the shifts in the patterns of financial intermediation and the efforts by both borrowers and lenders to adopt a more cautious approach to leverage. The most important change in financial intermediation has been the relative shrinkage of the depository sector. At the same time, perhaps the most important financial innovation has been the proliferation of information about the accessibility and liquidity of mutual funds. Both of these changes reflect the effects of a dramatic shrinkage of the government subsidy to the depository system in recent years. This shrinkage has manifested itself in a variety of ways, including higher deposit insurance pre miums, more stringent capital standards, and tighter supervision, regulation and examination. All this has resulted in a smaller amount of depository intermediation in the economy and a smaller share of the economy’s overall credit being recorded on the books of depositories. In addition, we have seen a smaller share of that smaller amount of depository credit being funded through deposits, in part because of increases in deposit insurance premiums. The pie charts shown by Cheryl Edwards and Sean Collins illustrate the decreasing impor tance of deposits as a share of total household assets. This is obviously just the flip side of the depository shrinkage. The banking system has been less aggressive in pursuing deposits in recent years, and households have reacted by reducing their deposits accordingly. The larger upshot of this, of course, has been major shifts in the relationships between broad monetary aggregates—composed primarily of the liabilities of the depository sector— and GDP. This has resulted in the recent record-high GDP velocities of the broad monetary aggregates. I believe this explains the major part of what has been happening. Although the steepening yield curve has certainly played a role, I never thought that was the key thing. Even though I’ve used yield curve spreads myself in modeling, I’ve always had problems with it on theoretical grounds. First, you have the theoretical problems arising from the expectations view of the yield curve. NOVEMBER/DECEMBER 1994 72 Saying people simply pick the highest point on the yield curve flies in the face of economic theory. But even more to the point, I think there’s nothing per se that prevents banks, if they want to, from going out the yield curve and pursuing longer-term CDs more aggressively. If you didn’t have the shrinkage in the banking system going on—which I think has been the fundamental thing—banks could and probably would have been more aggressive in pursuing longer-dated CDs with more aggressive pricing. The bottom line is, in my opinion, the shrinkage of depository sector. This has been accompanied by increases in the direct-placement credit market and by the growth of new financial intermediaries. Obviously, the most visible new intermediary has been the mutual fund industry that we are discussing. That is why, at a very simple level, it is extremely appealing to want to create an aggregate that adds in these mutual funds to M2. The charts in the paper by Athanasios Orphanides and his colleagues show clearly that we have had outflows from M2 and its components while we have had strong inflows into these mutual funds; from the surface it looks appealing to build a new aggregate including these mutual funds. Building a broad monetary aggregate that internalizes substitution among alternative assets also was the intent of the Federal Reserve Board staff when they defined the current M2 in 1980. At that time, the substitution was from M l to, primarily, small CDs and money market mutual funds (MMMFs). Including the close alternatives to M l meant that the new broader aggregate was much less sensitive to market interest rates. This is for two reasons. First, the rates paid on the deposits included in the non-M l component of the new M2 aggregate tended to adjust much more quickly to changes in market rates then did rates paid on M l. This sharply reduced the incentive to substitute away from M2, into assets not included in M2, when market rates changed. Second, the broader aggregate captured internally the substitution from liquid deposits into small time deposits when market rates increased and vice versa. When we look at the charts shown today, it is very tempting to say that a new, broader aggregate could capture some part of the substitu tion away from small time deposits and MMMFs and into bond and equity mutual funds. Even so, having stated that, I think the articles FEDERAL RESERVE BANK OF ST. LOUIS suggest— and I agree—that we must be skeptical about the usefulness of an aggregate that adds these stock and bond funds to M2. To avoid get ting into a tremendous amount of detail, I would just highlight a couple of points. Obviously, the capital gains issues that both Athanasios Orphanides and Cheryl Edwards mentioned is critical. It’s hard to know what might be the correct answer. It’s somewhat arbitrary to exclude capital gains completely. If you exclude them initially, however, you must decide when to include them in the aggregate. At what point do the capital gains start looking to households just like regular money? That’s a problem. On the other hand, when you include capital gains as they do, you bring up another whole set of other problems. The monetary aggregate is going to be extremely responsive to moves in the stock market, for example. Could you ever target such an aggregate? Scenarios under this which could lead to some pretty silly policy responses are not too hard to concoct. I believe that if you ever consider going oper ationally to such an aggregate, you would almost have to begin with a Q4 base period and try to differentiate over the course of the year what is contributing to that growth, separating how much is coming from capital gains, how much is coming from net inflows, and so on. I don’t know if you can do that in a timely enough way for policy analysis, however. Once you start following the components of an aggregate rather than the aggregate itself, I’m not sure that you gain a lot by defining the aggregate in the first place. That’s an obvious problem, but really the main reason we’re here. The second problem is one with the modeling exercise: Are there good, timely measures of the expected returns on these stock and bond mutual funds? These are the returns that belong in a money-demand equation. If you try to form these returns ex post in a backward-looking manner, you have a lot of theoretical problems in trying to justify it. If you try to look ahead, you have to specify the processes generating the data and how much people know about them. I know I’m not saying anything new here, but I want to highlight what I think is obviously a big problem. Finally, I would say that (as Athanasios pointed out) M2+ doesn’t really seem to have been a better indicator of GDP than M2, except perhaps during the last couple of years. We just don’t have a very long history, and we know 73 things are obviously still evolving. There may not, in fact, be stable demand function for M2+ that we can estimate. It doesn’t seem to have any history of being a better indicator of GDP. It’s more variable than M2, or at least not any less variable. Finally, the kinds of institutional changes that have affected the traditional rela tionships between M2 and nominal GDP are continuing. These changes even today are affecting the development of M2+. Their strength likely makes any kind of historical data not terribly useful for looking ahead. We should keep in mind that when things settle down—when the depository sector completes its shrinkage and stabilizes—the velocities of our current aggregates, particularly M2, might stabilize again, albeit at a permanently higher level. In addition, the short-run relationships between M2 and its opportunity cost that we relied on in the past might well re-emerge. In fact, unnoticed by many analysts, M2 velocity and its opportunity cost have increased in a roughly parallel way since early 1992, suggesting that this pattern might already be emerging. Unfortunately, at this point the jury is clearly still out and we really can’t answer that question definitively. The evidence that I’ve seen presented suggests that there is no other monetary indicator that is going to provide monetary policy with a long-term nominal anchor the way we thought M2 did prior to 1990. In the absence of such an anchor, it seems to me that monetary policy w ill continue to be made basically as it has been for some time now, without a monetary aggregate, by adjusting the nominal federal funds rate in response to activity in the real economy. NOVEMBER/DECEMBER 1994 75 G eorge G. Pennacchi George G. Pennacchi is professor of economics in the Department o f Finance at the University o f Illinois, UrbanaChampaign. Commentary INANCIAL INNOVATION CONTINUES to obscure traditional measures of monetary policy and efforts to find an improvement are now underway. A new, potentially superior monetary aggregate, M 2+, is carefully analyzed in the complementary papers by Sean Collins and Cheryl Edwards, and Athanasios Orphanides, Brian Reid and David Small. The Collins-Edwards (CE) paper provides a thoughtful discussion of issues dealing with the construction of M2+, while the Orphanides, Reid and Small (ORS) paper considers the empirical qualities of this new aggregate. I wish to complement both sets of authors for their work. Given their research objectives, these papers are professionally executed and insightful. As CE notes, M2 became uncoupled from nominal income beginning in 1990. During this period of slow M2 growth, funds invested in bond and stock mutual funds increased dramati cally. However, before simply assuming that M2 can be resurrected by adding in bond and stock mutual fund balances, CE sensibly investigate whether flows from M2 into these mutual funds were the only problems occurring with M2. Bearing in mind CE’s warning that it is “tricky business” to explain the weakness in M2 based on an analysis of its components, I would like to further discuss movements in the components of M2. I believe that a great deal of information is 1 Interest rates are averages of a Federal Reserve Board sur vey of approximately 460 commercial banks. This lowering of spreads between more competitive retail deposits (CDs) and less competitive retail deposits (NOWs) typically occurs present in the individual components of monetary aggregates. Table 1 presents a very simple descrip tion of recent changes in these components. The table confirms CE’s observation that the small time deposit component of M2 declined substantially during this period. Where did these funds go? Based on the negative correlation between small time deposits and liquid bank deposits, CE believe it is likely that funds moved into liquid liabilities such as demand deposits, other checkable deposits (OCDs), savings deposits and money market demand accounts (MMDAs). The deposit turnover rates that I give in the table provide additional evidence of this movement. Deposit turnover in OCDs declined substantially from 1990 to 1993, suggesting that these accounts were increasingly being used for savings rather than for transaction purposes. I believe that the cause of this intrabank shift was the reduction in the opportunity cost between small time deposits (for example, CDs) and OCDs (for example, NOWs). For instance, the average spread between six-month consumer CDs and NOW accounts was approximately 208 basis points in 1990. By November of 1993, that margin was reduced to approximately 103 basis points.1 Changes in intrabank opportunity costs that lead to flows from non-reservable to reservable deposits will increase the demand for high-powered money and thereby lead to monetary tightening, even when market rates fall. See Hutchison and Pennacchi (1992) for more discussion and empirical evidence. NOVEMBER/DECEMBER 1994 76 Table 1 Selected Components of Monetary Aggregates (in billions, seasonally adjusted) Quantity 1989.12 Change 1989.121993.11 % Change 222.7 97.2 44.6% Traveler’s checkst 6.9 1.1 15.9% Demand deposits* 279.9 105.4 Other checkable depositst 285.3 Savings and MMDAs* Money market funds (GP-BD)t Component Currencyt Turnover 1990 Turnover 1993' 37.7% 797.8 824.6 127.4 44.7% 16.5 11.7 891.0 323.6 36.3% 6.2 4.8 317.4 19 6.0% 3.02 3.02 Money market funds (IO )tt 108.8 87.9 80.8% Overnight repos & Euro$* 77.5 7.8 10. 1% Term repos & Euro$tt 178.4 Small time deposits* Large time deposits** 1152.7 548.8 -32.9 -364.5 -216.1 -18.4% -31.6% 1.0-1.53 1.0-1.53 -39.4% t Component of M1, M2, and M3 * Component of M2 and M3 t t Component of M3 1 Turnover rates are seasonally adjusted. Figure for 1993 is for month of September. 2 Turnover figure is an average for all MMMFs. See Gorton and Pennacchi (1993). 3 Reported in Collins and Edwards (1994), based on Federal Reserve Board Staff estimate. though the level of reported M2 (or M2+) may not change. Hence, these intrabank opportunity costs, in addition to the opportunity cost of holding M2 measured by the spread between the Treasury bill rate and the average rate on M 2’s components (see Figure 1 in CE), need to be considered. As the table points out, the components of current monetary aggregates have displayed very different movements over the last few years. However, a general observation that is consistent with each of the component changes in the table is that, on net, funds have flowed from non-demandable/redeemable assets into demandable/redeemable assets. This observation is also consistent with the net flow of funds into (openended) bond and stock mutual funds whose shares are redeemable as well. Perhaps this represents a permanent shift in investment behavior that is unrelated to the current slope of the term structure. Investments in money market and bond mutual funds can provide investors with rates of return that are nearly identical to investments in term CDs (the funds can hold these CDs themselves), but with the added convenience of mutual fund FEDERAL RESERVE BANK OF ST. LOUIS redeemability. One might argue that bond mutual funds now dominate a CD investment. A caveat to the redemption feature of bond and stock mutual funds is that liquidating shares may not always be costless. As ORS points out, redeeming bond and stock mutual fund shares can have capital gains tax implications (in addition to possible back-end loads). There are a number of somewhat complicated methods for calculating a mutual fund capital gain, each having possibly different tax implications. This could be a significant deterrent to frequent mutual fund withdrawals. Of course, this is not a problem affecting money market mutual funds, since they are permitted to use an “amortized cost” method of security valuation and the “penny rounding” method of share pricing that enables them to maintain a fixed share price. Therefore, I would predict that withdrawals from bond and stock funds would tend to be significantly less volatile than those of money market funds. For example, money market fund assets declined by over 30 percent during the 13-month period from December of 1982 to 77 January of 1984, a time when bank deposit interest rate ceilings were lifted. I doubt that redemptions of that magnitude are likely for bond and stock funds. However, as the empirical work of ORS suggests, the share price volatility of bond and stock funds, leading to capital gains and losses, will undoubtedly add volatility to M 2+, especially since these funds’ current 15 percent share of M 2+ is likely to grow. ORS examine the demand for balances in stock and bond mutual funds, as well as the demand for M 2+ as a whole. Following previous money demand formulations that include proxies for the opportunity cost of various monetary com ponents, they experiment with various quarterly measures of the “ex ante perceptions of returns” on stock and bond mutual funds. In my opinion, this is an exceedingly difficult empirical exercise and theoretically suspect as well. The empirical difficulty is that, unlike other components of M 2+ which have nearly risk-free returns, stock and bond funds have high rate-of-return volatility. Even if one assumed that the expected rates of return on these funds were constant, rather than varying on a quarterly basis, it could take decades of data before a reasonably accurate estimate of their expected returns could be found. These assets’ rates of return variances overwhelm their expected rates of return; that is, there is too much “noise” (variance) to be able to infer these assets’ desired “signal” (expected rate of return).2 Economic theory may also suggest that the opportunity costs of stock and bond investments will always be approximately zero. Unlike other bank liabilities that are not necessarily competi tively priced (for example, demand deposits, NOWs, MMDAs, small CDs), there is likely to be little “opportunity cost” of holding stock and bond investments, even via mutual funds. Demands for these assets should not depend on expected rates of return but risk-adjusted expected rates of return. A number of general equilibrium models predict that the best mea sure of this risk-adjusted return is the current short-term, risk-free rate, for example, the cur rent Treasury bill rate. But, of course, this then implies that the opportunity cost for competi tively priced assets will be zero. In the context of bonds, it is well known that a steeper yield curve will not necessarily indicate that (long term) bond funds are more attractive than when the yield curve is flat or inverted. A higher slope may simply reflect the expectation of rising short-term interest rates or a risk premium on more volatile long-maturity bonds. To justify their proxies, I believe the authors would need to argue that their measures reflect the actual “perceptions” of relatively unsophisticated mutual fund investors. But even if this were the case, these “m isperceptions” are likely to be temporary. Following a significant correction in stock and bond markets, investors are apt to quickly learn that recent stock performance or the steepness of the yield curve have little pre dictive power for future investment returns. Given the above difficulties in modeling and estimating risky asset demands, it is not surprising that ORS find evidence of instability in their estimated demand curve for M2+. I would not be surprised, however, if M 2+ turns out to better predict nominal GDP than other monetary aggre gates. The reason for this is that changes in M 2+ may not proxy for changes in “money” but for changes in nominal “wealth” due to the capital gains on its stock and bond components. If one views wealth as the capitalized value of future income, then changes in wealth may indeed be a good forecast of changes in nominal GDP. Hence, M 2+ may be a good indicator, but for the wrong reasons. My last comment concerns the general approach that is used to revise monetary aggre gates in response to financial innovation. CE is explicit in desiring a relatively simple aggregate that is constructed in a parallel fashion to previ ously existing aggregates. They imply that more complicated monetary indicators are problematic, because they make adjustments using a “modelbased procedure that the public (Congress in particular) might have tr o u b le understanding [quote from previous draft—Editor].” If this is the case, perhaps the Federal Reserve should (already does?) report an easily understood aggregate to the public, but then use a more complicated monetary measure for internal decision-making. In my opinion, if we truly wish to understand the inherently complex effects that financial innovation have on the conduct of monetary policy, we need to abandon the simple approach of merely adding new asset categories to old aggregates. As indicated by the turnover rates in the table and O RS’ estimated turnover rate of 2 See Merton (1980) on this point. NOVEMBER/DECEMBER 1994 78 bond and stock funds of 0.3, these new asset categories are very distant substitutes for highpowered money, the economy’s numeraire for pricing goods and services.3 As noted by ORS, the justification for adding bond and stock funds to M2 should not be based on these “balances having become a better transaction medium, but rather will be based on their substitutability for small time deposits or other M2 balances as savings vehicles.” But, if small time deposits are held for savings, rather than transaction purposes (as indicated by the turnover rate of 1.0-1.5), would it not make more sense to lessen the effects of small time deposits in a monetary measure rather than adding another, even less money-like asset? The approach represented by M 2+, which simply adds new non-monetary assets to correct problems with old non-monetary assets, is not unlike curing a hangover by having another drink. It may appear to be a good solution in the short run, but (as financial innovation continues) the ultimate consequence is a much larger hangover— or the need for an even larger drink. Developing aggregates based on their (historical) empirical fit is unlikely to be a successful endeavor in an environment in which new financial instruments continue to be developed. A more Bayesian or “model-based” approach, such as the work of Barnett (1980) and Spindt (1985) 3 Fama (1980) provides an insightful discussion of issues involving monetary policy and price-level control. FEDERAL RESERVE BANK OF ST. LOUIS would seem to be more appropriate. A general modeling of the demand for high-powered money could also potentially consider the effect of non-monetary transaction technologies, such as credit cards, or the effect of dollar-currency demand by foreigners. In other areas of economics, we do not insist that multi-good demand relations be a function of a linear combination of those goods, each having a coefficient of unity. Why do we continue this practice in monetary economics? REFERENCES Barnett, William A. “Economic Monetary Aggregates: An Application of Index Numbers and Aggregation Theory,” Journal o f Econometrics (summer 1980), pp. 11-48. Fama, Eugene F. “Banking in the Theory of Finance,” Journal of Monetary Economics (January 1980), pp. 39-57. Gorton, Gary B. and George G. Pennacchi. “Money Market Funds and Finance Companies: Are They the Banks of the Future?,” in Michael Klausner and Lawrence J. White, eds., Structural Change in Banking. Business One Inwin, 1993, pp. 173-214. Hutchison, David E., and George G. Pennacchi. “A Framework for Estimating the Value and Interest Rate Risk of Retail Bank Deposits,” Federal Reserve Bank of Chicago Working Paper WP-92-30 (December 1992). Merton, Robert C. “On Estimating the Expected Return on the Market: An Exploratory Investigation,” Journal o f Financial Economics (December 1980), pp. 323-61. Spindt, Paul A. “Money Is What Money Does: Monetary Aggregation and the Equation of Exchange,” Journal of Political Economy (February 1985), pp. 175-204. 79 Richard G. A nderson and William G. Dewald Richard G. Anderson is a research officer at the Federal Reserve Bank o f St. Louis. William G. Dewald is director of research at the Federal Reserve Bank of St. Louis. Replication and Scientific Standards in A pplied Economics a Decade A fter the Journal of Money, Credit and Banking Project L J INCE EARLY 1993, the Research Department of the Federal Reserve Bank of St. Louis has made the data and programs for articles published in the Bank’s R eview available to the public on its electronic bulletin board.1 During the first year, files from articles in the R eview were downloaded from the bulletin board more than 200 times. More recently, about 30 files have been down loaded each month. The Research Department of the Bank develops the program and data files on our bulletin board during a replication of each article prior to pub lication. A research analyst first checks the author’s data against original sources. Because databases may have been updated or revised after the research began, this can require searching for the original published data. In a few cases, data errors have been corrected, fortunately with only minor impact on the author’s results. Next, The bulletin board is advertised as the Federal Reserve Economic Database, or FRED. FRED’s phone number is (314) 621-1824. (The Federal Reserve System does not have a server on the Internet.) Dewald, Thursby and an annotated version of the computer program is prepared and all statistical results recalculated. Finally, bibliographic and other references are checked by the analyst against original source documents. We believe this practice both assures the accuracy of the empirical results and allows the interested reader to delve into the details o f th e a u th o r ’s re s e a r c h . THE ROLE OF DATA IN ECONOMIC EXPERIMENTS Although empirical knowledge in both the physical and social sciences arises from repeated experiments, the role of data differs. In the physical sciences, scientists control a relatively small number of variables such as temperature, atmospheric pressure, diet or family characteristics. Since some variables are neither observed nor controlled, no two repetitions of an experiment w ill be Anderson (1986) summarized the Journal of Money, Credit and Banking project mentioned in the title. NOVEMBER/DECEMBER 1994 80 identical. Response surface analysis and the newer field of research synthesis provide tools for ana lyzing the dependence of experimental results on the settings of the conditioning variables.2 In economics, however, unlike the physical sciences, researchers can only condition on the observed values of the environmental variables, not control them. Consider a simple model of an economic experiment: 1. Form hypotheses. 2. Collect data. 3. Develop theoretical and econometric framework. 4. Estimate. 5. Test hypotheses, draw conclusions. The values of the conditioning variables are collected in step 2. Published articles typically describe steps 1, 3 and 5, but are most often silent on step 2. In principle, a researcher armed with the values of the conditioning variables and the computer code for step 4 should be able to exactly reproduce an economic experiment.3 Unlike the physical sciences, the experiment is deterministic, given the data. Appraising the robustness of the results of an economic experiment requires knowing the values of the conditioning variables used by the researcher. Obtaining the data may sometimes be difficult. Datasets and programs may be mislaid or lost during the interval between completion of the research and publication of an article. Further, re q u e sts to a u th o rs fo r d ata may ra is e s u s p ic io n s that the reader hopes (or expects) to find errors in the authors’ research. An individual researcher has strong incentives not to share data and pro grams. If the materials are shared and results confirmed, the confirmation provides little (if any) reward to the researcher beyond the original publication of his findings. If results are found faulty, however, the researcher faces the likeli hood of some professional embarrassment. 2 See Cooper and Hedges (1994). 3 A complication not dealt with here are errors and inconsis tencies in econometric computer programs. In the Journal of Money, Credit and Banking project (described further below), we requested that authors provide the version, release date and serial number of the computer program used for their estimation. See Lovell and Selover (1994) for examples of the variation in econometric packages. 4 Various classification schemes and nomenclatures have been discussed by Kane (1984), Mittelstaedt and Zorn (1984), Hubbard and Vetter (1991), Lindsay and Ehrenberg (1993), and Fuess (1994). FEDERAL RESERVE BANK OF ST. LOUIS The trepidation of authors aside, scientific progress depends on challenging received wisdom. In applied economics, these challenges fall into three categories: replication of published results using the previous authors’ data and programs; applying new statistical methods or techniques to authors’ datasets; and application of existing statistical methods (including those used by previous authors) to new datasets.4 That most applied economic research falls within the third category is not surprising, since the first two depend on access to previous authors’ datasets. Only with the authors’ data may the reader repeat, or replicate, all five steps of the scientific experiment. Selecting a new set of values for the conditioning variables from published sources may yield results close to those obtained by the author, or results that are quite different. Unfortunately, it is difficult to predict the sensi tivity of authors’ results to variations in the values of the conditioning variables. For an example of the wide range of results that may arise when mixing different sources and vintages of data, see the computer simulation experiment reported in Dewald, Thursby and Anderson (1986). THE JMCB PROJECT AFTER 10 YEARS The Jou rn al o f M oney, C redit a n d B an kin g project, conducted from 1982-84 at the editorial offices of the JMCB at The Ohio State University in Columbus, was the first attempt by a profes sional journal to make authors’ programs and data available to its readers.5 During the project, the JMCB asked authors to submit data and pro grams to the journal’s office. Conceptually, we regarded the research reported in each article as the outcome of an experiment. A complete understanding of the experiment required the researchers’ data and computer programs, as well as the published summary descriptions and conclusions. For a subset of these submissions, we attempted to repeat steps 2 and 4 by collecting data from the sources cited by the authors and re-running the authors’ computer programs. 5 The primary research team was William Dewald, Jerry Thursby, Richard Anderson and Hashem Dezbaksh. The project’s findings are summarized in Dewald, Thursby and Anderson (1986). The project was supported in part by the National Science Foundation. 81 concluded that it is important for journals to request data from authors immediately following completion of the research, and for the journal to retain the data to avoid its loss during the interval between completion of the research and publication of the paper. T a b le 1 Datasets for the JMCB Project Datasets Requested by Month and Year through May 1989 1983 84 85 86 87 88 89 Jan. 0 0 12 3 9 5 8 Feb. 2 1 1 4 9 5 1 1 April 14 1 2 8 4 6 March 0 0 2 6 0 1 1 2 May 1 0 14 1 1 5 June 1 2 6 5 2 July 0 0 4 5 2 3 2 4 3 3 2 Aug. 0 2 1 2 4 Sept. 9 4 4 2 Oct. 11 6 7 38 Nov. 1 2 6 4 3 4 Dec. 0 1 6 0 3 1 Datasets Available by Issue as of May 1989 1980 81 82 83 84 85 86 87 88 Feb. 2 4 6 4 2 5 5 6 Aug. 1 4 6 2 6 2 6 Nov. 4 0 3 3 5 8 3 3 2 1 5 5 2 May 3 3 7 7 5 7 1 3 Total 8 10 12 24 24 17 19 10 17 SOURCE: JMCB editorial office staff. No later data are available from the JMCB staff. We found during the JMCB project that many authors could not furnish data and programs following publication of their articles. We in i tially requested data from the authors of 62 articles published during 1980-82, prior to the beginning of the project on July 1, 1982. About one-third of the authors did not respond to either a first or second request for data. Among the responding authors, one-half either could not locate their data or chose not to submit them. Most of these authors said that they could have done so if the materials had been requested when the manu script was first submitted to the JMCB. Data and programs were often apparently mislaid during the relatively long delay between completion of the research and publication of its findings. We next requested data from the authors of papers that either had only recently been accepted for publication or were under editorial review. More than three-fourths furnished their data. We In the second part of the JMCB project, we studied whether the materials submitted by authors were in fact sufficient to allow another researcher to repeat their experiment. For many articles, repeating step 2—searching for the authors’ data in their stated sources—was impossible. Descriptions of sources were either too vague to allow us to locate the data and/or the data were not included in the cited sources. Although 54 datasets were submitted to the JMCB during the project, we judged only eight as satisfactory and 14 as valueless in helping us understand the authors’ published work. Others were deficient in at least one important respect. For a few articles, we discovered data errors during comparison to published sources. In the most severe case, we found that an author’s con clusions were reversed (prior to publication of the article) when an error was corrected. Where the data were adequate, we usually obtained numerical results from authors’ programs very close to those reported by the authors. Beyond encouraging readers to explore authors’ methodology and the robustness of published results, we believed that requesting data and programs from authors would encourage them to exercise added care during their research. We also expected that other journals would adopt similar requirements to increase the value of their articles to readers. Although the JMCB project stimulated discussion of the role of replication in economics, no other journal adopted a policy of requesting data from authors during the 1980s. Some journals adopted editorial statements that authors should stand ready to provide data and programs to other researchers. Such statements, in and of themselves, may not solve the two major problems identified during the project: Data often are mislaid prior to publication of the article, and the author may be suspicious of the motives of a researcher requesting his data and programs. A decade after the JMCB project, the replication of previous studies as a part of new research seems an infrequent occurrence. During the last decade, no papers or notes in the JMCB have focused primarily on replication, and only about two papers per year have included a direct compari NOVEMBER/DECEMBER 1994 82 son of the authors’ results to those in previous studies, whether published in the JMCB or else where.6 The collection of data by professional journals also remains rare. The JMCB discontin ued requesting data from authors in 1993. To our knowledge, today only two journals—the Jou rn al o f A p p lied E con om etrics and the Jou rn al o f B u sin ess a n d E co n o m ic Statistics—routinely request data from authors, and neither requests their programming. From January 1983 through mid-1989, the JMCB received nearly 150 submitted datasets and about 300 requests, as shown in Table l . 7 Except for a surge in requests following the publication of Dewald, Thursby and Anderson (1986), on balance only a few datasets were requested each month even though the number of available datasets increased significantly during the decade. The higher request rate during the past two years for data from the St. Louis bulletin board may suggest that the modest costs of requesting data from the JMCB still exceeded the marginal value to an individual researcher of replicating a previous study. To obtain data for a JMCB article, a reader had to call the editorial office to ask the price of the data, submit payment by mail, wait for the data to be reproduced and mailed, and perhaps re-enter the data into a computer. By contrast, the St. Louis bulletin board is free, delivery is immediate, and data are machine-readable. THE ROLE OF THE NATIONAL SCIENCE FOUNDATION The economics program of the National Science Foundation (NSF) has sought to build a heightened awareness of the value of data collection, archival and distribution among economists during the last decade. Following publication of Dewald, Thursby and Anderson (1986), the NSF established an archive for the storage and distribution of authors’ data at the Inter-university Consortium for Political and Social Research (ICPSR) at the University of Michigan. Initially, some anticipated that the N SF’s effort would extend the JMCB’s practice of requesting and distributing authors’ data to 6 Replication also has been relatively rare even in journals that encourage submission of such papers. See Fuess (1994). 7 Recall that in 1982 we began requesting data for articles that had been published as early as 1980. 8 Some authors have since proposed models of how such col lective disinterest among professional journals might arise and be sustained. See Feigenbaum and Levy (1994, 1993). FEDERAL RESERVE BANK OF ST. LOUIS a much larger number of journals. The editors of 22 journals, however, declined invitations from N SF’s economics program to request that their authors place data in the ICPSR archive.8 The National Science Foundation has also adopted guidelines to reduce the cost of replica tions. The guidelines require that authors place any data used and/or developed in conjunction with an NSF-funded project in a public archive not later than six months following the end of the grant period. Applications for additional NSF funds must contain a statement of how the author has complied with this requirement. The ICPSR accepts data from any author who has received NSF funds.9 National Science Foundation initiatives have also assisted users of copyrighted and confidential data. Some data obtained by researchers from commercial vendors are copyrighted and may not be further distributed by the researcher without the vendor’s permission. One such vendor, the Center for Research in Security Prices (CRISP), has agreed to maintain researchers’ datasets as part of its own database and make them available to all licensed users of its data. For confidential data, the Bureau of the Census and the NSF are exploring opening regional offices that would allow researchers access to confidential data, including datasets used in previously published studies. A pilot office is operating in Boston. REPLICATION OF THE JMCB PROJECT IN ST. LOUIS Our experience at the JMCB during 1982-84 was itself only one trial of an experiment. Would another sample of authors also have difficulty providing data following publication of their articles? Or were our original findings anom alous, leading us to greatly exaggerate the problem, as some critics have suggested? During 1992-93, we repeated the JMCB experi ment at St. Louis, in part, by requesting data and programs from the authors of papers presented at the Bank’s annual economic policy conference in October 1992. We did not tell authors prior 9 Materials should be submitted to User Support, ICPSR, P.O. Box 1248, Ann Arbor, Ml 48106. Data may be retrieved from ICPSR via an Internet server; see Goffe (summer 1994; March 1994). 83 to the conference that we would be requesting their programs and data. Their responses were very similar to those we had observed at the JMCB a decade earlier. Authors who had com pleted their studies just prior to the conference promptly submitted their data and programs, while those who had largely completed their research as much as several years earlier found it costly to organize and submit data. The fol lowing year, we informed participants for our 1993 conference at the time their paper was invited that we expected data to be submitted with the manuscript. Authors generally found it imposed little burden to submit data and programs with their manuscripts so lon g as th ey w ere aw are o f the requ irem en t in ad v an ce, although the Bank’s staff had to make some follow-up calls to clarify documentation. THE FUTURE ROLE OF REPLICATION How should economists, as scientists, interpret their profession’s apparent, collective lack of interest in replicating prior studies? Does the failure of journals to request and distribute data and programs reflect a lack of scientific discipline in economics, as some have suggested? Do the costs to a reader of obtaining data from authors exceed the benefit? The JMCB project demonstrated that profes sional journals could reduce the costs of replica tion and improve the quality of applied economic research by collecting data and programs from authors. For the first time, the reader of an empirical article could obtain data and programs anonymously (with respect to the author) from the journal that published the article. Further, collection of the authors’ data and programs seemed to encourage additional care by authors to avoid inadvertent errors during the conduct of the research. Since January 1994, the Jou rn al o f A p p lied E conom etrics has accepted papers for publication conditional on authors furnishing an acceptable dataset. The publication reviews datasets for completeness and then places them on an Internet server at Queens University.10 The Jou rn al o f B u sin ess a n d E co n o m ic Statistics requests, but does not require, that authors submit data for distribution via an Internet server at Duke University. Tauchen (1993) argues, as we did in 1986, that readers of journals should be interested in authors’ data and programs, and a journal’s prestige (and circulation) should increase when such data and programs are made readily available to readers. We believe that published empirical research generally would benefit if the practices of the Jou rn al o f A p p lied E con om etrics, Jou rn al o f B u sin ess a n d E con om ic Statistics, and the St. Louis R eview w ere more widely adopted. REFERENCES Cooper, Harris, and Larry V. Hedges, eds. The Handbook of Research Synthesis. New York, 1994. Dewald, William G., Jerry G. Thursby, and Richard G. Anderson. “Replication in Empirical Economics: The Journal of Money, Credit and Banking Project,” American Economic Review (September 1986). Feigenbaum, Susan, and David M. Levy. T h e Self-Enforcement Mechanism in Science,” mimeo, paper presented at the American Economic Association Meetings, Boston, Massachusetts, January 1994. _____ , and_____ . “The Market for (Ir)Reproducible Econometrics,” Social Epistemology (no. 3,1993), pp. 243-44. Fuess, Scott M. “On Replication in Business and Economic Research: The QJBE Case,” paper presented at the Eastern Economic Association meetings, Boston Massachusetts, March 1994 (mimeo, Department of Economics, University of Nebraska, October 1993). Goffe, William L. “Computer Network Resources for Economists,” Journal of Economic Perspectives (summer 1994). _____ . “Resources for Economists on the Internet,” electronic document maintained on the Internet (March 1994 version). Contact bgoffe@whale.st.usm.edu. Hubbard, Raymond, and Daniel E. Vetter. “Replications in the Finance Literature: An Empirical Study,” Quarterly Journal o f Business and Economics (autumn 1991). Kane, Edward J. “Why Journal Editors Should Encourage the Replication of Applied Econometric Research,” Quarterly Journal of Business and Economics (winter 1984). Lindsay, Murray R., and A.S.C. Ehrenberg. ‘T he Design of Replicated Studies,” The American Statistician (August 1993), pp. 217-28. Lovell, Michael, and David Selover, “Econometric Software Accidents,” The Economic Journal (May, 1994), pp. 713-25. MacKinnon, James. “Guidelines for Users: Journal of Applied Econometrics FTP Archive Site,” Journal of Applied Econometrics (April-June 1994), pp. 229-30. Mittelstaedt, Robert A., and Thomas S. Zorn. “Econometric Replication: Lessons from the Experimental Sciences,” Quarterly Journal o f Business and Economics (winter 1984). Tauchen, George. “Remarks on My Term at JBES,” Journal of Business & Economic Statistics (October 1993). 10 For discussion of the Journal of Applied Econometrics server, see MacKinnon (1994). NOVEMBER/DECEMBER 1994 84 FEDERAL RESERVE BANK OF ST. LOUIS R E V IE W \UDEX 1994 JANUARY/FEBRUARY R. Alton Gilbert, “Federal Reserve Lending to Banks That Failed: Implications for the Bank Insurance Fund." David C. Wheelock and Paul W. Wilson, “Can Deposit Insurance Increase the Risk of Bank Failure? Some Historical Evidence.” James B. Bullard, “Measures of Money and the Quantity Theory.” JULY/AUGUST Daniel L. Thornton, “Financial Innovation, Deregulation and the “Credit View” of Monetary Policy.” Sangkyun Park, “Explanations for the Increased Riskiness of Banks in the 1980s.” MARCH/APRIL Patricia S. Pollard, “Trade Between the United States and Eastern Europe.” “Money Stock Measurement: History, Theory and Implications.” Proceedings of the Eighteenth Annual Economic Policy Conference. Richard G. Anderson, ed. John C. Weicher, “The New Structure of the Housing Finance System.” Richard G. Anderson and Kenneth A. Kavajecz, “A Historical Perspective on the Federal Reserve’s Monetary Aggregates: Definition, Construction and Targeting.” Kenneth A. Kavajecz, “The Evolution of the Federal Reserve’s Monetary Aggregates: A Timeline.” K. Alec Chrystai and Ronald MacDonald, “Empirical Evidence on the Recent Behavior and Usefulness of Simple-Sum and Weighted Measures of the Money Stock.” Douglas Fisher and Adrian Fleissig, “Money Demand in a Flexible Dynamic Fourier Expenditure System.” William A. Barnett and Ge Zhou, “Financial Firms’ Production and Supply-Side Monetary Aggregation Under Dynamic Uncertainty.” Jerome L. Stein, “Can the Central Bank Achieve Price Stability?” Michael J. Boskin, Philip H. Dybvig, and Bennett T. McCallum, “A Conference Panel Discussion.” (Commentaries by Charles W. Calomiris, Charles R. Nelson, James L. Swofford, William C. Brainard, and Frederic S. Mishkin.) Alvin L. Marty, “The Inflation Tax and the Marginal Welfare Cost in a World of Currency and Deposits.” SEPTEMBER/OCTOBER Joseph A. Ritter, “Job Creation and Destruction: The Dominance of Manufacturing.” Peter Yoo, “Boom or Bust? The Economic Effects of the Baby Boom.” Christopher J. Neely, “Realignments of Target Zone Exchange Rate Systems: What Do We Know?” R. Alton Gilbert, “A Case Study in Monetary Control: 1980-82.” NOVEMBER/DECEMBER “Mutual Funds and Monetary Indicators.” Proceedings from a symposium at the Federal Reserve Bank of St. Louis. Sean Collins and Cheryl L. Edwards, “Redefining M2 to Include Bond and Equity Mutual Funds.” MAY/JUNE Athanasios Orphanides, Brian Reid, and David H. Small, “The Empirical Properties of a Monetary Aggregate That Adds Bond and Stock Mutual Funds to M2.” Helmut Schlesinger, “On the Way to a New Monetary Union: The European Monetary Union” (the Eighth Annual Homer Jones Memorial Lecture). (Commentaries by William A. Barnett and Ge Zhou, Jacob S. Dreyer, Josh Feinman, John V. Duca, and George G. Pennacchi.) Clemens J.M. Kool and John A. Tatom, “The P-Star Model in Five Small Economies.” Charles W. Calomiris, “Is the Discount Window Necessary? A Penn-Central Perspective.” FEDERAL RESERVE BANK OF ST. LOUIS Richard G. Anderson and William G. Dewald, “Replication and Scientific Standards in Applied Economics A Decade After the Journal of Money, Credit and Banking Project.” Federal Reserve Bank of St. Louis Post Office Box 442 St. Louis, Missouri 63166
@FedFRASER