The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Federal Reserve Bank of New York Quarterly Review A utum n 1991 1 Volum e 16 N um ber 3 Do In te rn a tio n a l R eactions of S tock and Bond M arkets R eflect M a cro e co no m ic F u n d a m e n ta ls? 14 The C ost of C a p ita l fo r S e cu ritie s Firm s in the U nited S tates and Japan 28 F inancial L ib e ra liza tio n and M o n e ta ry C ontrol in Japan 47 E xpected Inflation and Real Interest R ates Based on Indexlinked Bond P rices: The U.K. E xperience 61 Tracking the E conom y w ith the P urchasing M anagers’ Index 70 Treasury and Federal R eserve Foreign E xchange O p e ra tio n s Federal Reserve Bank of New York Quarterly Review Autumn 1991 Volume 16 Number 3 Table of Contents 1 Do International Reactions of Stock and Bond Markets Reflect Macroeconomic Fundamentals? Eli M. Remolona Using quarterly data for the United States, Japan, and the United Kingdom, this article investigates the relationship between changes in fundamental economic conditions and the international reactions of stock and bond markets. The author assesses the degree to which the transmission of information about fundamentals prompts the stock or bond markets in different countries to move together. 14 The Cost of Capital for Securities Firms in the United States and Japan R obert N. M cCauley and Steven A. Zim m er The authors use stock market valuations to construct estimates of the cost of capital for five U.S. and four Japanese securities firms in 1982-91. They seek explanations for the observed capital cost differences in mac roeconomic, risk, policy, and industrial organization factors. Their analysis also contrasts the gap in capital costs between U.S. and Japanese securities firms with the corresponding gap for industrial firms and banks. 28 Financial Liberalization and Monetary Control in Japan Bruce Kasman and Anthony P. R odrigues This article examines the effects of financial market reform on monetary control in Japan during the past two decades. The authors identify changes in the Bank of Japan’s operating strategy and evaluate the Bank’s ability to influence key policy variables in the current liberalized environment. Table of Contents 47 Expected Inflation and Real Interest Rates Based on Index-linked Bond Prices: The U.K. Experience G abriel de Kock Some analysts have argued that indexed bonds convey important informa tion for the formulation of monetary policy. This article investigates whether a market measure of expected inflation derived from British indexed gilt prices would be useful in predicting future inflation and real economic activity. 61 Tracking the Economy with the Purchasing Managers’ Index Ethan S. Harris The purchasing managers’ index is a widely watched but virtually untested indicator of manufacturing activity. This article examines how well the index lives up to its billing as a leading indicator. The author also explores whether the index supplies information about the economy beyond that already provided by other indicators. 70 Treasury and Federal Reserve Foreign Exchange Operations A report for the period May-July 1991. 76 List of Recent Research Papers Do International Reactions of Stock and Bond Markets Reflect Macroeconomic Fundamentals? by Eli M. Remolona Much of the movement in stock and bond markets appears to be independent of changes in fundamental economic conditions. Stock prices in particular often fluctuate more sharply than would be justified by shifts in the underlying fundamentals. Such market volatility may distort the economy’s allocation of capital and on occasion lead to liquidity crises and macroeconomic instability. This article investigates one possible source of the volatility: a tendency of market participants to overreact to developments in other markets. The analy sis yields some evidence that stock and bond markets move together to an extent not easily explained by changes in macroeconomic fundamentals. Empirical work by Campbell and Shiller, Poterba and Summers, West, and Bulkley and Tonks supports the notion that the volatility of U.S. and U.K. stock prices exceeds the volatility implied by fundam entals.1 Schwert finds that the timing of volatility in the U.S. stock market often fails to coincide with that of mac roeconomic fundamentals.2 A related finding by Bennett and Kelleher is that high volatility in stock markets 1For U .S. stocks, see John Y. C am p b ell and Robert J. Shiller, "C o integration and Tests of Present Value M odels," J o u rn a l o f P o litic a l E co n om y, vol. 95 (1987), pp. 1062-88; Jam es M. Poterba and Law rence H. Sum m ers, “M ean Reversion in Stock Prices: Evidence and Im p lic atio n s ,” J o u rn a l o f F in a n c ia l E c o n o m ic s , vol. 22 (1988), pp. 2 7 -5 9 ; and Kenneth D. West, “A Specification Test for S p eculative B ubb les," Q u a rte rly J o u rn a l o f E c o n o m ic s , vol. 102 (1987), pp. 5 5 3 -8 0 . For U.K. stocks, see G eorge Bulkley and Ian Tonks, "Are U.K. Stock Prices Excessively Volatile? Trading Rules and V ariance Bound Tests," E c o n o m ic J o u rn a l, vol. 9 9 (1989), pp. 1 08 3 -9 8. 2G. W illiam Schw ert, “Why D oes Stock M arket Volatility C hange over Time?" J o u rn a l o f F in a n c e , vol. 44 (D e c e m b e r 1989), pp. 1115-53. tends to be associated with high return correlations across the markets.3 This result may arise because stock markets have overreacted to one another. When one market departs from fundamentals, the mistake is propagated to other markets, thus giving rise to the excess volatility. Moreover, Shiller and Beltratti find that domestic stock markets often overreact to domestic bond markets.4 Hence, unaccountable movements in one market can be compounded as they spread to other markets at home and abroad. Consider an extreme example of high volatility: the October 1987 stock market break. The global nature of the event could be seen as evidence that stock markets were overreacting to one another. Alternatively, some would argue that stock investors at the time were simply responding to information contained in price changes in other markets, particularly information on expectations of real activity abroad.5 But conclusions based on a 3Paul B ennett and Jea n ette Kelleher, "The International Transmission of Stock Price D isruption in O c to b e r 1987," this Q u a rte rly R eview , Sum m er 1988, pp. 17-33. See also Paul K upiec, "Financial Liberalisation and International Trends in Stock, C orp o ra te Bond, and Foreign E xchange M arket Volatilities," O E C D D ep artm e n t of Econom ics and Statistics Working Papers, no. 94, February 1991. ♦Robert J. Shiller and A ndrea E. B eltratti, "Stock Prices and Bond Yields: Can Their C om ovem ents Be E xplained in Terms of Present Value M odels?" National Bureau of Econom ic R esearch, Working Paper no. 3 4 6 4 , 1990. sSee R ichard W. Roll, "The International Crash of O c to b e r 1987," in Robert Kam phuis, R oger K orm endi, and J.W. H en ry W atson, eds ., B la c k M o n d a y a n d th e F u tu re o f F in a n c ia l M a rk e ts (M id -A m e ric a Institute, 1988), pp. 3 5 -7 0 ; and M ervyn A. King and Sushil W adhani, “Transmission of Volatility betw een Stock M arkets," R evie w o f F in a n c ia l S tu d ie s , vol. 3 (1990), pp. 5 -36 . FRBNY Quarterly Review/Autumn 1991 1 sin g le event are u n co n vin cin g , and for this reason, it is im p o rta n t to co n sid er w h e th e r the broader experience of sto ck and bond m arket co-m ovem ent is c o n siste n t w ith m o ve m e n ts in real a c tiv ity , in te re s t rates, and inflatio n. T his a rtic le exam ines n e a rly tw o decades of m arket expe rie n ce in the U nited S tates, Japan, and the U nited K ingdom . The d iscu ssio n first focuses on the degree to w hich m acroe co nom ic fu n d a m e n ta ls and foreign m arket m o ve m e n ts e xp la in d o m e s tic m a rke t m o ve m e n ts. It then asks w h e th e r a h yp o th e sis based on m arket m ove m ents conveying in fo rm a tio n about fu n d a m e n ta ls can explain the strong in te rn a tio n a l reactions m arkets have to one another. Of the d om e stic fu n d a m e n ta ls, future real a c tivity is found to be the main force d rivin g the stock m arkets, and future inflation, the main force driving the bond m ar kets. Foreign fu n d a m e n ta ls a ppear to exe rt no direct im pact on do m e stic m a rkets. A fte r taking account of the e ffe cts of both do m e stic and foreign fu n d a m e n ta ls, the a n a lysis finds tha t fo re ig n m arket returns still explain m uch of the rem aining variance of do m e stic returns. S u p e rfic ia lly , the s u b s ta n tia l re sid u a l e x p la n a to ry pow er of foreign m arket m ovem ents m ight appear ipso fa cto e vid e n c e th a t d o m e s tic m a rke ts o ve rre a ct to m ovem ents in foreign m a rkets. But there is a fu rth e r possibility, nam ely th a t the m ovem ents in foreign m ar k e ts c o n v e y a d d itio n a l in fo rm a tio n re le v a n t to th e d o m e stic m arkets. T h is in fo rm a tio n may bear on those d o m e stic fu n d a m e n ta ls th a t influence d o m e stic m ar kets. T hus, before a ve rd ict of “ o ve rre a ctio n ” by dom es- Chart 1 Links among Markets and Fundamentals Note: (A) and (B) represent the links between markets and fundamentals, (C) represents the markets' reactions to each other, and (D) represents the links between domestic and foreign fundamentals. Digitized for 2 FRASER FRBNY Q uarterly Review/Autumn 1991 tic m arkets to fo re ig n m a rke ts can be b ro u g h t, this p o s s ib ility m ust be exam ined. The sig n ific a n t re la tio n s h ip s found to exist am ong do m e stic and foreign fu n d a m e n ta ls and d o m e s tic and foreign s e c u ritie s m a rke ts are illu s tra te d in C h a rt 1. The link (A) represents the im pact of d o m e s tic fu n d a m e n ta ls on the d o m e stic m a rke t and (B) the im p a ct of foreign fu n d a m e n ta ls on the foreign m arket. T he lin k (C) rep re se n ts the re a c tio n s of th e m a rk e ts to e ach other. B ecause fu n d a m e n ta ls in one c o u n try are found to have no direct bearing on the m a rke t in the o th e r country, a p p ro p ria te reactions of d o m e stic m a rke ts to foreign m arkets should reflect links betw een d o m e s tic and fo r eign fundamentals, shown as (D). The hypothesis implied by these relationships is that the domestic market tries to infer information about foreign fundam entals from foreign markets to better predict domestic fundamentals. To d e te rm in e w h e th e r the actual m a rke t re a ctio n s are co n siste n t w ith th is h yp o th e sis, each of th e id e n tified links is estim a te d . Testing the h y p o th e s is involves c o m paring e stim a te s of the a ctual m a rke t re a ctio n s w ith the reactions im plied by the e stim a te s of the links betw een fu n d a m e n ta ls and the links betw een m a rke ts and fu n d am entals. The a n a ly s is fo cu se s on real a c tiv ity as the main fu n d a m e n ta l fo r sto ck m a rke ts and on in fla tio n as the co rre sp o nd in g fu n d a m e n ta l fo r bond m a rke ts. The estim ates su g g e st th a t the U.S. and J a p a n e se sto ck m arkets react to each o th e r’s m o ve m e n ts on average more than w ould be ju s tifie d by the in fo rm a tio n th e se m arkets convey a bout real activity, w h ile the U .S ., J a p anese, and U.K. bond m a rke ts react to one a n o th e r’s m ovem ents more than w ould be ju s tifie d by the in fo rm a tion these m a rke ts convey a b o u t in fla tio n. Explaining stock and bond returns Id e n tify in g the fu n d a m e n ta ls M arket p rices of sto cks and bonds can be sp e cifie d as the present d is c o u n te d values of stre a m s of expected cash flow s. H ence sto ck and bond re tu rn s sh o u ld be affected by m a cro e co n o m ic fa cto rs th a t re fle ct d is c o u n t rates or expected cash flow s. A t the sam e tim e , the strength of th e se e ffe cts sh ould d iffe r betw e e n sto ck and bond m a rkets, p a rtic u la rly the e ffe c ts of exp ecte d real a c tivity and expected in fla tio n. D iffe re n ce s b etw een the m arkets allow us to iso la te the e ffe c ts of one fu n d a m ental on one m arket by using the o th e r m a rke t to control for the e ffe cts of o th e r fu n d a m e n ta ls . For stocks, cash flow s m ay ta ke the form of d ivid e n d paym ents, pro ce e d s from sto ck bu yb a cks by the issu ing firm , or p roceeds from sto ck p u rch a se s by ta ke o ve r fir m s .6 P re v io u s s tu d ie s s h o w th a t s e v e ra l m a c 6Cash flows from stock bu ybacks and takeovers becam e im po rtan t in the U nited States in the 1980s. ro e cono m ic va ria b le s — the d ivid e n d yield, the spread betw een long and sh o rt inte re st rates, inflation, and real a c tiv ity — s ig n ifica n tly influ e nce re tu rn s.7 The d ivid e n d yield and term spread should reflect d isco u nt rates and b u siness co n d itio n s. E xpected inflation should affect real d isco u n t rates, but it may also affect cash flow s. Of th e se m a c ro e c o n o m ic v a ria b le s , fu tu re real a c tiv ity sh o u ld e xe rt the s tro n g e s t in flu e n c e on sto ck cash flow s. U nlike stocks, bonds in th is study have cash flow s th a t are fixed in nom inal term s. C hanges in the s h o rt term real in te re st rate a ffectin g the d isco u nt rate should have a s tro n g e r e ffe ct on bond returns than on stock re tu rn s .8 E x p e c te d in fla tio n s h o u ld e x e rt an even stro n g e r in flu e nce on bond returns b e ca u se of its direct effect on the real value of in te re s t and p rin c ip a l pa y m ents. By co n tra st, real a c tiv ity sh o u ld have re la tive ly little e ffe ct on bond cash flow s, p a rtic u la rly in the case of g o v e rn m e n t b o n d s not s u b je c t to d e fa u lt ris k , although it may affect d is c o u n t rates. D e s c rip tio n o f key va ria b le s The key v a ria b le s fo r th is s tu d y are e x c e s s s to c k returns, excess bond returns, real a c tiv ity g row th, and inflation rates (Table 1). E xcess returns are the returns over a q u a rte r m inus a th re e -m o n th in te re s t rate at the b e g inning of the q u a rte r.9 The a n a ly s is relies on quar- 7See in p a rticu la r Nai-Fu Chen, Richard Roll, and Stephen A. Ross, "E conom ic Forces and the Stock M arket," Journal of Business, vol. 59 (1986), pp. 383-403; and Eugene F. Fama, "Stock Returns, E xpected Returns, and Real A ctivity," Journal of Finance, vol. 45 (1990), pp. 1089-1108 Chen, Roll, and Ross also identify the yield spread between low -grade and high -grade bonds as a significant variable, but Fama finds that the d ivide nd yield does just as well as the the risk spread. The estim ates presented here will use the divide nd yield because data for this variable are available for foreign markets. Shiller som etim es uses the price-earnings ratio instead of the d ivide nd yield. 8See John Y. C am pbell and John Ammer, "W hat Moves the Stock and Bond Markets? A Variance D ecom position for Long-Term Asset Returns," paper presented at the annual m eeting of the A m erican Finance A ssociation, W ashington, D C.. D ecem ber 1990. 9Stock returns are measured as the divid e n d yield plus the change in the log stock price index Bond returns are ap proxim ated by the yield at the beginning of the qu arte r minus the cha nge in log yields. Table 1 Statistical C haracteristics of Key Variables (Q uarterly Data at an Annual Rate, June 1973 to S eptem ber 1989) Excess Stock Returns* U nited States Japan U nited K ingdom Excess Bond Returns* Mean Standard Deviation Mean Standard Deviation 5.27 15.65 19.98 37.56 32.64 50.30 0.23 3,97 0 82 30.30 38.66 40.10 U nited States Japan U nited States 0.62 0.38 0.35 0.45 0.20 Stock Return C orrelations Japan U nited K ingdom B ond Return C orrelations 0.42 Real Growths Mean U nited States Japan United Kingdom Inflation Standard Deviation 2,50 3.79 1.88 4.08 3.24 8.90 United States 0,30 0 05 S tandard Deviation Mean 6.42 5.26 9.84 Real Growth C orrelations Japan U nited Kingdom Japan 3.89 6.19 7.26 Inflation C orrelations Japan 0.07 U nited States 0.57 0.55 Japan 0.52 ^Excess stock returns are stock returns minus a three-m onth interest rate. ^Excess bond returns are bond returns minus a three-m onth interest rate. § lo g change in real GNP or GDP. FRBNY Q uarterly R eview/Autum n 1991 3 te rly d ata from June 1973 to S eptem ber 1990 for the U nited S tates, Japan, and the U nited K in g d o m .10 C hart 2 show s the beh avior of excess stock returns and C hart 3 the beh avior of excess bond returns over the p e rio d .11 U sing excess returns a llow s us to abstract from possi10The analysis looks at stock and bond returns only up to the third q u a rte r of 1989 because these returns are related to leads of m acroeconom ic variables that go up to the third qu arter of 1990. 11The lack of movement in excess bond returns in Japan in the early 1970s seems to reflect a m arket sub je ct to “ g u id a n c e ” by the m onetary authorities. Efforts to take acco unt of this period with the use of dum m y variables did not alter the analysis. 4 FRBNY Q uarterly Review/Autumn 1991 ble e ffe c ts of c ro s s b o rd e r s h o rt-te r m in te re s t-ra te a rb itra g e and to focus on the m a cro e co n o m ic v a ria b le s m ost relevant to sto ck and bond m a rke ts. Real grow th is m easured by GNP or GDP, in fla tio n by th e co n su m e r price index. As m easured by th e ir sta n d a rd d e v ia tio n s , excess stock returns are o n ly so m e w h a t m ore vo la tile than excess bond re tu rn s . S to c k re tu rn s , how ever, seem m ore c o rre la te d a cro ss m a rk e ts th a n bon d re tu rn s, except betw een Japan and the U nited K in g d o m . Real grow th and in flation seem a b o u t e q u a lly v o la tile , but they are c le a rly less v o la tile th a n s to c k and bond returns. Real grow th also te n d s to be m uch less corre- lated across co u n trie s than does inflation. S tock returns tend to be sig n ific a n tly more co rrelated than real grow th across co u n trie s, w hile bond returns tend to be less co rrelated than inflation. E x p la in in g s to c k re tu rn s To id e n tify the s ig n ifica n t links am ong sto ck m arkets and fu n d a m e n ta ls, th is a n a lysis evaluates the degree to w hich m a cro e co no m ic varia b le s and foreign sto ck m ar ket m ovem ents explain do m e stic stock m arket m ove m ents. Table 2 rep orts a d ju ste d R -squared s ta tis tic s as m easures of the e xp la na to ry pow er of reduced-form sto ck retu rn e q u a tio n s . The d e p e n d e n t v a ria b le is excess stock returns, or the returns m inus a three- m onth in te re st rate. The e x p la n a to ry va ria b le s com m on to all th e e q u a tio n s are th e la g g e d d iv id e n d y ie ld , excess bond returns, and fo u r q u a rte rly leads of real grow th. The a d ju ste d R -squared s ta tis tic s fo r the basic equation are 20 p e rce n t fo r the U.S. m a rke t, 16 p e rce n t for the Ja p a n e se m arket, and 25 p e rc e n t fo r the U.K. m arket. A lthough th e se s ta tis tic s show th a t a large p ro p o rtio n of the m ovem ent of sto ck re tu rn s is left u nexplained, the e x p la n a to ry va ria b le s used here are som ew hat more su c c e s s fu l than va ria b le s used in p re v ious s tu d ie s .12 12Using U.S. data for 1953-87, Fama estim ates a sim ilar equation that gives an unadjusted R-squared of 23 pe rcent for q u a rte rly stock Chart 3 Bond Market Excess Returns Quarterly Returns at an Annual Rate Percent Source: Japanese and U.K. data, Bank for International Settlements. Note: Excess returns are bond returns over a quarter minus a three-month interest rate at the start of the quarter. FRBNY Q uarterly Review/Autum n 1991 5 To evaluate the im p o rta n ce of future real a c tiv ity re la tive to oth e r fu n d a m e n ta ls, a second equation is e s ti m ated fo r each m arket, th is tim e including the change in the sh o rt-te rm real in te re st rate and fo u r leads of inflatio n. The sh o rt-te rm real rate is based on ex post inflation. The a d d ition a l va ria b le s produce a sig n ifica n t gain in e xp la na to ry pow er fo r the Japanese and U.K. m a rkets. But for U.S. sto ck returns, the in clu sion of in fla tio n reduces the a d ju ste d R -squared s ta tistic, in d i c a tin g o n ly a w e a k e ffe c t. S u b s e q u e n t e s tim a te s inclu de in flation leads for the Japanese and U.K. m ar kets but not for the U.S. m arket. The use of leads fo r real grow th and inflation allow s fo r th e e x tra in fo rm a tio n m a rk e t p a rtic ip a n ts m ay have.13 However, the e q u a tio n s in effect leave out fo re cast errors by p a rtic ip a n ts , an om itted va ria b le problem that w ould bias the c o e fficie n t e stim ates. A lthough this om issio n w ould not a ffe ct the evaluation of e xp la na to ry power, the a n a lysis of the in fo rm a tio na l c o n siste n cy of m arket reactions in the next section requires a p ro c e dure tha t gives unbiased co e ffic ie n t e s tim a te s .14 Footnote 12 (continued) returns, com pared with 27 percent here. The explanatory power is lower for m onthly returns and higher for annual returns. See Fama, "S tock Returns " 13See Fama, “ Stock Returns," and C am pbell and Ammer, “ What Moves the Stock and Bond M arkets?" 14Suppose we have the relationship r, = a + bE,(yl t t ), where the return r, depends on the current expectation of a future value y, + (. The realized value would be y , = E,(y,t ,) + e, ,,, where e ,+ , is the expectational error. Since expectations are unobservable, we m ight run the regression on the realizations, estim ating r, = a + b y ,t , + u,. Unfortunately, doing so would give a biased estim ate of Table 2 Explaining Stock M arket Excess Returns: Adjusted R-Squared Statistics Japan United States Equation with real growth ieadsf United Kingdom 0.42 (4.5**) 0.18 (0.8) 0.18 (0 5) 0.45 (1.3) A ddition of foreign excess returns 0.42 (12.2**) o b 0.24 (2.3**) A ddition of foreign real growth leads o 0.25 0.20 iv> 0.16 A ddition of inflation 0.18 (0.1) leads* ) 0.46 (1.8) Notes: Parentheses contain F-statistics indicating how each addi tion affects the explanatory power. Double asterisks indicate significant addition to explanatory power at the 5 percent level com pared with the previous equation. ^Equations include lagged dividend yield, domestic excess bond returns, and four leads of real growth. 6 FRBNY Q uarterly Review/Autumn 1991 Next let us te st the p o ssib le relevance of foreign fu n d a m e n ta ls to d o m e s tic sto ck returns. Foreign real a ctivity, fo r e xa m p le , m ay in flu e n c e d o m e s tic s to c k returns to the e xtent th a t som e of the firm s tra d e d in the m arket c o n d u ct an im p o rta n t p a rt of th e ir b u sin ess abroad. Table 2 re p o rts the a d ju ste d R -squared s ta tis tics for eq u a tio n s th a t ta ke a cco u n t of fo re ig n fu n d a m entals, here co n s is tin g of fo u r le a d s of real g ro w th in each of the tw o o th e r co u n trie s . In no in s ta n c e is the gain in e xp la n a to ry pow er s ta tis tic a lly s ig n ific a n t, a fin d ing that in d ica tes little d ire ct relevance of fo re ig n real a c tivity to d o m e stic m a rk e ts .15 T h is m eans th a t if fo r eign m arkets do convey in fo rm a tio n relevant to the fu n d a m e n ta l d e te rm in a n ts of m ovem ents in d o m e stic m arkets, the path of in flu e nce m ust link m o ve m e nts in foreign fu n d a m e n ta ls to d o m e stic fu n d a m e n ta ls , w ith m ovem ents in foreign m a rke ts se rv in g as the cha n n e l through w hich th is in fo rm a tio n is conveyed. A p re lim in a ry te s t fo r m a rke t o ve rre a ction is to d e te r mine w h e th e r returns in fo re ig n m a rke ts add s ig n ific a n t e x p la n a to ry p o w e r to s to c k re tu rn e q u a tio n s th a t alre a d y ta ke a c c o u n t of m a c ro e c o n o m ic fu n d a m e n ta ls .16 Table 2 re p o rts a d ju ste d R -sq u a re d s ta tis tic s for e q uations that in clu d e fo re ig n excess sto ck re tu rn s. The gain is q uite im pressive fo r the U.S. and Ja p a n e se stock m arkets, s u g g e s tin g th a t p a rtic ip a n ts in the se m arkets may often overreact to d e ve lo p m e n ts in m ar kets a b ro a d .17 E xp la in in g b o n d re tu rn s To id e n tify the s ig n ific a n t links am ong bond m arkets and fu n d a m e n ta ls , th e a n a ly s is now e v a lu a te s the degree to w hich m a cro e co no m ic v a ria b le s and fore ign bond m arket m ovem ents explain d o m e s tic bond m a rke t m ovem ents. Table 3 re p o rts a d ju ste d R -sq u a re d s ta tis tics as m easures of e x p la n a to ry pow er fo r re d u ce d -form bond re tu rn e q u a tio n s . T h e d e p e n d e n t v a ria b le is Footnote 14 (continued) b, because the residual term u, contains e(+J, w hich w ould be correlated with yl +1. 1®The finding is consistent with the con clu sion that do m estic factors dom inate international factors in explaining stock returns of individual firms, even of m ultinationals. See Bruno Solnik, International Investm ents, 2d ed. (Reading, M assachusetts: Addison-W esley, 1991), pp. 135-40. 16Robert P indyck and Julio R otem berg use this approach to co n clu d e that there is overreaction in m arkets for inte rnatio nally tra ded com m odities. See “ The Excess C om ovem ent of C om m odity P rices," Econom ic Journal, vol. 100 (1990), pp. 1173-89. 17As to the effects of the individ ual foreign stock m arkets, the Japanese m arket is significa nt for the U.S. m arket, as is the U.S. market for the Japanese m arket. In the case of bond m arkets, the Japanese market matters sig n ifica n tly for the U.S. m arket, the U.S. and U.K. m arkets for the Japanese m arket, and the Japanese market for the U.K. market. excess returns on lon g-term g o vernm ent bonds, th a t is, the returns m inus a th re e -m o n th rate. The com m on e x p la n a to ry v a ria b le s are e xce ss sto c k re tu rn s, the change in the real s h o rt-te rm rate, and fo u r q u a rte rly leads of inflation. As w ith stocks, the e xp la na to ry power of the se va ria b le s is good by the standards of the lite rature . The a d d ition of fo u r leads of real grow th does not help sig n ific a n tly for any of the bond m arkets. S ub se q u e n t e stim a te s of bond return e q uations leave out real grow th leads. The a n a lysis confirm s th a t future in fla tion is a m ore im p o rta n t fu n d a m e n ta l for bond m ar kets than is future real activity, w h ile future real a c tiv ity is m ore im p o rta n t fo r stock m arkets. W hen foreign fu n d a m e n ta ls in the form of fo u r in fla tio n le a d s in each of the tw o o th e r c o u n trie s are in clu d e d , the gain in e xp la n a to ry power is s ta tis tic a lly in s ig n ific a n t in every case. A gain foreign fu n d a m e n ta ls seem to have little d ire ct relevance to d o m e stic m ar kets. T his finding m eans th a t if foreign m arkets convey in fo rm a tio n a b o u t fu n d a m e n ta ls , th e re le v a n c e to d o m e stic m arkets is like ly to arise from links betw een foreign fu n d a m e n ta ls and d o m e stic fu n dam entals. Table 3 also reports the a d ju ste d R -squared s ta tistics fo r e q u a tio n s in co rp o ra tin g foreign excess bond returns. T he su b sta n tia l gain fo r all three bond m arkets s u g g e sts th a t p a rtic ip a n ts in th e se m arkets could be over reacting to m arket m ovem ents abroad. In te rp re ta tio n The strong e xp la n a to ry pow er of foreign m arket returns by its e lf does not m ean th a t m arket overreaction exists. It is easy to im agine the m arkets m oving to g e th e r in Table 3 Explaining Bond M arket Excess Returns: Adjusted R-Squared Statistics United States Japan United Kingdom Equation with inflation leads* 0.26 0.14 0.24 Addition of real growth leads* 0.23 (0.6) 0.11 (0.6) 0.23 (0.7) Addition of foreign inflation leads 0.28 (1.2) 0.12 (0.9) 0.19 (0.5) Addition of foreign excess returns 0.37 (4.6**) 0.33 (9.0**) 0.25 (3.0**) Notes: Parentheses contain F-statistics indicating how each a d di tion affects the explanatory power. Double asterisks indicate significant addition to explanatory power at the 5 percent level compared with the previous equation. ^Equations include change in real short-term rate, domestic excess stock returns, and four leads of inflation. response to com m on or c o rre la te d in fo rm a tio n a bout fu n d a m e n ta ls not ca ptured by o u r m a cro e co n o m ic v a ri ables. Such in fo rm a tio n m ay so m e tim e s cause the d if fe re n t m a rk e ts to m ake s im ila r m is ta k e s a b o u t th e future, so th a t the e x p la n a to ry pow er of fo re ig n returns may a rise sim p ly from the co rre la tio n of fo re ca st errors. In particular, foreign sto c k and bond m a rke ts m ay throw out s ig n a ls a b o u t fo re ig n m a c ro e c o n o m ic fu n d a m e n ta ls th a t in turn help p re d ict d o m e s tic fu n d a m e n ta ls . The s ig n a ls may at tim e s turn o u t to be fa lse , so th a t the im plied future d e ve lop m e n ts do not show up in the data, b u t th e d o m e s tic m a rk e ts w ill have b e e n rig h t to respond to the sig n a ls. The a n a ly s is in th e next se ctio n a s k s w h e th e r a c tu a l m a rk e t b e h a v io r c a n be so ju s tifie d . Testing for inform ational consistency A n a ly tic a l a p p ro a c h If m arkets are in fact reacting to in fo rm a tio n related to fu n d a m e n ta ls and if, as the d ata in d ica te , fo re ig n fu n d a m entals have little d ire ct relevance to d o m e s tic m arkets, then the m arkets m ust be reacting to one a n o th e r’s m ovem ents la rg e ly b ecause of re co g n ize d lin ks betw een d o m e stic and fo re ig n fu n d a m e n ta ls . T he d o m e s tic m ar ket w ill be try in g to in fe r in fo rm a tio n a b o u t foreign fu n d a m e n ta ls from the o th e r m a rke ts in o rd e r to m ake b e tte r fo re ca sts of d o m e s tic fu n d a m e n ta ls . If the actual reaction is c o n s is te n t w ith the links betw een fu n d a m e n ta ls and w ith the links betw een fu n d a m e n ta ls and m ar ket returns, then in fo rm a tio n a l c o n s is te n c y h o ld s .18 The e m p irica l a n a ly s is below p ro ce e d s by e s tim a tin g the various s p e cifie d links am ong m a rke ts and fu n d a m entals to te st fo r in fo rm a tio n a l c o n s is te n c y .19 The d is c ussion fo cu se s on real a c tiv ity as the fu n d a m e n ta l m ost im p o rta n t to sto c k m a rke ts and on in fla tio n as the fu n d a m e n ta l m ost im p o rta n t to bond m a rke ts. Three q u e stio n s are a d d re sse d : How c lo s e ly w ould sto ck m ar kets move to g e th e r if the d o m e s tic m a rke ts were relying on foreign sto ck m a rke ts s im p ly fo r in fo rm a tio n a b o ut foreign real a c tiv ity ? How c lo s e ly w ould bond m a rke ts move to g e th e r if the d o m e s tic m a rke ts were relying on foreign bond m a rke ts sim p ly fo r in fo rm a tio n a b o u t fo r eign in fla tio n? How c lo s e ly do th e m a rke ts a ctu a lly move to g e th e r? S to ck re tu rn s a n d re a l g ro w th To o btain unbiased e s tim a te s of the e ffe c t of real grow th on stock returns, we can re e stim a te the sto ck return e q u a tio n s by replacing the fo u r leads of real grow th and 18The appendix offers a form al m odel for this analysis. 1®Unlike a test based on statistica l exp la nato ry power, this test does not require com plete data on the markets' inform ation, only enough data to produce good coe fficients on the various links. FRBNY Q uarterly Review/Autum n 1991 7 in fla tio n w ith th e ir p redicted va lu e s.20 The d ivid e n d yield and excess bond returns are kept in the eq u a tio n s to control fo r oth e r fu n d a m e n ta ls, p a rtic u la rly the effects of chan ges in the real d is c o u n t rate. Inflation is om itted from the U.S. e stim a te s because it lacks a d d ition a l e xp la n a to ry power. The e stim a te s in Table 4 su g g e st th a t the Japanese stock m arket is the m ost sensitive to predicted real grow th and the U.K. m arket the least sensitive. The sum of the c o e fficie n ts on real grow th indicates that w hen exp ecte d d o m e stic real grow th over the next fo u r q u a r te rs increases by a point, U.S. excess sto ck returns rise 2.5 p e rce n ta g e points on average, Japanese excess returns rise 4.3 points, and U.K. excess returns rise 0.6 of a point. In te rn a tio n a l re a l g ro w th lin ks If real grow th abroad has no d ire ct effect on d o m e stic sto ck returns, a reaction to foreign m arkets should in d i cate a link betw een grow th abroad and grow th at home. To e stim a te th is link, an index of real grow th leads is co n stru cte d fo r each m arket using w eights p ro p o rtio n a l to the real grow th c o e ffic ie n ts in the e stim ated stock 20The instrum ents used to predict real growth and inflation are the divide nd yield, excess bond returns, contem poraneous real growth and inflation, and four lags each of dom estic inflation, real growth in each of the three countries, the change in log dollar oil prices, and the changes in the two relevant log exchange rates. return e q u a tio n s .21 The m ovem ents in th is index for other c o u n trie s are p re su m a bly w hat d o m e s tic m arket p a rtic ip a n ts can in fe r from m ovem ents in fo re ig n sto ck prices. The upper se ctio n of Table 5 re p o rts the e stim a te d effe cts of foreign fu tu re real grow th on d o m e s tic fu tu re real g ro w th .22 The e stim a te s are based on the c o n stru cte d indexes and co n tro l fo r c u rre n tly o b s e rva b le v a ria b le s th a t m ay a lso h e lp p re d ic t d o m e s tic real g ro w th . To a llo w fo r the jo in t d e te rm in a tio n of real grow th in d iffe re n t co u n trie s , in s tru m e n ts are used fo r the foreign real grow th in d e xe s.23 The e s tim a te s in d i cate th a t w hen U.S. real grow th over the next fo u r q u a rte rs rises a p e rce n ta g e point, J a p a n e se real grow th can be expected to rise 0.15 of a p o in t and U.K. real grow th 0.60 of a point. W hen fu tu re J a p a n e se real grow th rises a point, U.S. real grow th can be expected to rise 0.60 of a p oint and U.K. real grow th to fa ll 0.3 6 of 21For exam ple, the index for U.S. real growth w ould have a 60 percent w eight for the first lead because the co e fficie n t on this lead is 60 percent of the sum of the four lead coe fficients (Table 4). 22The control variables are the current values and four lags each of dom estic real growth and inflation. 23The instrum ents are four lags of real growth and inflation in each of the three countries. Table 4 Stock Return Equations Estimated by Two-Stage Least Squares (D ependent Variable Is Excess Stock Return) United States C onstant D ividend yield E xcess bo nd return P redicted real growth First lead Second lead Third lead Fourth lead Sum of c oe fficients U nited K ingdom - 1 9 .5 0 2.88 0.38* ( - 1 .2 9 ) (1 31) (2.71) - 2 0 .2 5 2.73 0.15 ( -0 .5 8 ) (1.09) (1.21) 17.66 0.87 0.49* 1.51 3.76 - 3 .3 9 * 0.64 (0.74) (1-55) (2 69) (0.32) 3.32 1.54 - 0 .9 4 0.34 (1.45) (0.74) ( - 0 .4 5 ) (0.17) 0.79 0.51 -0 .9 1 0.24 4.27 2.52 P redicted inflation First lead Second lead Third lead Fourth lead R-squared A djuste d R-squared Japan - 1 .1 2 0.14 0.05 ( 1.83 -3 .4 1 * 1.81 0.28 0.15 (0.51) (1.21) (3.02) (0.79) (0.51) (-0 .9 8 ) (0 24) 0.63 - 0 .8 7 ) (1.31) ( -2 .4 4 ) (1.14) 2.10 2.76* - 4 .7 2 * -1 .8 9 (1.47) (1.70) ( - 3 .0 4 ) ( - 1 .3 2 ) 0.48 0.38 Notes: Instrum ents are d ivide nd yield, excess bond returns, contem poraneous real growth and inflation, and four lags each of dom estic inflation, real growth in each of the three countries, percentage change in d o lla r oil prices, and pe rce n ta g e cha nge in the two d o lla r exchange rates. T-statistics are in parentheses. Asterisks indica te significa nce at 10 pe rcent level. 8 FRBNY Q uarterly Review/Autumn 1991 a p o in t.24 U.K. real grow th seem s to have little e ffe ct on real grow th in the other econom ies. The large stan dard erro rs indicate th a t these e s ti m ates are not very precise. The usual sig n ifica n ce tests w ould su g g e st that there are no real grow th links. N o n e th e less, this a n a lysis m ust give the m arkets the benefit of the dou b t by a llow in g them a reason to react to one a n o th e r’s m ovem ents. M arket p a rtic ip a n ts presum ably do not lim it th e ir responses to only those influences that su rvive strin g e n t s ta tis tic a l te sts. Hence the a n a lysis w ill proceed on the a ssu m p tio n th a t the e stim a te s of real grow th links in Table 5 are our best e stim ates. Im p lie d s to c k m a rk e t re a c tio n s The m iddle section of Table 5 show s the m agnitudes of sto ck m arket in te ra ctio n s im plied by the links betw een m a rke ts and real a c tiv ity and the links betw een d o m e s tic and foreign real activity. The stro n g e st im plied reac tio n s are betw een the U.S. and Japanese m arkets. The various links am ong m a rke ts and fu n d a m e n ta ls im ply th a t if U.S. m arket p a rtic ip a n ts saw Ja panese stock re tu rn s rise 1 p e rc e n ta g e p o in t w hile o th e r fa c to rs re m a in e d u n c h a n g e d , th e y w o u ld in fe r a ris e in expected Jap anese real grow th of 0.23 of a point (1 d ivid e d by 4.3) and thus a rise in expected U.S. real grow th of 0.14 of a point (0.23 tim e s 0.60), so th a t U.S. 24Because real growth is correlated across countries, the sum of the coe fficients provides better estim ates than do the individual coefficients. In the case of U.K. growth, for example, the sum of the estim ated effects of U.S. and Japanese growth of 0.24 is more reliable than the individ ual effects of 0.60 and -0.36, respectively. stock returns w ould rise 0 .3 5 of a p o in t (0.14 tim e s 2.5). S im ila rly, if J a p a n e s e m a rk e t p a rtic ip a n ts saw U.S. stock returns rise 1 point, th e y w ould react so that Japanese sto ck returns rise 0 .25 of a p oint. The various links do not im ply strong p o sitive re a ctio n s in the case of the U.K. m arket. A c tu a l s to c k m a rk e t re a c tio n s T h e te s t o f in fo r m a tio n a l c o n s is te n c y u s e d h e re involves e xtra ctin g th a t p a rt of fo re ig n sto ck m arket returns reflecting m ovem ents in fo re ig n exp e cte d real activity. To th is end, the e stim a te d fo re ig n sto ck return e q u a tio n s are cle a n se d of the e ffe cts of o th e r fu n d a m e n ta ls to cre a te p re d ic to rs of fo re ig n real a c tiv ity g ro w th .25 The p re d icto rs of fo re ig n real a c tiv ity grow th are s u b stitu te d into the real grow th link e q u a tio n s, and the resulting p re d ictio n s of d o m e s tic real grow th are in turn s u b s titu te d into the d o m e s tic sto ck return e q u a tions. In p rin cip le , th e se e q u a tio n s co n tro l fo r the m ove m e n ts o f fu n d a m e n ta ls o th e r th a n re a l a c tiv ity , p a rtic u la rly those reflected in excess bond returns. The estim ated c o e ffic ie n ts on the fo re ig n real grow th p re d ic to rs in th e m o d ifie d s to c k re tu rn e q u a tio n s p ro v id e m easures of sto ck m a rke t re a ctio n s to the relevant foreign m arket m o ve m e n ts.26 25From the excess stock return equations in Table 4, the terms involving the divide nd yield, excess bond returns, and inflation are subtracted, so that only the real growth term s are left 26The regression is based on equation A .5 of the model developed in the appendix Table 5 Real Growth Links and Implied and Actual Stock Market Reactions United States Japan U nited Kingdom Effect on dom estic real growth of real growth in: United States Japan United Kingdom 0.60 0.03 (0.50) (0.08) 0.35 0.12 (0.30) (0.32) 0.62* 0.04 (0.13) (0.09) 0.15 (0.13) - 0 .0 8 (0 05) 0.60 - 0 .3 6 (0.78) (1.72) 0.15 - 0 .0 5 (0.20) (0.25) 0.14 0 29 (0.17) (0.21) Im plied dom estic stock m arket reaction to stock m arket in: U nited States Japan United Kingdom 0.25 (0.22) -0 .5 4 (0.34) Actual dom estic stock market reaction to stock market in: U nited States Japan United K ingdom 0.38* (0.08) 0.08* (0.07) Notes: S tandard errors are in parentheses. Asterisks indica te significa ntly greater actual over im p lie d reaction at the 10 pe rcent level. FRBNY Q uarterly R eview/Autum n 1991 9 The bottom sectio n of Table 5 reports the e stim a te s of actual sto ck m arket reactions. The estim ates show s ig n ifica n t o ve rrea ction by the U.S. and Ja panese stock m a rkets to each other. In response to a 1 point rise in Ja p a n e se sto ck returns, U.S. sto ck returns rise 0.62 of a p oint on average, an increase nearly tw ice the m a g n i tude ju s tifie d by the in fo rm a tio n about real a c tiv ity co n veyed by the Jap anese sto ck m arket. S im ilarly, the size of the Japa nese sto ck m arket reaction to U.S. stock m arket m ovem ents is ha lf again as great as the size im plied by in fo rm a tio n a l consistency. A lthough the J a p anese sto ck m arket is show n to have a s ta tis tic a lly s ig n ific a n t overreaction to the U.K. m arket, th is result is b a s e d on an im p la u s ib ly la rg e n e g a tiv e im p lie d reaction. B o n d re tu rn s a n d in fla tio n To o b ta in unbiased e stim a te s of the e ffe ct of inflation, the bond return eq ua tio n s are reestim ated by replacing the inflation leads w ith th e ir predicted va lu e s.27 Excess sto ck returns are kept as a se parate variable in the eq u a tio n s to co ntrol fo r o th e r fu n d a m e n ta ls, p a rtic u la rly for the po ssib le e ffe cts of future real grow th on the d isco u n t rate. The estim a te s reported in Table 6 show that the U.S. and Ja p a n e se bond m arkets are very s e n sitive to inflatio n, w hile the U.K. m arket in e xp lica b ly re s p o n d s p o s itiv e ly to in fla tio n . A 1 p o in t ris e in expected inflation over the next fo u r q u a rte rs reduces U .S . b o nd re tu rn s 3 .8 p o in ts and J a p a n e s e bond returns ne arly 3.0 points. 27The instrum ents used to predict inflation are excess stock returns, current real growth and inflation, and four lags each of dom estic real growth, inflation in each of the countries, oil price inflation, and currency deprecia tion rates. In te rn a tio n a l in fla tio n lin ks To estim a te the in fla tio n links, an index of in fla tio n leads is co n stru cte d fo r each country. The w e ig h ts are d erived from the bond return e q u a tio n s in the sam e way th a t they were draw n from the sto ck return e q u a tio n s fo r the real grow th indexes. The m ovem ents in th is in fla tio n index are w hat can be inferred from bond m a rke t m ove m ents. Based on th e se indexes, the e s tim a te s in the upper se ctio n of Table 7 m easure the e ffe c ts of foreig n on d o m e stic fu tu re in fla tio n. T he e s tim a te s co n tro l for o th e r fa cto rs and fo r the jo in t d e te rm in a tio n of inflation in the d iffe re n t c o u n trie s .28 Here a 1 p o in t rise in the J a p a n e se in fla tio n rate o ve r th e next fo u r q u a rte rs raises the expected U.S. rate 0 .29 of a p oint, and a 1 point rise in the U.S. rate ra ise s the J a p a n e se rate 0.43 of a point. A 1 p oint rise in both the U.S. and Ja p a n e se rates raises the U.K. rate 1.46 p o in ts. T h e se e stim a ted in te rn a tio n a l in fla tio n links tend to be s ta tis tic a lly s ig n ifi cant and thus more re lia b le than the e stim a te d real grow th links. Im p lie d b o n d m a rk e t re a c tio n s The various lin ks am ong bond m a rke ts and expe cted inflation rates im p ly the m a rke t re a ctio n s c a lc u la te d in the m iddle p a rt of Table 7. H ence, if U.S. bond m arke t p a rtic ip a n ts saw J a p a n e se bond returns rise a p e rc e n t age point, they w ould in fe r a fa ll in the e xp e cte d J a p anese in flation rate of 0 .34 of a p o in t (1 d ivid e d by 2.98) and a fall in the expected U.S. rate of n early 0.10 of a p oint (0.34 tim e s 0.29), so th a t U.S. bond 28The control variables are current and four lags each of dom estic real growth and inflation in the three countries. To allow for the joint determ ination of inflation, four lags each of real growth and inflation in the three countries are used as instrum ents. Table 6 Bond Return Equations Estimated by Two-Stage Least Squares (D ependent Variable Is Excess Bond Return) United States 23.39 0.17 (2 45) (1.45) Japan 12.86 0.16 U nited K ingdom (1.47) (0.92) -1 8 .7 9 0.47* ( - 1 .4 8 ) (4.02) (0.54) (0 20) (0.21) (--2 .3 6 ) 1.79* - 3 .7 5 * 3.38* -0 .4 1 (1.74) ( - 3 .6 4 ) (2.81) ( - 0 .4 3 ) C onstant Excess stock return Predicted inflation First lead S econd lead Third lead Fourth lead -4 .4 9 * 3.19 - 2 .0 8 - 0 .4 4 Sum of c oe fficients - 3 .8 3 - 2 .9 8 1.01 0.32 0.26 0.18 0.11 0.21 0.14 R-squared A djuste d R-squared ( - 2 .3 2 ) d 62) ( - 1 00) (-0 2 1 ) 0.84 0.32 0.40 - 4 .5 4 * Notes: Instrum ents are excess stock returns, contem poraneous real growth and inflation, and four lags each of d o m estic real growth, inflation in each of the three countries, percentage change in dollar oil prices, and pe rcentage cha nge in the two do lla r exch ange rates. T-statistics are in parentheses. A sterisks indica te significance at the 10 percent level. 10 FRBNY Q uarterly Review/Autumn 1991 returns w ould rise 0.37 of a p oint (about 0.10 tim e s 3.8). T he im plied reaction of the Japanese bond m arket to the U.S. m arket of 0 .33 is a lm o st as strong. The im plied re a ctions involvin g the U.K. m arket are much weaker, if not negative. A c tu a l b o n d m a rk e t re a ctio n s The bottom section of Table 7 reports e stim a te s of the relevant actual bond m arket reactions. To extract the p a rt of foreign bond m arket returns that reflects m ove m ents in foreign expected inflation, the e stim ated bond m arket eq u a tio n s are used to c o n stru ct p redictors of foreign in fla tio n .29 These pre d icto rs are then su b stitu te d into the inflation link e qua tio n s, w hich in turn are s u b stitu te d into the d o m e stic bond return equations. The d o m e stic bond return eq u a tio n s are then reestim ated, w ith the foreign bond m arket m ovem ents in e ffe ct h e lp ing predict d o m e stic infla tio n. The equations co n tro l for the m ovem ents of fu n d a m e n ta ls other than inflation, p a rtic u la rly those reflected in excess stock returns. The e stim a ted co e fficie n ts on the foreign bond m arket va ri ables then m easure the actual bond m arket reactions to the relevant m ovem ents in the foreign m arkets. T he e stim a te s show a sig n ific a n t overreaction by the J a p anese bond m arket to m ovem ents in the U.S. m ar ket. Japan ese returns rise 0.54 of a point instead of 0.33 of a p oint in response to a 1 point rise in U.S. “ S pecifically, we subtract from the excess bond return equations in Table 6 all the term s but those for inflation returns. The e stim a te s also show s ig n ific a n t o ve rre a c tio n s by the U.K. and Ja p a n e se bond m a rke ts to each o th e r’s m ovem ents. The a p p a re n t o v e rre a ctio n s in v o lv ing the U.K. m arket sh ould be tre a te d w ith m ore s k e p ticism , however, because they re fle ct an in e x p lic a b ly positive e ffe c t of U.K. in fla tio n on U .K. bond returns. Conclusion The beh a vio r of sto ck and bond m a rke ts is of co n ce rn to e co n o m ists because th e se m a rke ts set p rice s a ffe c t ing the co st of ca p ita l fo r the c o rp o ra te s e c to r and because excess m a rke t v o la tility m ay lead to fin a n c ial s tra in s and m a cro e co no m ic in sta b ility. T h o se w ho w o rry a bout excess v o la tility have reco m m e n d e d such p o lic ie s as ta xin g se c u ritie s tra n s a c tio n s , ta x in g s h o rt-te rm c a p ital ga in s more than lo n g -te rm c a p ita l g a in s, and ra isin g m argin re quirem ents on sto ck p u rc h a s e s .30 T h is stu dy asks w h e th e r excess c o rre la tio n s a cro ss m a rke ts are a like ly source of excess vo latility. The evid e nce presented s u g g e sts som e te n d e n c y by p a rtic ip a n ts in the U.S. and J a p a n e se sto ck m arkets and in the U .S., Ja p a n e se , and U.K. bond m a rke ts to overreact to one a n o th e r’s m a rke t m ovem ents. A lth o u g h the e stim a te s are im p re cise , th e y in d ic a te th a t the tw o “ Lawrence Summers and V ictoria Summ ers s u p p o rt the securities transactions tax; see "When Financial M arkets Work Too Well: A Cautious Case for a S ecurities Transactions Tax,” Jou rnal of Financial Services Research, vol. 3 (1989), pp. 261-86. Gikas Hardouvelis advocates raising m argin requirem ents; see "M argin Requirements and Stock Market V olatility," this Q uarterly Review, Summer 1988, pp. 80-89 Table 7 Inflation Links and Implied and Actual Bond Market Reactions United States Japan U nited K ingdom Effect on dom estic inflation of inflation in: U nited States Japan United K ingdom (0.18) 0,43* 0,29* - 0 .0 6 (0.12) (0.02) 0.37 0.23 (0.15) (0.08) 0.42 - 0 .0 7 (0.10) (0,10) 0.00 (0.03) 0.33 (0.14) -0 .0 1 (0.09) - 0 .5 4 2.00* (1 04) (0.96) 0.14 -0 .6 8 (0 27) (0.33) - 0 .0 8 0.24* (0.14) (0.11) Im plied dom estic bond market reaction to bond market in: U nited States Japan United Kingdom A ctual dom estic bond m arket reaction to bond m arket in: U nited States Japan U nited Kingdom 0.54* (0.13) 0.22* (0.11) Notes: S tandard errors are given in parentheses. Asterisks on inflation links indica te significa ntly positive coe fficients, and asterisks on m arket reactions indica te s ignifica ntly greater actual over im plied reaction at the 10 pe rcent level, FRBNY Q uarterly Review/Autum n 1991 11 stock markets move together more closely than would be expected from the information the markets convey about future real activity and from the links between domestic and foreign real activity. The evidence on bond markets is less consistent, but the three markets seem to move together more closely than would be expected from the information they provide on inflation and from the links between domestic and foreign inflation. An important limitation of the present study is that it analyzes market reactions on the basis of average behavior over the period. In fact, the stock markets sometimes move very closely together, while at other times they move independently. When the Japanese stock market plunged in the spring of 1990, the U.S. and U.K. stock markets shrugged off the event; by contrast, in October 1987 the three markets fell as one. It is as if the markets have “moods,” so that a shock is sometimes transmitted to other markets more forcefully than at other times. If overreaction helps drive market prices away from fundamental values with some frequency, the resulting market volatility may pose needless risks to investors 12 FRBNY Quarterly Review/Autumn 1991 and raise the cost of financing in the form of publicly traded debt or equity.31 International comparisons of the cost of capital suggest that U.S. corporations are placed at a competitive disadvantage by the relatively high costs of equity financing in the United States, costs that some observers attribute in part to stock market volatility.32 Worse, the high volatility may make markets vulnerable to a global crash. In a world where market prices can occasionally take on a life of their own, the various markets may at times inflate together and then burst like enormous bubbles. 31The underlying behavior m ight be c h a ra c te riz e d as a form of international noise trad ing . B radford D e Long, A ndrei Shleifer, Lawrence Sum m ers, and R obert W aldm ann show that in g e n e ra l the presence of noise trad ers can m ake the m arkets too risky for investors who rely on fun dam entals, so that prices m ay deviate from fun dam entals for e x te n d e d periods of tim e. See "N oise Trader Risk in Financial M a rk e ts ,” J o u rn a l o f P o litic a l E c o n o m y , vol. 98 (1990), pp. 7 0 3 -3 8 . “ R obert N. M cC auley and Steven Zim m er, “ Explaining International D ifferences in the Cost of C a p ita l,” this Q u a rte rly R e vie w , Sum m er 1989, pp. 7-28 . Appendix: A Model of Domestic Stock Markets’ Reactions to information from Foreign Markets This model formalizes a possible role for foreign stock markets as conveyors of information about real activity relevant to domestic stock markets. Real activity can be thought of as determining the cash flows of publicly traded firms. With stock prices assumed to be the pres ent values of the future streams of cash flows, we can write the stock market return as a function of future real activity growth and of variables tracking the discount rate: (A.1) r\ N1 = Y4 + k=1 ^ 'k Y 't + k + V 't< where z't is the vector of discount rate variables, y ‘t+k is the kth lead of real activity growth, and v'\ is noise in returns. The number of leads is N‘. Real activity in one country could affect real activity in other countries through international trade. To construct an index of real activity that reflects stock market behav ior, we can derive the weights from the lead structure implicit in the returns equation. The index collapses real growth in several periods into a single variable: (A.2) # = XT' w* > & ky't+k, param eters in equation reduces to ( A .3 ) r\ = y Z ' + 8fyi + A .1 . The returns equation v\, and the co-variation of returns and real activity is mea sured by a single parameter, 81. (A.4) y ‘t = a‘(L)Xt + . &Jy { + <4. where y ‘t is our index of future real activity growth in country / as of time t, y{ our index of future real activity growth in country j, ot!(L)x‘t a vector polynomial in the lag operator L, X, a vector of observable variables helping to predict y ‘t, and ift the unpredictable part of real activity growth. The 0,y coefficients measure the co-movement with real activity growth in other countries after we have controlled for other macroeconomic variables. The hypothesis of rational expectations allows us to assume that stock market investors in country / know equations A.2, A.3, and A.4. At time t, they observe the returns r{ and other fundamentals t t in the other coun tries as well as x\ and 4 in their own country. The hypothesis gives (A .5) r 1, = Y 4 + h'a'(L)x‘, + S' in which investors infer y{ from — (rj 5* hi come from the where 0L = 81/8* and 8j We now measure the links in real activity across coun tries by estimating Pf(r{ - 7 '>*< 8i + v'„ iA )- Once 8' and S' are estimated from equation A.1 and p* is estimated from equation A.4, a regression of stock mar ket returns in country / on stock market returns in other countries, as in equation A.5, should yield a coefficient not significantly different from 8,3'//8/ for country /. Other wise, we can conclude that international stock market correlations fail to reflect macroeconomic fundamentals. FRBNY Q uarterly R eview/Autum n 1991 The Cost of Capital for Securities Firms in the United States and Japan by Robert N. McCauley and Steven A. Zimmer Recent studies of international differences in capital costs have focused on industry and banks. In the 1980s U.S. firms seemed to be losing ground internationally, whether measured by semiconductor trade, industrial investment, manufacturing trade, or market share in U.S. commercial lending. This slipping competitiveness prompted economists to investigate whether U.S. industry and banks were laboring under a cost of capital disadvantage. By contrast, the cost of capital for U.S. securities firms received little attention during the last several years because these firms appeared to perform more creditably. They defended their home turf, mounted major expansions into foreign markets, and staked out market share and profit in trading government bonds and equities abroad.1 U.S. securities firms invested much more abroad that their foreign competitors invested in the United States. In other industries, espe cially banking, foreign direct investment into the United States dominated U.S. investment abroad (Table 1).2 ’ The firms showed best results in trad ing Jap anese and G erm an governm ent bonds in Tokyo and London, respectively, and in trad ing J a p a n e se equities and equity derivatives in Tokyo. See John J. R uocco, M aureen LeB lanc, and Patrick D ignan, "C om petitiveness in G overnm ent Bond M a rk e ts ,” and M artin Mair, M ichael Kaufman, and Steven Saeger, “C om petitiveness in Equity M a rk e ts ,” in In te rn a tio n a l C o m p e titiv e n e s s o f U.S. F in a n c ia l F irm s : A S ta ff S tu d y (N ew York: Federal R eserve Bank of New York, 1991), pp. 130-72. 2Perhaps as a result, the pu blic policy discussion of the securities industry has focused on ensuring that U .S. firms enjoy equal a c c e s s to foreign financial m arkets. S ee, for instance, D ep artm ent of the Treasury, N a tio n a l T re a tm e n t S tudy, 1990 U p d a te , pp. 225-41; Staff of the Board of G overnors of the Federal R eserve System and the Federal Reserve Bank of New York, R e p o rt o n Im p le m e n ta tio n o f th e P rim a ry D e a le rs A ct, m em orandum , August 15, 1989; and 14 FRBNY Quarterly Review/Autumn 1991 Reversing the procedure of earlier studies, this article takes the respectable performance of U.S. securities firms as its rationale for exploring cost of capital differ ences between countries. If U.S. firms achieved some degree of success in spite of higher capital costs, then this disadvantage is clearly not a decisive one. But if the disadvantage faced by securities firms is smaller than that faced by U.S. industry and banks, then capital costs may help to explain differences in competitive outcomes. Our investigation begins with a comparison of the capital costs faced by U.S. and Japanese securities firms in the 1982-91 period. We measure the cost of capital to five U.S. and four Japanese securities firms as the multiple that their respective stock exchanges assigned to the economic earnings of the firms. Our findings indicate that U.S. equity investors placed a lower value on a given stream of earnings of U.S. securities firms than Japanese equity investors placed on a comparable stream of earnings of Japanese secu rities firms. As a result, U.S. securities firms needed to earn more on a given sum of capital underpinning any line of business. The gap in valuation of securities firms’ earnings in the New York and Tokyo stock exchanges nevertheless appears to be narrower than the gaps we found between U.S. and Japanese industries and banks in our own earlier studies of cost of capital differences.3 If F o o tn o te 2 (c o n tin u e d ) "Japan, U .K . and Sw itzerland: Prim ary D ea le rs A ct U p d a te ," m em orandum , D e c e m b e r 3, 1990. 3R obert N. M cC auley and Steven A. Zim m er, "Explaining International D ifferences in the C ost of C a p ita l," this Q u a rte rly U.S. se cu ritie s firm s c a rry a s m a lle r d is a b ility in ca p ita l co sts than o th e r U.S. firm s, then it m akes sense that any adva ntage s in oth e r d im e n sio n s of co m p e titio n , such as e xpe rien ce w ith derivative products or a p p lic a tion of technology, could be d e cisive in overall c o m p e ti tive outcom es. In see king to explain ca p ita l cost diffe re n ce s, we e m p ha size ge neral fa cto rs a cco u n tin g fo r a lower J a p anese cost of e q u ity in the la tte r 1980s. These include h ig h e r h o u s e h o ld s a v in g s and s m o o th e r e c o n o m ic grow th. O u r a n a ly s is also c la rifie s w h y the gap betw een m easured e q u ity costs in New York and Tokyo m ight be s m a lle r fo r se cu ritie s firm s than fo r banking and other in d u strie s . On the one hand, Japanese se c u ritie s firm s’ co st of e q u ity may be h ig h er than that of Ja panese n o n fin a n cia l firm s or banks because the m arket per ceives a re la tively severe th re a t to the se c u ritie s firm s’ revenues and e arnings in the ongoing trend toward fin a n cial d eregulation . On the o th e r hand, the lower e q u ity costs for U.S. s e cu ritie s firm s relative to other U .S. com p a n ie s may be influenced by the choice of s a m p le p eriod. The m id-1980s were boom years for the s e c u ritie s business, and U.S. investors, seized w ith the g row th p o s s ib ilitie s created by the fin a n cial innovators and e nginee rs of Wall S treet in in cre a sin g ly g lobal m ar kets, m ay have priced U.S. s e c u ritie s firm s’ e a rnings at a prem ium . Footnote 3 (continued) Review, vol. 14 (Summer 1989), pp. 7-28; and Steven A. Zimmer and Robert N. McCauley, "Bank Cost of C apital and International C o m p e titio n ,” this Q uarterly Review, vol. 15 (Winter 1991), pp. 33-59. Measuring the cost of capital S e cu ritie s firm s p rovide p ro d u cts and e n gage in a c tiv i ties of varying risk a g a in s t w hich th e y m ust hold e q u ity ca p ita l. The required return on th is e q u ity c a p ita l w ill be im p o rta n t in d e te rm in in g the c o m m issio n or fee th a t a firm m ust ch a rg e fo r a s e rv ic e or the return it m ust earn a rb itra g in g m a rke ts or in ve stin g on its own a cco u n t. We define the co st of ca p ita l fo r a s e c u ritie s firm as the m inim um required fee the firm m ust ch a rg e , or the return it m ust m ake, to cover the required return on the e q u ity ca p ita l a llo tte d to an activity. O ur d e fin itio n of cost of ca p ita l fo r a s e c u ritie s firm , like our d e fin itio n of the co st of ca p ita l fo r a bank, excludes d ebt co sts. T he reason fo r th is e xclu sio n is that in te rn a tio n a lly active s e c u ritie s firm s face s im ila r borrow ing costs. For in sta n ce , J a p a n e se firm s’ s u b s id i aries in New York sh o u ld be able to fin a n ce th e m se lve s at m uch the sam e rates as U.S. firm s. Indeed, th is arg u m e nt may be more firm ly g ro u n d e d fo r s e c u ritie s firm s than fo r banks. The m ost im p o rta n t so u rce of borrow ed fu n d s fo r a large s e c u ritie s firm is the sale and fo rw a rd re p u rc h a s e of s e c u ritie s . T he se cu re d nature of th is fin a n c in g te c h n iq u e le s s e n s c re d ito r dem ands fo r su b s ta n tia l d iffe re n c e s in in te re s t rates b a s e d on th e c r e d itw o r t h in e s s o f th e b o rro w e r. R e p u rc h a s e a g re e m e n ts have g e n e ra lly p e rm itte d se c u ritie s firm s in the U nited S tates to fin a n ce th e m selves at rates below in te rb a n k ra te s .4 O ur d e fin itio n of co st of c a p ita l fo r s e c u ritie s firm s fo llo w s the d e fin itio n of bank c o st of c a p ita l p re se n te d in our e a rlie r stu d ie s, and we w ill p roceed in a s im ila r fashion. The first step in asse ssin g c o st of c a p ita l differ4For the last year, the overnight repurchase rate has on average exceeded the federal funds rate in the U S money market. Table 1 Foreign Direct Investment Flows into and out of the United States, 1985-89 (In B illions of Dollars) Total M anufacturing Inflow Outflow 232.9 90.5 2.6 110.4 45.2 2.4 Ratio of Inflow to Outflow Banking 9.1 0.1 109.6 Finance (except banking) 6.9 12.6 0.5 Source: "Foreign Direct Investm ent in the United States" and “ U.S. Direct Investment A b ro a d ,” Survey o f C urrent Business, vol. 70 (A ugust 1990), pp. 54, 55, 97, 98. Notes: M anufacturing, banking, and finance do not sum to total. Direct investment flows relating to the N etherlands A ntilles and to the U.K. C aribbean Isles are sub tracte d from U.S. direct investm ent abroad and foreign direct investm ent in the U nited States, respectively. These adjustm ents are made because outflows to the N etherlands A ntilles in this period essentially reflect repaym ents of E urobonds sold through shell finance affiliates and because outflows to the U.K. C aribbean Isles reflect onle ndin g of the proceed s of com m ercial p a p e r and bond sales by U.S. finance affiliates of nonfinancial foreign corporations via tax havens in the C aribbean. The removal of these flows reduces cum ulative U.S. dire ct investm ent outflows by $20.3 billion and boosts foreign direct investm ent inflows by $2.2 b illion for both the total and the finance com ponent. FRBNY Q uarterly Review/Autum n 1991 15 ences is to estimate the required return on equity— the “cost of equity”— to securities firms in the United States and Japan. Our analysis of a small sample of key publicly traded firms suggests that Japanese securities firms enjoy a substantial cost of equity advantage over U.S. firms. The second step is to show how differences in the cost of equity translate into differences in the cost of capital. Because securities firms, unlike banks, do not have uniform international capital requirements, this step requires care. One complication is that both observed and required shareholder-equity-to-asset ratios of Japanese securities firms are higher than those of U.S. securities firms. The co st of equity We define the cost of equity as the ratio of a firm’s sustainable profits to the market value of its equity. We cannot observe sustainable profits, but we can observe reported profits for a sample of firms and make adjust ments to them. In addition to making reported profits better reflect economic profits, these adjustments make the stated profit measures internationally comparable. Our sample of firms for the United States is neces sarily limited to those whose shares have been publicly traded throughout the sample period. First Boston and Shearson-Lehman are thus excluded because their public shareholders were bought out by their respective parents, Credit Suisse and American Express; Goldman Sachs, Drexel, and Prudential-Bache are excluded by virtue of their private ownership. That leaves Merrill Lynch, Morgan Stanley, and Salomon Brothers of the “bulge bracket,” or lead underwriter, firms and Bear Stearns and Paine Webber of the remaining top ten firms. The selection of Japanese securities firms is quite obvious in light of their dominant status: Daiwa, Nikko, Nomura, and Yamaichi, the so-called Big Four. The sample period runs from 1982 to 1991. The nine and one-half fiscal years covered cannot be syn chronized across the two countries. For all U.S. firms except Bear Stearns and Paine Webber, fiscal years correspond to calendar years and the 1991 observation covers only the first half.5 For the Japanese firms, the half year covers October 1988 to March 1989, an accounting period that permitted their fiscal years to be aligned with general practice in Japan. Because Bear Stearns and Morgan Stanley made their initial public offerings in October 1985 and March 1986, respectively, 5For B ear Stearns, d a ta for fiscal years ending in April through 1987 and in June from 1988 on are a g g re g a te d with the other firms' data for the previous D ecem ber. For Paine W ebber, d a ta for the fiscal year ending in S e p te m b e r are a g g re g a te d with the other firms’ data for the following D ec e m b e r through 1986; in 1987 the firm switched to fiscal years ending in D ecem ber. 16 FRBNY Quarterly Review/Autumn 1991 1985 is the first sample year for each (Morgan Stanley’s public offering price is taken to be its December 1985 price). Altogether, this study’s cost of equity calculations rely on forty-three observations of U.S. securities firms’ share prices, earnings statements, and balance sheets and forty corresponding observations for Japanese securities firms. We adjust reported profits for the following:6 d epreciation— stated earnings are lowered to offset the upward bias introduced when depreciation expenses are based on historical cost during a period of inflation; equity/inflation— the increase in the nominal value of shareholder equity necessary to maintain the real value of shareholder equity is subtracted from stated profits; crossholding— the undistributed profits associated with equity shares held by Japanese firms are con solidated into income; and restru ctu rin g c h a rg e s — U.S. firms’ restructuring charges are spread out over three years. The crossholding adjustment is performed for Jap anese securities firms but not for U.S. securities firms even though both hold significant amounts of equities. The reason for the asymmetry in this adjustment is that U.S. firms mark their equities to market, while Japanese firms do not. Over time, U.S. firms’ marked-to-market equity values reflect retained earnings on equity hold ings insofar as these earnings are embodied in share prices. Japanese firms not only value their equity hold ings at historical cost, but also hold and rarely realize large and growing stakes in their investment accounts for strategic purposes. It is this combination of low turnover and historical cost accounting that requires the crossholding adjustment. Taken together, the adjustments performed on the raw observed ratios of after-tax earnings to market capitalization narrow the differences between the U.S. and Japanese firms significantly (Table 2 )7 Making «Com pare the adjustm ents to bank profits in Z im m er and M cCauley, “Bank Cost of C a p ita l,” pp. 3 6 -4 2 . ^The rows do not sum for U.S. firms in the years 1984, 1 9 8 8 -9 0 , and the average owing to our constraining the cost of e quity to be non negative. This constraint a dds 0 .4 p e rc e n ta g e point to the average cost of equity. Excluding firm -years of com p u te d negative cost of equity would yield an average cost of equity of 8 .6 pe rce n t. Treatm ent of the industry as a single firm — a d d in g earn in g s across firms in a given year and c o m paring the total with sum m ed m arket c a p ita liza tio n s — results in an ave ra g e cost of e quity of 7 .4 p ercen t. a llo w a n c e s fo r in f la t io n ’s e ro s io n o f d e p re c ia tio n e xp e n se s and of s h a re h o ld e rs ’ e q u ity re d u ce s U.S. se c u ritie s firm s’ e a rn in g s to a gre a te r extent than th e ir Ja p anese co u n te rp a rts ’ ea rn in g s, la rg e ly because of the h ig h er rate of inflation e xperienced in the U.S. econom y in the sam ple period. S preading out U.S. firm s’ e x tra o rd in a ry reserves should in p rin cip le sim ply sm ooth th e ir cost of e q u ity but in practice th is a d ju s t m ent in te ra cts w ith share price m ovem ents to w iden the gap a bit. The cro ssh o ld in g a d ju stm e n t narrow s the gap s u b stantially, a fin ding in line w ith previous w ork on d iffe r ences in e q u ity va lu a tion s in the tw o m a rke ts.8 The cro s sh o ld in g a d ju stm e n t fo r Japanese se c u ritie s firm s in the late 1980s is more c o n s iste n t than fo r Japanese b a n ks, e s p e c ia lly c ity b a n ks, in the sam e p eriod. B ecause the city banks cam e u nder pressure to m eet new in te rn a tio n a l ca p ita l s ta n d a rd s and responded in »See James M. Poterba, "C om parin g the Cost of C apital in the United States and Japan: A Survey of M ethods," this Quarterly Review, vol. 15 (W inter 1991), pp. 20-32, and references contained therein. p a rt by re a lizin g m assive g a in s on c ro ssh e ld shares, the cro ssh o ld in g a d ju s tm e n t a c tu a lly su b tra c te d e a rn ings in the three fisca l years to M arch 1990.9 In the sam e period, Ja p a n e se s e c u ritie s firm s, a ctin g like th e ir co rp o ra te cu sto m e rs, te n d e d to e sch e w re a lizin g ga in s on e q u itie s in th e ir in ve stm e n t p o rtfo lio s — and th e re b y avoided the ta xe s a sso cia te d w ith such re a liza tio n s. The resulting co st of e q u ity se rie s show som e v o l a tility but c a rry a c le a r m e ssa g e (C h a rt 1). The J a p anese se c u ritie s firm s in o u r sa m p le face an average cost of e q u ity of 5.1 p e rce n t in the sa m p le p e riod as ag a in st 7.8 p e rce n t fo r the U.S. s e c u ritie s firm s. Such a diffe re n ce is u n like ly to be w ith o u t im p lic a tio n s fo r in te r n ational co m p e titio n . At the sam e tim e , th e adva n ta g e of J a p a n e s e s e c u ritie s firm s is s m a lle r th a n th a t enjoyed by Ja p a n e se b anks (3.1 p e rc e n t co m pared w ith 11.9 pe rce n t fo r U.S. b a n k s )10 or J a p a n e se in d u s tria l firm s (4.5 pe rce n t co m pared w ith 11.2 p e rc e n t fo r U.S. aZimmer and McCauley, "B ank Cost of C a p ita l,” p. 40. 10Zim mer and McCauley, "B ank Cost of C a p ita l,” p. 42. Table 2 Summ ary of Adjustm ents to Cost of Equity (Cross-Firm Averages in Percent) tt tp * lr m s 10.01 11.57 3.67 8.30 10.74 13.55 12.37 8.28 3.31 19.30 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 Averages Japanese Firms A djustm ents Profit/ Market C apitalization 10.11 Profit/ M arket C apita lization 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 Averages E quity/ D epreciation Inflation - 0 .9 8 - 0 .9 7 - 1 .0 0 - 0 .6 9 - 0 .7 3 -1 .5 7 - 1 .4 4 - 1 .3 9 -1 .7 1 -1 .2 1 -1 .1 7 CrossH olding R estructuring Cost of Equity - 1 .7 2 - 1 .5 5 - 1 .7 7 -1 .1 4 - 1 .2 3 - 1 .9 6 - 2 .2 5 - 2 .3 5 -2 .1 6 - 1 .9 4 0 0 0 0 0 0 0 0 0 0 0.21 0 02 0.94 - 0 .2 5 - 0 .1 7 0.78 - 0 .2 6 0.97 3.02 - 2 .3 4 7.51 9.07 1.90 6.22 8.62 10.80 8.57 6.79 4.77 13.82 -1 .8 1 0 0.29 7.81 A djustm ents D epreciation Equity/ Inflation CrossH olding R estructuring 5.12 5.43 6.31 5.28 3.81 3.98 3.61 4.51 7.37 3.06 - 0 .1 0 - 0 .0 8 -0 .0 8 - 0 .0 5 - 0 .0 2 - 0 .0 2 -0 .0 2 - 0 .0 2 - 0 .0 3 - 0 .0 5 - 0 .7 7 - 0 .2 2 - 0 .7 2 - 0 .6 5 - 0 .0 4 - 0 .0 4 - 0 .1 6 - 0 .4 8 - 1 .1 6 —1.17 1.66 1.18 1.15 0.84 0.43 0.56 0.48 0.34 0.60 0.76 0 0 0 0 0 0 0 0 0 0 4.85 - 0 .0 5 -0 .5 4 0.80 0 Cost of Equity 5.91 6.30 6.66 5.42 4.18 4.47 3.91 4.35 6.78 2.60 5.06 Sources: Annual reports; Toyo Keizei Inc., Japan Com pany H andbook; Federal Reserve Bank of New York staff estim ates. FRBNY Q uarterly Review/Autum n 1991 17 in d u s try ).11 T he se fin d in gs are c o n s is te n t w ith m anagers’ actions in the 1980s. C o n sid e r the m atch betw een the observed p attern of fu n d -ra isin g in the e q u ity m arkets and the pattern of a b solute and relative advantage in equity co sts of U.S. and Jap a n e se firm s across industry. First, the a b so lute advantage of Japanese firm s in equity co sts in 1985-89 was reflected in the c o n tra s tin g behav ior of n on fin a n cia l co rp o ra tio n s in the U nited S tates and Japan: U.S. no n fin a n cia l c o rp o ra tio n s retired (net) $500 b illio n w h ile th e ir Jap a n e se co u n te rp a rts issued ¥ 11.4 trillio n , or $80 billio n , n e t.12 S econd, p a rtic u la rly low e q u ity costs help e xplain w hy Ja panese banks raised m o re e q u ity th a n a n y o th e r in d u s try in J a p a n ,13 alth ough ca p ita l regu la tio n also played a role. (U.S. banks were co n stra in e d by regulation from jo in in g th e ir co rp o ra te cu sto m e rs in share re p urchases.) Finally, the "M c C a u le y and Zimmer, "E x p la in in g ," p. 12. 12M argaret H astings P ickering, “A Review of C orporate R estructuring Activity, 1980-90," Board of Governors of the Federal Reserve System Staff Study, no. 161, May 1991; and Bank of Japan, flow of funds data in Econom ic S tatistics Monthly. 13Robert Zielinski and Nigel Holloway, U nequal E quities: Power and Risk in J a p a n ’s Stock Market (Tokyo: Kodansha International, 1991), pp. 184-86. 18 FRBNY Q uarterly Review/Autum n 1991 U.S. se c u ritie s in d u s try stood out as an is s u e r of new eq u ity in the 1980s: B ear S te a rn s, M organ Stanley, and o th e rs m ade in itia l p u b lic o ffe rin g s ,14 and G o ld m a n S achs, S h e a rso n L e h m a n , and P aine W e b b e r so ld e q u ity to S u m ito m o Bank, N ippon Life, and Y asuda Trust, respectively. Moreover, the issu e s of th e U.S. firm s clu ste re d in the m id-1980s, w hen o u r m easured co st of e q u ity was m ost favorable. A llo c a tin g e q u ity to fin a n c ia l a c tiv itie s The required fee or return on a given p ro d u c t o r a c tiv ity is de te rm in ed by the required return on e q u ity and by the am ount of e q u ity a llo tte d to th e p ro d u c t o r activity. If both a U.S. and a J a p a n e se s e c u ritie s firm a llo t the sam e e q u ity to a given p ro d u ct or activity, th en the required fee or return w ill be an equal fra c tio n of each firm ’s cost of equity. A ny d iffe re n ce in th e co s t of e q u ity is then reproduced in the c o st of c a p ita l fo r the p ro d u ct or activity. If U.S. s e c u ritie s firm s lever up th e ir s h a re h o ld e rs’ e q u ity w ith more a sse ts than J a p a n e se s e c u ritie s firm s, it m ight seem safe to co n c lu d e th a t th e y a llo t less e q u ity to a given a c tiv ity than d oes th e ir c o m p e titio n . T his c o n clu sio n does not follow , however. At the o u ts e t, it is easy to overstate the d iffe re n ce in le ve ra g e b ecause U.S. acco u n tin g sta n d a rd s leave s e c u ritie s sold un der agreem ents to re p u rch ase on the b a la n ce s h e e t, w h ile J a p a n e s e a c c o u n tin g ta k e s th e m o ff. E ve n if o n e a d ju sts fo r th is d iscrepancy, however, J a p a n e se s e c u ri tie s firm s rem ain less leveraged, w h e th e r m easured at book or m arket value (Table 3, lin e s 5 and 7). To som e extent, Ja p a n e se s e c u ritie s firm s’ low er le v erage o ffse ts the h ig h e r risk of th e ir a sse ts. By h is to r ical acco u n tin g , U.S. and Ja p a n e se s e c u ritie s firm s have 3 to 4 p e rce n t of th e ir asse ts invested in e q u itie s. By m arket value, however, the J a p a n e se firm s have a lm ost tw ice the e q u ity (Table 3, lin e s 1 and 2). S till, a d iffe re n t m ix of e q u itie s in a sse ts d oes not p ro vid e a fu ll a ccount of the le verage d iffe re n ce . If e q u ity h o ld in g s are s u b tra c te d fro m s h a re h o ld e rs ’ e q u ity, J a p a n e s e firm s rem ain s ig n ific a n tly le ss le v e ra g e d (T able 3, line 6). Even the rem aining d iffe re n ce in le ve ra g e need not im ply that J a p a n e se firm s a llo c a te m ore e q u ity to a given a c tivity in a g ive n m arket. T he lack of in te rn a tio n a l c o o rd in a tio n in the re g u la tio n of th e s e c u ritie s business m ust be re co g n ize d . S e c u ritie s firm s in Japan m ust hold s h a re h o ld e rs’ e q u ity eq u a l to 10 p e rce n t of assets. D espite the a p p lic a tio n of th is s ta n d a rd to both d o m e stic and fo re ig n firm s o p e ra tin g in Tokyo, U.S. firm s have co m p la in e d th a t so high a c a p ita l require14Chris J. M uscarella and Michael R. Vetsuypens, "A S im ple Test of Baron's Model of IPO U n d e rp ric in g ,” Jou rnal o f F inancial E conom ics, vol. 24 (1989), pp. 125-35. m ent is re strictive .15 W hatever the w eight of th is c o n te n tio n , U.S. and Ja pane se firm s in Tokyo require the sam e e q u ity in a given a ctiv ity to, say, a rb itra g e betw een cash and fu tu re s m arke ts in stock. In New York the s u b s id ia rie s of Japanese s e cu ritie s firm s are not bound by Jap a n e se ca p ita l s ta n d a rd s but need o n ly s a tis fy U.S. Treasury and S e cu ritie s and E xchange C o m m ission ca p ita l requirem ents. Indeed, the Big F o u r’s U.S. s u b sid ia rie s operate w ith leverage m ore like that of U.S. firm s than that of th e ir parents (Table 4; Table 3, line 5). W hen in New York, th e se firm s do as New Y orkers do. T he overall d iffe re n ce in leverage, therefore, can be a scrib ed la rg e ly to d iffe re n ce s in capital requirem ents and in the g e o g ra p h ica l m ix of business. Indeed, capital 15Foreign securities firm s have faced the same capital requirements as Japanese firm s since the mid-1980s. See U.S. Treasury, N ational Treatment Study: 1986 U pdate, p. 78; and Report on Primary Dealers Act, A ttachm ent 3, Summ ary of Public Comm ents, pp. 7-8. requirem ents b e tte r e xplain the d iffe re n c e s in le verage than the degree of leverage of e ith e r U.S. or Ja p a n e se firm s since firm s in both c o u n trie s te n d to hold ca p ita l in excess of re q uirem ents. S im ila r leverage w ith in a m a rke t m akes fo r c o st of ca p ita l d iffe re n ce s th a t re fle ct co st of e q u ity d iffe re n ce s. G iven th a t a 10 p e rc e n t e q u ity -to -a s s e t ra tio is required in Japan, if U.S. firm s face a required return on e q u ity of 10 pe rce n t w h ile J a p a n e se firm s face a required return of 5 percent, then the fo rm e r need to earn 1 p e rc e n t on assets in Tokyo w h ile the la tte r can get away w ith 1/2 p e rc e n t. If th e c a p ita l re q u ire m e n t w o rk s o u t to 2 p e rce n t in the U.S. m arket, th e n th e U.S. firm needs to earn 20 basis p o in ts per annum on its a sse ts w hile the Ja p a n e se firm needs to earn o nly 10 b asis p o in ts. In th is m anner the c o st of e q u ity d iffe re n c e s c a rry over into cost of ca p ita l d iffe re n ce s. Explaining cost of capital differences for securities firms The fin d in gs so fa r raise tw o q u e s tio n s : W hy do J a p anese s e c u ritie s firm s cla im an a d va n ta g e in the co s t of e q u ity over th e ir U.S. c o u n te rp a rts ? A nd w h y is the advantage s m a lle r than th a t fo u n d fo r J a p a n e se nonfinancial firm s and banks? Table 3 Selected Balance Sheet Characteristics of U.S. and Japanese Securities Firms (Percent) Japanese Firms U.S. Firms Equity holdings in perspective 1. Equity portfolio/total assets (security holdings at book value) 3.0 3.7 2. Equity portfolio/total assets (security holdings at market value) 6.9 3.7 26.3 87.4 44.5 87.4 11.7 4.3 3. Equity portfolio/shareholder equity (security holdings at book value) 4. Equity portfolio/shareholder equity (security holdings at market value) Leverage 5. Shareholder equity/total assets (security holdings at book value) 6. Shareholder equity less equity holdings/ total assets less equity holdings (security holdings at book value) 7. Shareholder equity/total assets (security holdings at market value) 9.0 14.7 0.38 4.3 Sources: Annual reports; Toyo Keizai Inc., Japan Com pany H an d b o o k; Federal Reserve Bank of New York staff estim ates. Notes: Data are averages for 1986-89. Assets for Japanese firm s includ e gensaki and repurchase agreem ents. For Daiwa, Nikko, and Yamaichi, the market value of securities p o rtfolio is estim ated from net assets at market value less unconsolidated shareholder equity from the Japan Com pany Handbook. For Nomura, whose annual reports detail the market value of secu rities, this difference overstates unrealized gains on secu rities by an average of 6 percent, with a range of 1 to 9 percent. U nrealized gains on D aiw a’s, Nikko's, and Yam aichi’s eq uity ho ld ings alone are estim ated as the product of the diffe ren ce above and .905. M a c ro e c o n o m ic e x p la n a tio n s fo r U .S .-J a p a n e s e d iffe re n c e s 16 Ja panese se c u ritie s firm s share in the re la tive ly low e q u ity costs th a t c h a ra c te riz e d the w h o le Ja p a n e se co rp o ra te s e cto r in the la tte r 1980s. T h e se low co sts are tra ce a b le in la rg e p a rt to m a c ro e c o n o m ic fa cto rs. E ve n th o u g h th e in te r n a tio n a l m o b ility o f c a p ita l increased in the 1980s (as e vid e n ce d by s u b s ta n tia l c ro s sb o rd e r tra n s a c tio n s in e q u ity), c a p ita l co s ts were fa r from e q u a lize d a cro ss c o u n trie s and n a tio n a l fa cto rs still played a p re d o m in a n t role. In Japan, h ig h e r h o u s e hold savings m ade fo r low er e q u ity co sts. In a d d ition , sm o o th e r grow th in Japan, re su ltin g in p a rt from s u c ce ssfu l m a cro e co no m ic policy, m eant low er risk in p ro f its, and low er risk in profits m eant low er c o st of equity. S a fe ty n e t d iffe re n c e s b e tw e e n U.S. a n d J a p a n e se s e c u ritie s firm s We have argued e lse w h e re th a t the risk faced by in v e s tors in the e q u ity of banks d e p e n d s on th e nature of the safety net provided by o ffic ia ls of various c o u n trie s to th e ir banks. Investors in the shares of s e c u ritie s firm s also face s y s te m a tic a lly d iffe re n t risks ow ing to n a tio n a l diffe re n ce s in s a fe ty -n e t c h a ra c te ris tic s . In p a rticu la r, investors in J a p a n e se s e c u ritie s firm s have m ore reason 16M acroeconom ic explanations of U .S .-Japanese cost of ca p ita l differences are discusse d at length in M cCauley and Zimmer, "E xp la in in g ,” pp. 16-20. FRBNY Q uarterly Review/Autum n 1991 19 to su p pose th a t th e ir d ow nside risk is s u b s ta n tia lly le sse ned by the p o s s ib ility of governm ent in te rve n tio n than do investors in the shares of U.S. se c u ritie s firm s. P otential investo rs try in g to im agine the w orst that m ight happen to the value of th e ir shares in a se cu ritie s firm are liable to co n ju re up diffe re n t sce n a rio s for losses in Ja pane se and U.S. se c u ritie s firm s. If they are co n sid e rin g in vesting in shares of a Japanese firm , they m ay w ell call to m ind the d istre ss of Yam aichi S e cu ritie s in the 1960s; if the y are co n sid e rin g investm ent in a U .S. firm , th ey may re a d ily recall the ba n kru p tcy filing of Drexel in 1990. The e sse n tia l fe atu re s of Y a m a ich i’s d iffic u ltie s may be re la te d b rie fly : lo s s e s on sto c k m a rke t h o ld in g s im paired the firm ’s ca p ita l; cu sto m e rs w ithdrew liq u id ity ; the B ank of Japan w orked w ith the M in is try of F inance to p ursue a rescue plan in volving la rg e ly unsecured advances by the B ank of Japan; eventually Yam aichi recovered and repaid the loans over fo u r ye a rs.17 The e sse n tia l featu re s of D re xe l’s d iffic u ltie s may be related w ith equal bre vity: losses on ju n k bonds and bridge loans im paired the firm ’s ca p ita l; providers of w h o le sa le fun d in g w ith d re w liq u id ity ; the S e cu ritie s and E x c h a n g e C o m m is s io n w o rk e d w ith th e F e d e ra l R eserve Bank of New York to achieve an o rd e rly reduc tion of the bala nce sh e e ts of the registered brokerde a le r and the g o vern m en t se cu ritie s su b s id ia rie s ; the firm so u g h t pro te ctio n from its cre d ito rs u nder C hapter 11 of the B a n kru p tcy C ode; and the fate of unsecured 17A ppe ndix to statem ent of E. Gerald Corrigan, President of the Federal Reserve Bank of New York, in D eposit Insurance Reform a n d F inancial M odernization, H earings before the Senate Comm ittee on Banking, Housing, and Urban Affairs, 101st Cong., 2d sess (W ashington, D.C.: Governm ent Printing Office, 1990), pp 82-86, reprinted as "How Safety Nets W ork," Central Banking, Autumn 1990, pp. 61-63. creditors, like th a t of h o ld ers of the firm ’s (u n tra d e d ) equity, rem ains u n c le a r at th is ju n c tu re .18 The s trikin g co n tra s t betw een th e se tw o e p is o d e s, of course, p rovides no c e rta in g u id e to how a tro u b le d s e cu ritie s firm w ould be han d le d in the fu tu re . C e rta in ly the contexts of the o fficia l a ctio n s d iffe re d : g e n e ra lly low share p rices reflected g e n e ra l e co n o m ic w e a kness in Japan in 1962, w h ile D re xe l’s d iffic u ltie s cam e late in an e co n o m ic upsw ing. N e ve rth e le ss, m a rke t p a rtic i pants may well view the e q u ity of a m a jo r U.S. s e c u ri ties firm as s u b je ct to one more risk than th a t of a m ajor Japanese se c u ritie s firm . M a rk e t m easures o f ris k M arket m easures of risk show Ja p a n e se s e c u ritie s firm s to be, if anyth in g , a bit ris k ie r than th e ir U.S. c o u n te r parts. B ecause Ja p a n e se s e c u ritie s firm s are m uch less leveraged than U.S. firm s, th e y sh ould e x h ib it low er stock betas, given equal ris k in e s s of a s s e ts .19 B ut in fact the sto ck betas of J a p a n e se s e c u rity firm s have averaged 1.46 over the p e riod 1987-91, as co m pare d w ith 1.29 fo r U .S. s e c u ritie s firm s o ve r th e p e rio d 1986-91, and the d iffe re n ce is even m ore s trik in g for 18C hristopher Byron, “ Drexel's Fall: The Final D ays," New York, M arch 19, 1990, pp 32-38. 19S tarting with the relationship ba = w x be + ( 1 - w ) x bd, where ba = w = be = 0d = asset beta equity/asset ratio equity beta bond beta, we have dt>e/dw = w ~ 1 x { b a - be + [(1 - w) x dt>d/dw]}. Given that ba and d£>d/dw are small and of opposite sign, we have dbe/dw < 0. If we further assum e that bond betas are ge nera lly n egligible, we have dbe/dw = - b j w. Table 4 Shareholders’ Equity as a Share of Total Assets for U.S. Affiliates of Japanese Securities Firms (Percent) Date Daiwa S eptem ber 1985 S eptem ber 1986 S eptem ber 1987 S eptem ber 1988 M arch 1989 M arch 1990 M arch 1991 1 38 1.05 0.92 0,99 0.85 1.17 Period average 1.06 Nikko Yamaichi Average 2.96 2.90 2.22 1.64 1.90 5.91 2.30 1.48 2.00 1.57 1.38 1,41 Nomura 2.45 1.84 0.90 0.92 0.96 5.91 1.84 1.99 1.92 1.42 1.20 1.36 2.32 2.29 1.41 1.77 Source: Annual reports. Note: For Nomura and Yamaichi, M arch figures for 1989 and 1990 are averages of S eptem ber 1988 and S eptem ber 1989 and S eptem ber 1989 and S eptem ber 1990, respectively. 20 FRBNY Q uarterly Review/Autumn 1991 years oth er than 1990 (Tables 5 and 6). F in a n c ia l d e re g u la tio n a n d the in s e c u rity o f Ja panese s e c u ritie s firm s’ e a rn in g s Investors in the Big F ou r’s shares may well perceive a risk of more con cern than b a n kru p tcy or the shares’ e xa ggerated response to g e neral m arket m ovem ents. P rospective dere g u la tio n is w id e ly viewed as a threat to the firm s’ revenues, and the risk of an adverse change in the rules can boost the m easured cost of e q u ity for Ja p anese se cu ritie s firm s relative to Japanese firm s in g e n eral. In ad d ition , if inve sto rs a n ticip ate a d e clin e in the p ro fita b ility of Jap ane se se cu ritie s firm s, then the cu rre nt relation of th e ir e a rn in g s to the m arket valuation of th e ir shares w ill tend to ove rsta te th e ir cost of equity unless the stock m arket is ve ry m yopic. E vidence s u g g ests that inve sto rs in the shares of the Big Four s e c u ri ties firm s do fe a r lower p ro fita b ility going forward. Japan ese se cu ritie s firm s resem ble U.S. secu ritie s firm s in the m id-1970s in th e ir d e p e n d e n c e on e q u ity co m m issio n s as a so u rce of revenue. U.S. s e c u ritie s firm s drew about half of all revenues from e q u ity c o m m issions w hen th e y were lib e ra liz e d in May 1975 (C h a rt 2). S ince then, the share of co m m is s io n s in in d u s try revenues has fa lle n below a fifth . By co n tra s t, the large Japanese s e c u ritie s firm s have d e p e n d e d and co n tin u e to depend on e q u ity co m m is s io n s fo r a b o u t h a lf of th e ir revenue (C h a rt 3). Investors need o n ly e x tra p o la te a trend to foresee th a t th e se revenues w ill sh rin k over the m edium term . The Japanese a u th o ritie s have been re ducing e q u ity co m m issio n rates g ra d u a lly (C h a rt 4). O ver the last decade, co m m issio n rates fell at an an n u a l rate of 1 perce n t fo r tra d e s of 1 m illio n yen (a b o u t $7000), 1.6 perce n t for tra d e s of 10 m illio n yen ($7 0 ,0 0 0 ), 5.5 per cent for tra d e s of 100 m illio n yen ($70 0 ,0 0 0 ), 13.4 per cent fo r tra d e s of 1 b illio n yen ($7 m illion), and 18.9 perce n t fo r tra d e s of 10 b illio n ($70 m illion). Table 5 Relation of U.S. Securities Firms’ Share Returns to Returns on the Standard and Poor’s 500 Index M errill Lynch Morgan Stanley Salom on Brothers Period Beta S tandard Error R2 Beta S tandard Error R2 Beta S tandard Error R2 1986-91 1986 1987 1988 1989 1990 1991 (26 weeks) 1.30* 0.81 1.29 0.88 2.09* 1.45 1.67 0.085 0.19 0.15 0 15 0.23 0.24 0.34 45 .26 60 40 .61 .41 .50 1.14 1.17 1.30 0.87 0 88 1.06 1.37 0.088 0.21 0.17 0 20 0.25 0.22 0 36 38 .45 .53 .27 .19 .32 .38 1 44* 1.64* 1 55* 1.33 1.06 1.31 1.61 0.089 0.21 0.20 0.21 0.22 0.20 0.38 48 .56 .55 .45 .33 .47 .43 Source: S tandard and Poor’s Note: Data are weekly. 'B e ta is significa ntly different from one on a tw o-tailed test at 5 percent significance Table 6 Relation of Japanese Securities Firms’ Share Returns to Returns on the TOPIX Index Daiwa Nikko Yamaichi Nomura Period Beta S tandard Error R2 Beta Standard Error R2 Beta S tandard Error R2 Beta S tandard Error R2 1987-91 1987 1988 1989 1990 1991 (25 w eeks) 1.62* 2 07* 2.7 9’ 2.03* 1.05 2.05* 0.09 0.21 0.27 0.27 0.13 0.28 .57 .68 69 .54 .58 71 1.34* 1.53* 2.44* 1.99* .91 1.33 0.09 0.21 0.24 0.25 0.11 0.31 .50 .53 .66 .55 56 .45 1.42* 1.63* 1 99* 1.68* 1.10 1.95* 0.07 0.16 0.21 0.22 0.12 0.19 .62 .70 .63 .53 64 82 1.44* 1.79* 2.53* 1.89’ .98 1 39 0.09 0 22 0.23 0 26 0.11 0.24 .55 59 .71 .52 61 .60 Source: Daiwa and Dow Jones Tradeline International. Note: Data are weekly. ’ Beta is significa ntly different from one on a two-tailed test at 5 percent significance. FRBNY Q uarterly R eview/Autum n 1991 21 N ote th a t lib e ra liz a tio n of c o m m is s io n s h u rts the s e c u ritie s firm s more than the lib e ra liza tio n of interest rates ever h u rt Japa n e se banks. C om p e titio n am ong th e b an ks fo r borrow ers kept the spread betw een aver age d e p o sit rates and prim e lending rates fa irly narrow by in te rn a tio n a l stand a rd s. R egulation of com m ission rates proved m uch more e ffe ctive in protecting the reve nues of the s e cu ritie s firm s. R e in forcing the trend tow ard co m m issio n d eregulation was a 1988 regulation th a t sh ra n k the Big F our’s share of e q u ity b ro ke ra g e . The M in is try of F in a n ce was re p orted to have a dvised s e c u ritie s firm s not to perform m ore than 30 percen t of d a ily trading in any single share. T h is g uid ance , aim ed at excesses a ssociated w ith th e m a tic pro m otio ns of the Big Four, co n trib u te d to a d e clin e in th e ir share of e q u ity brokerage from 60 p e rce n t in 1981 to 46 p e rce n t in the m iddle of the d ecade to 33 percent at the end of the d e ca d e .20 As a ^ S a to sh i Takeuchi, "B ig Four's Transaction Share No Longer So Big; 30% Cap on Trade Volume H obbles Strategy to Promote Selected Issues,” Japan E conom ic Journal, O ctober 28, 1989, p. 2. The article notes that “ the gu id elines em erged after the U.S. governm ent's special body on stock trading, the Brady Comm ission, sharply c ritic iz e d the Big Four’s o lig o p o lis tic control [and] accused the Big Four of m anipulating stock price s by con ducting con certe d result of co m m issio n cu ts and lo st m a rke t share, Big Four c o m m issio n s show ed little of the bu o ya n cy of the trading value of J a p a n e se e q u itie s (C h a rt 5). N ote th a t th e v a lu e o f s h a re tu r n o v e r on th e T o k y o S to c k E xchange reflected not o n ly the p e rfo rm a n c e of share prices but also the c le a r dow nw ard trend in share v o l ume from the b e g inn in g of 1988. F u rth e r a n a lysis of the Big Four co m m is s io n incom e confirm s the e rosion of th e ir revenue base in the m id st of the boom m a rke t of the late 1980s. We relate the log of annual co m m issio n incom e fo r each of th e Big Four for 1983-91 to a tim e trend and to the log of the value of shares tra d e d on the Tokyo S to ck E xch a n g e . The e s ti m ated co e ffic ie n t fo r tim e s u g g e sts th a t w hen the value of tra d e s on the Tokyo S to ck E xch a n g e is held co n sta n t, c o m m issio n revenue te n d s to d e clin e 4.7 p e rc e n t per year. T his rate lies w ith in the sp e ctru m of ra tes of d e clin e fo r regulated co m m is s io n s over the d e ca d e , as o u tlin ed a b o ve — 1 p e rce n t to 18 p e rc e n t— and is close to the rate of d e clin e fo r co m m is s io n s a s s o c ia te d w ith Footnote 20 (continued) buying operations based on sp e cific them es." See R eport o f the P residential Task Force on M arket M echanism (W ashington, D C.: Government Printing Office, 1988), p. I-8. Chart 2 Composition of Revenues of the U.S. Securities Industry Commissions q I------1— I------1------1------1------1------ 1------ 1------ 1— I------ 1------1------ 1------1------1— 1 1975 77 79 81 83 85 87 89 91 1984 85 86 87 88 89 90 Source: Securities Industry Association. Note: Data for 1991 cover first half of year only. 22 FRBNY Q uarterly Review/Autumn 1991 Sources: Daiwa, Nikko, Nomura, and Yamaichi annual reports. 91 an 80 m illio n yen trade. A llo w in g fo r 10 perce n t grow th in tra d in g value and other, no n co m m issio n revenues, in ve sto rs may readily foresee co m m issio n incom e d ro p ping to less than a q u a rte r of the Big F our’s revenues over the next fifte e n ye a rs.21 The Big F our’s co m m issio n incom e is q uite re sp o n sive to the sto ck m a rk e t’s p e rform ance. O ur regression a n a ly sis su g g e sts th a t a 10 p e rce n t rise in the value of sto ck m arket tra d in g yie ld s a 7.1 pe rce n t increase in Big Four co m m issio n revenues (Table 7). Big Four c o m m is sions did not respond one-fo r-o n e to the value of tra d in g because rising share p rice s tended to push tra n sa ctio n values along the d e clin in g sch e d u le of co m m issio n s and b ecause th e ir m a rket share was d e clin in g . In vestors in the shares of Ja p a n e se se c u ritie s firm s m ust pay a tte n tio n to the la rg e r agenda of dere g u la tio n th a t in clu d e s a re co n sid e ra tio n of the A rtic le 65 b a rrie rs betw een se cu ritie s and b anking businesses. A lready 21lf tra d in g value rises at 10 pe rcent per annum, if the elasticity of com m issions with respect to tra ding value is .71, and if deregulation continues to put a 4.7 percent per annum drag on com m issions, then com m issions will grow at 2.1 pe rcent per annum. If other revenues start off equal to com m issions and grow at 10 pe rcent per annum, then it will take fifteen years for com m issions to fall to a qu arte r of revenues. In other words, (1.021 )x = (1.1 )x/3; solving for x, we have fifteen. the J a p a n e s e c ity b a n ks have e q u ity s ta k e -o u ts in s m a lle r s e c u ritie s firm s th a t co u ld be c a p ita liz e d upon were A rtic le 65 repealed or m odified to p e rm it bank e n try in to b ro k e rin g J a p a n e s e s h a re s . Even if the change in the law now th o u g h t m ost lik e ly w ill not perm it banks to b ro ke r shares, in ve sto rs n e ve rth e le s s w ill have borne the risk th a t a m ore sw eeping d e re g u la tion poses to s e c u ritie s firm s’ revenues and p ro fits .22 Finally, in ve sto rs may p e rce ive th a t th e e n try of fo r eign se c u ritie s firm s may p re se n t a th re a t to the c o m m ission revenue of the large J a p a n e se s e c u ritie s firm s. Foreign firm s have b ro u g h t w e ll-d e v e lo p e d te c h n ic a l tra d in g ta c tic s and more c ritic a l re search to th e ir b id ding for in s titu tio n a l tra d e s . W ith th e s e a d vantages, they have raised th e ir share of tra d in g on the Tokyo S tock E xchange from 1.5 p e rc e n t in 1986 to 5.4 p e rce n t in 1989 and 7.3 p e rce n t in the first h a lf of 1990.23 22“ While the entry of the banks into ce rta in areas of secu rities business is now a foregone conclusion, the speed with w hich such reforms will be im plem ented, the scope of the banks' new b u si nesses, the form w hich en try will take, and the new questions surrounding the banks’ ab ility to expand aggressively w hile international capital ad equa cy requirem ents still seem a problem for them, all com bine to sug gest a pictu re w hich is not as b la ck as orig inally pe rce ive d ” (A licia Ogawa, "D aiw a S e cu ritie s ,” S. G. W arburg Securities, M arch 26, 1990, p. 12). 23Business Week, July 9, 1990, p. 60. In N ational Treatment Study, 1990, p. 236, the U.S. Treasury cites the "m arket pow er” of the Big Chart 4 Equity Commission Rates in Japan Effective Dates ---------- April 1 ,1 9 7 7 ---------- April 15, 1985 ---------- November 25, 1986 —— - October 5, 1987 June 4, 1990 1 10 100 1,000 10,000 Value of trade in millions of yen Source: Tokyo S tock E xchange Fa ct Book, various issues. FRBNY Q uarterly Review/Autum n 1991 23 O ne m easure of the loss of fra n ch ise value of the Ja p a n e se s e cu ritie s firm s is the ratio of m arket value to b o o k v a lu e . T h e s e firm s ’ m a rk e t-to -b o o k ra tio has de clin e d as com m issio n s have been reduced (C h art 6). Note th a t the spate of p u b lic share o ffe rin g s in late 1985 and ea rly 1986 by U.S. s e c u ritie s firm s, in cluding B ear S te arn s and M organ Stanley, were well tim ed by this m easure. The p o s s ib ilitie s of a d d itio n a l com m ission cuts, Ja p anese bank co m p e titio n , and fu rth e r pen e tra tio n by foreign firm s all represent risks that investors in Big Four shares take into account. It is u n d e rsta n d a b le if inve sto rs in the shares of the Big Four d isco u nt current ea rn in g s som e w hat to allow for ch e a p e r sto ck trading for Japa nese househ o ld s and in s titu tio n s .24 As a result of the Big F ou r’s p ro b le m a tic grow th prospects, the m easured cost of eq u ity for th e se firm s may be higher than th a t of Jap anese firm s in general. In d u s tria l o rg a n iza tio n A n o th e r fa cto r je o p a rd iz in g the e a rnings of the J a p anese se cu rity firm s is the perip h e ra l p o sition of the J a p a n e s e s e c u rity firm s in th e c o u n try ’s in d u s tria l org a n iza tio n . A Japa n e se c ity bank is at or near the Footnote 23 (continued) Four to acco unt for the minim al shares acco rded foreign firms in un derw riting syndicates in Tokyo. In underw riting carve-outs of U.S. firms, however, U .S.-based underw riters have played im portant roles. See Ted Fikre, “ Equity Carve-Outs in Tokyo,” this Quarterly Review, vol. 15 (W inter 1991), pp. 60-64. 24A major rating firm cited “ concerns about future profitability in light of structural changes that are currently taking place in the dom estic Japanese financial m a rket,” including “ lower dom estic equity brokerage com m ission rates and ongoing discussions about financial reform s,” in w arning investors of possible dow ngradings. Standard and Poor's C redit Week, May 13, 1991, p. 19. c e n te r of a k e ire ts u , a n e tw o rk of firm a ffilia tio n s tha t a p p ro x im a te a c ro s s -s e c tio n of th e e co n o m y. T h is arra n g em e n t a ffe cts the co st of e q u ity d ire c tly th ro u g h the sto ck m arket: extensive cross sh a re h o ld in g w ith in the keiretsu may s ta b iliz e and p e rh a p s even elevate share prices. Indirectly, the keiretsu stru c tu re assures ste a d ie r bu sin ess flow s and p ro vid e s im p lic it g u a ra n tees of a ssista n ce to tro u b le d m em bers, b e n e fits th a t in turn help to s ta b iliz e profit flow s. The p e rip h e ra l p o sitio n of the Big Four s e c u ritie s firm s is e vid e nt in the reference w ork In d u s tria l G ro u p in g s in Ja p a n .25 Three of the Big Four a p p e a r o n ly once each and no group a ffilia tio n is given. By co n tra s t, nine of the eleven Ja p a n e se banks e xam ined in o u r s tu d y of bank c o s t of c a p ita l a n c h o r w e ll-d e fin e d in d u s tria l groups. E ven th e e x c e p tio n a l J a p a n e s e s e c u r itie s firm broadly co n fo rm s to the patte rn . N ikko S e c u ritie s is listed as a sso cia te d w ith the M itsu b ish i group, but the a ffilia tio n is d e scrib e d as w eak. The a g g re g a te e q u ity stake in N ikko held by M its u b is h i g ro u p c o m p a n ie s, m easured a g a inst the o ve ra ll c o n c e n tra tio n of s h a re h o ldings in the s e c u ritie s firm , s u p p o rts th a t ch a ra c25Eighth ed. (Tokyo: Dodwell M arketing C onsultants, S eptem ber 1988), pp. 34-35, 49, 128, 304, 306, 506, 512. Chart 6 Ratio of Market to Book Value for U.S. and Japanese Securities Firms Ratio 7 -------------------------------------------------------------------------------------------------Japanese firms with equity holdings at book value / « / \ ~ % \ Table 7 Regression Analysis of Japanese Securities Firms’ Com m ission income September, 1984-91 Dependent variable Natural Log of Commission Independent variables Time Natural log of Tokyo Stock Exchange trading value Intercept -.0 4 7 (.012) .713 (0 5 0 ) 9.31 (.146) .87 R2 Degrees of freedom 0 I I I I 1i 1 1 I I I I I I 1I I I 1I M 1I I I 1I I I 1I I 1■ 1-L-J 1982 83 84 85 86 87 88 89 90 91 33 Note: Standard error of coefficients is given in parentheses. 24 FRBNY Q uarterly Review/Autum n 1991 Sources: Annual reports; Toyo Keizai Inc., Ja pa n C om pany Handbook-, and Federal Reserve Bank of New York staff estimates. te riza tio n . The M itsu b ish i g ro u p ’s aggregate holding of N ik ko ’s e q u ity am o unts to no more than a third of the top ten s h a re h o ld e rs’ co lle ctive stake. By co n tra st, the M itsu bish i g ro u p ’s h o lding of M itsubishi B a n k ’s shares bulks m uch large r: alm o st tw o -th ird s of the top ten sh a re h o ld e rs’ s ta k e .26 O nly 13 of the 128 firm s in the M itsu bish i group show lower group “ in flu e n c e ” ratios than does N ikko S e cu ritie s. Moreover, N ikko has no d ire cto rs from M itsubishi group com panies w hile M it subishi Bank has two. R eversing the p e rsp e ctive to exam ine fin a n cial firm s’ h o ld in g s o f e q u itie s c o n firm s th a t s e c u ritie s firm s rem ain m uch less well co n n e cte d than Ja panese banks. The se cu ritie s firm s chan n e le d p a rt of th e ir strong flow of retained ea rn in g s during the boom years of the 1980s into accu m u la tin g e q u ity stakes. As a result, se cu ritie s firm s increased th e ir stra te g ic share of e xch a n g e -liste d firm s fa ste r than banks did in the 1980s, e sp e cia lly if “ m ost of the increase in bank e q u ity o w n e rs h ip ” was “ not .. . fo r stab le share-o w n in g purposes [but rather] fo r sh o rt-te rm investm en t p u rp o se s.” 27 In M arch 1990, alm o st fo u r-fifth s of N o m u ra ’s e q u ity holdings by value were held in the in ve stm e n t a ccount; such shares “ are acquired for the C o m p a n y’s o p e ra tin g purposes and are rarely sold under a C om pany policy.”28 S till, Japanese b a n k s ’ s ta k e in firm s lis te d on th e T o kyo S to c k E xchange rem ains ab out ten tim e s d e e p e r than th a t of Ja p ane se s e cu ritie s firm s (Table 8). At the firm level, exam in a tio n of the se c u ritie s firm s’ m ajor ho ld in g s in J a p a n ’s top com panies show s the ho ldin gs to be few er and more c o n ce n tra te d than those of the banks. A lth o u g h som e o b se rve rs contend that “ N om ura is a ctive ly bu ild in g its own keiretsu of n o n in d u s tria l co m p a n ie s in a va rie ty of sectors in clu d in g real estate, in su ra n ce , d is trib u tio n , research, tra in in g , and a d v e rtiz in g ,” 29 N om ura has not broken into the to p tie r of o w n e rsh ip of firm s tra d e d in the firs t s e ctio n of the Tokyo S tock E xchange. A se a rch of the to p e ig h t o r ten sh a re h o ld e rs in each of the 1254 firm s listed on the Tokyo S tock E x c h a n g e ’s firs t se ctio n fo u n d o n ly th irty three s h a re h o ld in g s of the Big Four s e c u ritie s firm s (Table 9). N om ura a cco u n te d fo r half of the se , but its h o ld in g s w ere q u ite c o n c e n tra te d in fin a n c ia l firm s , in cluding the shares of tw o of its own m a jo r s h a re holders, Daiwa Bank and Toyo Trust. By co n tra st, the oth e r 3 se c u ritie s firm s were not re p re se nte d am ong the top sh a re h o ld e rs of any of th e ir own top sh a re h o ld e rs. W hatever the d iffe re n ce s am ong th e m a jo r s e c u ritie s firm s, none of them has h o ld in g s a p p ro a ch in g the near c ro ss-se ctio n of co rp o ra te Japan ow ned by the city banks. The u se fu ln e ss of the lim ite d e q u ity sta ke s th a t the Big Four do p o sse ss is s u g g e ste d by th e ir role as un d e rw rite rs fo r 22 out of 24 of the firm s in w h ich th e y hold m ajor sh a re h o ld in g s. In all but tw o ca se s fo r w hich an u n d e rw rite r is liste d , the s e c u ritie s firm w ith the e q u ity stake is at least co -le a d u n d e rw rite r, u su a lly m ain u n d e rw rite r, and o fte n s o le u n d e rw rite r. T h is strong pattern su g g e sts th a t e q u ity s ta ke s ce m e n t b u s i ness relations and c o n s e q u e n tly u n d e rs c o re s the th re a t to u n d e rw ritin g incom e a risin g from exp a n d e d pow ers fo r banks. C om bined w ith p ro sp e ctive d e re g u la tio n , the more ce n tra l p o sition of banks in the stru c tu re of co rp o ra te n etw orks renders the e a rn in g s of the s e c u ritie s firm s insecure. If banks are allow ed to e n te r the w h o le sa le se c u ritie s m a rkets, c o rp o ra tio n s m ay well favor th e ir banks in the face of ro u g h ly co m p a ra b le p ricin g of p rospective deals. For th is reason, u n d e rw ritin g reve nues could be p a rtic u la rly at risk. A co m p a riso n of the re sp o n se s to Y a m a ic h i’s d istre ss 26M itsubishi group com panies held 8.8 percent of Nikko Securities' shares, w hile the top ten held 26.4 percent. M itsubishi group com panies held 18.8 percent of M itsubishi Bank's shares, while the top ten held 29.5 percent. MRichard W. W right and G unter A. Pauli, The S econd Wave (New York: St. Martin's Press, 1987), p. 71. M artin French, “ Japan's Great Finance Plan," Aslamoney, July-A ugust 1991, p. 35, also suggests that Nomura might establish itself at the cen ter of a m ajor industrial group. The a rticle also associates Daiwa S ecurities with the Sumitomo group and Yamaichi S ecurities with the Fuyo group. 27W. Carl Kester, Japanese Takeovers (Boston: Harvard Business School Press, 1991), p. 207. 28Nomura Securities Company, Annual Report 1990, p. 23. Table 8 Share of Tokyo Stock Market Owned by Japanese Securities Firms and Banks . S ecurities firms Banks ; " : . :a: 1982 1983 1984 1.6 1.7 1.7 17.5 18.0 17.7 ; 4 || .... . . :, : " . . : . >. , ... : ■: ^ 1985 1986 1987 1.8 17.4 1.9 2.1 2.3 2.3 2.0 18.4 19.3 19.8 21.3 21.3 1988 1989 1990 Source: Tokyo Stock Exchange. FRBNY Q uarterly R eview/Autum n 1991 25 and the tro u b le s of a w e ll-co n n e cte d a u tom obile m aker h ig h lig h ts the gre a te r risk a tte n d a n t on the se cu ritie s firm s’ re lative ly perip h e ra l p o sitio n (although the sheer size of Y am a ichi’s problem may have had so m e th in g to do w ith the diffe re n ce in h a n d lin g the tw o cases). W hile the a u to m o b ile firm M azda was helped through a period of d istre ss by its main bank and affiliated c o m p a n ie s,30 Yam aichi had to resort d ire ctly to the governm ent. Conclusions U.S. s e cu ritie s firm s m ust c le a r a higher cost of equity hu rd le in pricing th e ir p ro d u cts and s e rv ic e s than th e ir “ R ichard Pascale and Thomas P. Rohlen, “ The Mazda Turnaround,” Journal o f Japanese Studies, vol. 9 (Summer 1983), pp. 219-63. Japanese c o u n te rp a rts . H ig h e r c a p ita l re q u ire m e n ts in Japan may put U.S. firm s at a p a rtic u la r d is a d v a n ta g e in com p e tin g there. Factors c o n trib u tin g to low er co sts fo r J a p a n e se firm s in th e 1 9 8 0 s w e re h ig h e r h o u s e h o ld s a v in g s and sm o o th e r e co n o m ic g row th. In a d d itio n , a c o m p a riso n of the exp e rie n ce of tro u b le d s e c u ritie s firm s in the U nited S tates and Japan s u g g e sts a w id e r sa fe ty net in Japan th a t m ay low er e q u ity costs. Ja panese s e c u ritie s firm s seem to have a s m a lle r cost of e q u ity a dvantage over th e ir U.S. c o u n te rp a rts than J a p a n e se n o n fin a n cia l firm s and banks have over th e ir respective c o u n te rp a rts . In p a rt, J a p a n e se in ve s tors bear a risk of low er e a rn in g s fo r Ja p a n e s e s e c u ri ties firm s in a d e re g u la te d e n v iro n m e n t, and th is risk Table 9 Japanese Securities Firms’ Equity Stakes in Firms Listed on the Tokyo Stock Exchange First Section U nde rw rite r Status S ecurities Firm S ector Firm Nomura Financial Daiwa Bank Toyo Trust Dai-Tokyo Fire & M arine C hiba Bank Osaka Securities Finance Japan S ecurities Finance Kokusai S ecurities Sanyo S ecurities N onfinancial M anufacturing Retail trade C om m unications C onstruction Nikko Transport Fishing Financial M anufacturing Daiwa Yamaichi M anufacturing Retailing C onstruction Financial Percent Stake Hokko C hem ical Nissho (m edical equipm ent) Toyo Denki, Seizo (railroad equipm ent) Sogo (d epartm ent store) N ippon Television Network Nissan C onstruction Daiwa Danchi H itachi Transport Hoko Tokyo Securities Toyo S ecurities Maruman S ecurities Kosei S ecurities Japan Securities Finance Tateho Chem ical Ikegai (m achine tools) Nissan Nohrin Kogyo (plyw ood) Kyodo Printing N ippon Conveyor Nihon Matai (food containers) Senshukai Morimoto N ippon Trust Bank K ita-N ippon Bank Taiheiyo S ecurities 3.1 6.9 9.2 1.7 17.0 3.4 32.5 8.1 Sole Main Co Sub Not X X X X X 4.9 1.4 X 2.4 3.9 4.3 2.1 6.5 0.8 5.1 33.6 6.4 4.9 4.1 5.0 4.2 1.7 X X X x X X X X X x 4.2 2.5 1.8 X 3.8 3.5 3.5 1.6 4.1 4.1 X X X X x X X Source: Toyo Keizai Inc., Japan Com pany H a n d b o o k -F irs t Section, W inter 1990. Notes: Nomura com prises Nomura S ecurities and Nomura Land and B uilding, and Nikko com prises N ikko S ecurities, N ikko B uilding , and N ikko Investm ent Trust. No underw riters are listed for the securities firms in w hich the Big Four own stakes. 26 FRBNY Q uarterly Review/Autumn 1991 boosts their measured cost of equity. In addition, the distance of Japanese securities firms from corporate networks of mutual support may render their shares more risky than the shares of firms secure within such networks. FRBNY Quarterly Review/Autumn 1991 27 Financial Liberalization and Monetary Control in Japan by Bruce Kasman and Anthony P. Rodrigues The last fifteen years have witnessed a substantial liberalization of Japan’s financial markets. Controls on cross-border capital flows have been gradually dis mantled and restrictions affecting competition and price flexibility in domestic financial markets have been relaxed. As a result, the range of free market assets has grown significantly, as has the range of credit sources available to domestic borrowers. The experience of other industrial countries indicates that changes in financial structure can have important implications for the conduct of monetary policy. A num ber of countries substantially revised their operating procedures during the past decade as financial market changes altered the relationships between policy tools and objectives. This article examines the effects of financial reforms on Japanese monetary policy. In the first section of the article we discuss how the Bank of Japan has altered its operating strategy in response to the evolving financial environment. We focus in particular on changes in the intermediate objectives of monetary policy and in the instruments used to implement policy. Our analysis sug gests that the complex system of controls prevailing in the m id -1970s supported an operating strategy designed to influence the supply of bank credit. With the relaxation of these controls, monetary policy authorities shifted their strategy away from the control of credit aggregates and, in recent years, have increasingly em phasized interest rates as an agent of policy transmission. The a rticle ’s second section offers an empirical http://fraser.stlouisfed.org/ 28 FRBNY Quarterly Review/Autumn 1991 Federal Reserve Bank of St. Louis assessment of the monetary control mechanism in the current liberalized environment. Specifically, we evalu ate the degree to which the Bank of Japan has been able to influence market interest rates and broad money through interbank interest rates, its chief operating tar get. We find that monetary policy changes have elicited strong and consistent interest rate responses across the term structure in recent years. In particular, long term bond yields are much more responsive to mone tary policy actions than in the past. In contrast, our analysis of the relationship between money and interest rates indicates that as financial reform has reduced policy makers’ direct influence over banks, the link between policy and broad money may have weakened. Our results do not address the extent to which mone tary policy actions have been transmitted to real activity or prices. Nonetheless, our findings suggest that the Bank of Japan has successfully adapted its operating strategy to the changing financial environment. Evolution of monetary control in Japan In Japan, as in most countries, the ultimate goals of monetary policy are output growth and inflation man agement. The authorities typically tighten monetary pol icy to reduce inflationary pressures and ease policy to stimulate activity. Output and prices are controlled only indirectly and with lags, however. Policy actions first affect financial markets and only over time can be expected to influence real activity and prices. Because financial markets play a central role in trans mitting monetary policy, policy makers generally base th e ir op e ra tin g stra te g y on fin a n cia l v a ria b le s .1 In par ticular, a fin a n cial varia ble su b je ct to a high degree of co n tro l by a u th o ritie s u sua lly se rve s as the ta rg e t for d a y-to-day o p e ra tio n s. Borrow ed reserves and the fed funds rate are g e n e ra lly view ed as the cu rre n t o perating ta rg e ts em ployed by the Federal R eserve in the U nited S ta te s.2 In Japan, the reserve progress ratio, the ratio of reserves accu m ula ted w ith in a m onthly m aintenance period to tota l required reserves, and in te rb a n k interest ra te s h a v e s e rv e d a s im ila r fu n c tio n s in c e th e m id-1970s. F in ancia l varia b le s are also em ployed as in te rm e d ia te ta rg e ts or in d ica to rs. As the term “ in te rm e d ia te “ s u g g ests, th e se varia b le s fit betw een the in stru m e n ts and o p e ra tin g ta rg e ts of policy, on the one hand, and the u ltim ate po licy goals, on the other. To be e ffective, an in te rm e d ia te va riab le should provide inform ation about p o licy goals and bear som e relation to op e ra tin g ta r g ets. In the late 1970s and ea rly 1980s a num ber of c e n tra l banks used a m o n e ta ry aggregate as a key in te rm e d ia te variable, in m any cases se ttin g e xp licit ta rg e ts fo r its annual grow th. In recent years, reliance on m o n e ta ry aggregates as e xp licit in te rm e d ia te ta rg e ts has d im in ish e d and a tte ntio n has sh ifte d to a w id e r set of fin a n cial m arket variables. 1A more com plete d e scription of the role of targets and indicators in the im plem entation of m onetary p o licy can be found in Richard G. Davis, "In term ediate Targets and Indicators for M onetary Policy: An Introduction to the Issues,” this Q uarterly Review, Summer 1990. 2Borrowed reserves are obtained by banks directly from the Federal Reserve discou nt window. The federal funds rate is the rate d e pository institutions cha rge one another to borrow reserves. The m ovem ent away from m o n e ta ry ta rg e tin g has la rg e ly stem m ed from c h a n g e s in the fin a n c ia l e n v iro n m ent. The re m a in d e r of th is s e ctio n c o n s id e rs how dere g u la tio n , g lo b a liz a tio n , and in n o vatio n in J a p a n ’s fin a n c ia l m a rk e ts o ve r th e p a st tw o d e c a d e s have s h a p e d th e B a n k of J a p a n ’s p o lic y a n d o p e ra tin g strategy. The m on e ta ry c o n tro l m e c h a n is m : m id-1970s Until the m id-1970s, the J a p a n e se fin a n c ia l syste m was highly regulated. A com plex system of c o n tro ls had evolved, lim itin g in te re st rate m ovem ents and the a c tiv i ties of m arket p a rtic ip a n ts .3 T h is syste m ensured that la rg e p e rs o n a l s e c to r s u r p lu s e s w e re tr a n s fe r r e d through banks to large c o rp o ra tio n s to prom ote high rates of d o m e stic ca p ita l fo rm a tio n .4 In th is highly regulated e n v iro n m e n t, the B ank of J a p a n ’s o p e ra tin g s tra te g y was d e s ig n e d to c o n tro l bank cre d it to the n o n fin a n cia l private sector. The m o n e ta ry policy co n tro l m e ch a n ism d u rin g the m id-1970s is sum m arized in C h a rt 1. D a y-to -d a y p o licy o p e ra tio n s 3For a detailed discussion of the structure Japanese financial system, see Robert A. Financial M arkets: D eficits, Dilemm a, an d MIT Press, 1986): and Yoshio Suzuki, ed., System (London: IFR Books, 1987). and evolution of the Feldman, Japanese D eregulation (C am bridge: The Japanese Financial ♦Generally, households were able to invest the ir savings only in bank deposits, and banks had few alternatives to lending these funds to corporations. N either the corpora te nor the banking sector had significant direct recourse to raising funds in open cap ital markets, which consequently remained undeveloped. Because interest rates were adm inistratively controlled and often held below m arketclearing levels, major corpora tions cou ld borrow cheaply, while sm aller firms and individ uals faced strin gent cre dit constraints. Chart 1 Japan’s Monetary Control Mechanism: Mid-1970s Instruments Intermediate Objectives Operating Targets Policy Goals Bank of Japan lending Interbank market operations Discount rate I Reserve requirements Window guidance Interest rate controls FRBNY Q uarterly Review/Autum n 1991 29 too k the form of inte rb a n k m arket a ctivitie s or Bank of Japan lending to banks. These o p e ra tio n s affected the rate at w hich banks accu m u la te d reserves during a m a in te n a n ce period as m easured by the reserve p ro g ress ra tio .5 B ecause le nding by the Bank of Japan m ade up a large co m p o n e n t of bank reserves and banks were a lm ost e xclu sive ly lim ite d to the interbank m arket as an a lte rn a tive source of fun d s, the response of in te rb a n k in te re st rates to chan g e s in bank reserves was strong and h ighly p re d icta b le .6 C h anges in in te rb a n k in te re st rates, in turn, in flu enced the q u a n tity of cre d it provided by banks. A d m in istra tive c o n tro ls on loan (and deposit) rate m ovem ents lim ite d banks’ a b ility to pass on in te rb a n k rate changes to th e ir c u s to m e rs .7 Thus, h ig h er in te rb a n k rates led banks to ration credit and, given the heavy d e p e ndence of the c o rp o ra te se cto r on bank lending, prom pted c u t backs in e xpend itures. The Bank of Japan actively used se ve ra l o th e r s u p p le m e n ta ry in s tru m e n ts , in c lu d in g q u a n tita tive lending lim its on in d ivid u a l banks (w indow guidance), d isco u nt rate ch a n g e s, and a d ju stm e nts in reserve re quirem ents, to secure a desired level of bank le ndin g, p a rtic u la rly in p e rio d s of tig h te n in g . F in a n c ia l lib e ra liz a tio n : 1974-89 E con om ic grow th slow ed m a rke d ly a fte r 1973 and was a cco m pa nied by a sharp d e clin e in the share of output devoted to investm ent. Net c o rp o ra te borrow ing as a share of GNP fell by m ore than half from its 7 percent average share over 1965-74 (Table 1). At the sam e tim e, the dem and fo r bo rrow ing by the p ublic se cto r more than d o ubled du ring the 1970s, and J a p a n ’s te n d e n cy to run p e rs is te n t cu rre n t acco u n t surp lu se s, in te rru p te d o nly by oil price sho cks, becam e more pronounced. T hese m a cro e co no m ic ch a n g e s d ra m a tica lly altered the flow of fu n d s in the Ja panese econom y, creating pressures that eroded the tig h t re strictio n s on fin a n cial 5lf banks fulfill their requirements along an average path, the reserve progress ratio increases by 3.3 pe rcentage points each day. The Bank of Japan adjusts aggregate reserves to determ ine this ratio and transm it actions to the interbank market. 6Banks and securities corpora tions exchange funds in two interbank markets: the call market, a short-term market analogous to the U.S. federal funds m arket; and the bill discount market, where com m ercial bills are rediscounted. Interest rates in the interbank market are theoretically free from control. Nevertheless, because money market brokers have until recently set interbank rates in close consultation with the Bank of Japan, the Bank has had con sid era ble short-term influence on interbank rates. 7H igher interbank interest rates were passed on to corporate borrow ers in the form of higher deposit-to-loan ratios and of increases in loan rates tied to the Bank of Japan's discount rate. A lthough these rate movements allowed policy to affect expenditure de cisio ns through financial price changes, they were less significa nt than the effects of credit rationing. http://fraser.stlouisfed.org/ 30 FRBNY Q uarterly Review/Autumn 1991 Federal Reserve Bank of St. Louis activity.8 In p a rticu la r, the large increase in g o v e rn m en t borrow ing was pivotal in the d e ve lo p m e n t of active s e c o n d a ry m a rke ts in se c u ritie s . D uring th e 1960s and early 1970s, in itia l issues of g o ve rn m en t b onds were bought by a syn d ica te of fin a n c ia l in s titu tio n s at p rice s fixed by the Bank of Japan. T h e se in itia l fixed prices, c o m b in e d w ith th e B a n k o f J a p a n ’s p r o m is e to repurchase the bonds, s ig n ific a n tly lim ite d the d e v e lo p m ent of s e c o n d a ry s e c u ritie s m a rk e ts .9 A fte r 1975, however, the large scale of g o ve rn m en t bond issues th reatened to u n d e rm in e m o n e ta ry co n tro l (b e ca u se of the Bank of Japan pro m ise to re p u rch ase ) and force d banks to raise the share of b onds in th e ir p o rtfo lio s at a tim e w hen a ttra ctive a lte rn a tiv e in ve s tm e n t o p p o rtu n i ties were becom ing a vailable. T h e se d e ve lo p m e n ts led to a nu m b e r of reform s lib e ra liz in g bond issue rates, rem oving re s tric tio n s on the sale of bonds in s e c o n d a ry m arkets, and exp a n d in g the b onds’ m a tu rity ra n g e .10 By 8Good discussions of Japanese financial reform through the m id-1980s can be found in the OECD E conom ic S urvey— Japan (Paris: OECD, 1984); Suzuki, The Japanese F inancial System: and Thomas F. C argill, "Japanese M onetary Policy, Flow of Funds, and Domestic Financial Lib e ra liza tio n ," Federal Reserve Bank of San Francisco E conom ic Review, Summer 1986, pp. 21-32. For analysis of the more recent liberalization process, see K. O sugi, "Japan's Experience of Financial D eregulation since 1984 in an International P erspective," BIS E conom ic Papers, no. 26, January 1990; Masaaki Nakao and Akinari Horii, "C hanges in the M onetary Control Techniques and Procedures by the Bank of Jap an," Bank of Japan Research and S tatistics D epartm ent, S pecial Paper no. 195, 1991; and Kumiharu Shigehara, "Japan's E xperience with the Use of Monetary Policy and the Process of Lib e ra liza tio n ," Bank of Japan M onetary and E conom ic Studies, vol. 9, no. 1 (M arch 1991). »The high com m issions prom ised to the synd ica te upon resale of the bonds after the holding period raised the effective interest rate earned by subscribers. 10For a more detailed discussio n of these issues, see Suzuki, The Japanese Financial System, or M ichael Dotsey, "Japanese M onetary Table 1 Net Lending by Sector (As a Percentage of Nom inal GNP) C orporate business Personal sector Public sector Rest of world Memo Real GNP growth 1965-74 Average 1975-84 Average 1985-90 Average -7 .1 9.4 -2 .6 -0 .7 -2 .9 10.3 - 7 .1 -0 .8 -4 .3 9.0 -1 .3 -2 .9 8.1 4 0 4.8 Sources: "Flow of Funds in Japan in 1990," Bank of Japan Research and S tatistics D epartm ent, S pecial Paper no. 204, July 1991; "Flow of Funds in Japan in 1989," Bank of Japan Research and S tatistics D epartm ent, S pecial Paper no. 191, A ugust 1990. the early 1980s, turnover in Japan’s secondary govern ment bond market had become the second largest in the world.11 The increased supply of government bonds also encouraged the development of short-term money mar kets. In the late 1960s, the gensaki market, involving repurchase transactions largely using government bonds, arose as a vehicle for nonbank short-term financing. Liquidity in this market was significantly boosted by the growth in government bond issuance, and by the mid-1970s, the gensaki market had become a major unregulated short-term money market for nonfi nancial corporations. The growth of the gensaki market made it difficult for the Bank of Japan to maintain deposit rate ceilings. Attracted by rising market interest rates in the late 1970s, corporations were shifting their bank deposits to gensaki assets. Pressure by banks led authorities, in May 1979, to permit banks to issue certificates of deposits (CDs).12 The emergence of freer domestic capital markets coincided with the loosening restrictions on interna tional capital transactions. Capital outflows were gradu ally liberalized to contain upward pressure on the yen after 1973 while capital inflows remained highly restricted throughout the decade. However, after the second oil price shock placed downward pressure on the yen, a more general relaxation of controls was implemented under the Foreign Exchange and Foreign Trade Control Law in December 1980.13 The rise in Japan’s global surpluses in the first half of the 1980s, particularly its bilateral surplus with the United States, placed increased international pressure on Japan to accelerate financial liberalization. In 1984, F o o tn o te 10 ( c o n tin u e d ) Policy, A C om parative A na ly s is ,” B a n k o f J a p a n M o n e ta ry E c o n o m ic S tu d ie s , vol. 4, no. 2 (1986). " A c c o rd in g to the O E C D E c o n o m ic S u rv e y -J a p a n , turnover in the J ap a n e se bond m arket reached 2 0 0 trillion yen in 1981, about oneq u arte r the size of the turnover in U .S. bond m arkets and almost three tim es the turnover in the U.K. bond m arket. 12S eco n d a ry m arket trading in C D s did not begin until May 1982. 13The lifting of these capital controls resulted in large increases in both inward and outward c ap ital flows. In addition, the lifting of controls on nonresident transactions in Jap a n e se money m arkets led to c onsid erably closer integration of Jap a n e se money m arkets with those in Europe and the U nited States. As a num ber of studies have shown, interest rates in Euroyen m arkets and in the dom estic gensaki m arket be c a m e virtually e q u a lized by 1982. For exam ple, see B ruce Kasm an and C harles Pigott, “Interest R ate D ivergences am ong the M ajor Industrial N ations," this Q u a rte rly R eview , Autumn 1988, pp. 2 8 -4 4 ; and Jeffrey Frankel, “International Financial Integration: Relations am ong Interest Rates, Exchange Rates, and M on etary In d ic a to rs ,” in In te rn a tio n a l F in a n c ia l In te g ra tio n a n d the C o n d u c t o f M o n e ta ry P o lic y , Federal Reserve Bank of New York, 1990. a package based on the findings of a special committee set up by the U.S. Treasury and Japanese Ministry of Finance was announced. Most notably, new measures reduced restrictions on Euroyen activities, including Japanese resident borrowing and bond issues by Jap anese and foreigners. In addition, limits on forward foreign exchange transactions and swap limit rules on Japanese banks were abolished. Subsequently, limits on the purchase of foreign securities by Japanese non bank institutional investors were lifted. The second half of the 1980s saw continued efforts to deregulate domestic markets. The liberalization of inter est rates on bank time deposits began in 1985 and is expected to be completed in 1993. Money market certifi cate deposits were introduced in 1985; restrictions on the minimum denomination, length of maturity, and amounts issued have been steadily relaxed on these accounts as well as on CDs and time deposits.14 The changing financial m arket environm ent Financial liberalization and the associated process of financial innovation have had far-reaching effects on Japan’s financial system. Many constraints on portfolio and expenditure choices have been removed, altering the tightly controlled flow of funds patterns that sup ported the m onetary control m echanism of the mid-1970s. Three changes have been particularly sig nificant in the evolution of the Bank of Japan’s operating strategy: First, the importance of bank loans as a source of funds has greatly declined. Second, the range of instruments used by banks to raise funds has expanded dramatically. Third, assets with market-deter mined prices now predominate in the portfolios of all sectors of the economy. We have seen that bank credit was employed as an intermediate target of policy in the mid-1970s largely because of its central role in channeling funds between lenders and borrowers. Before 1974, bank lending accounted for close to three-quarters of intermediated funds in Japanese markets (Table 2). In the second half of the 1970s, however, the importance of bank lending declined sharply as public sector bond issues increased and corporate sector capital spending growth slowed. Recent years have seen a further decline in the size of domestic loans in Japan’s flow of funds. The interna tionalization of Japan’s financial activities has combined with the corporate sector’s steady move towards securitization to reduce the share of domestic loans to less than half of all intermediated funds flowing through 14A num ber of actions have also been taken to prom ote d e e p e n in g of short-term m oney m arkets. A yen -d e n o m in ate d bankers’ a c c e p ta n c e m arket was launched in June 1985 and a com m ercial p a p e r m arket op en ed in 1987. In addition, a variety of short-term governm ent bond issues have been introduced, and m easures have b een taken to e xp and the maturity structure in the in terbank m arket. FRBNY Quarterly Review/Autumn 1991 31 J a p a n .15 At the sam e tim e that d o m e stic cre d it d e clin e d in im p o rta n ce, the Bank of J a p a n ’s control over bank le n d ing d e c is io n s w e a k e n e d . T h e g ra d u a l re m oval of re strictio n s on bank beh a vio r enabled banks to expand th e ir fu n d in g sources (both at hom e and abroad) and to ad ju st p rices of th e ir s e rv ic e s more independently. As a result, banks’ re lia nce on Bank of Japan credit d e clined significantly, along w ith the B a n k ’s leverage in using w in dow g u id an ce or o th e r a d m in istra tive co n tro ls to a ffe ct bank behavior. T he d e v e lo p m e n t of E uroyen and CD m a rke ts in recent years has been p a rtic u la rly im p o rta n t in this p rocess (Table 3). Both m a rkets, free from official con- trols, have expanded d ra m a tic a lly : E uroyen lia b ilitie s have grow n more th a n fo u rfo ld and o u ts ta n d in g CDs more than do u b le d since 1985. C urrently, th e y re p resent nearly half of J a p a n e se m oney m a rke ts and exceed the size of d o m e stic in te rb a n k m a rkets. The in cre a se d a v a ila b ility of m a rk e t-p ric e d a s s e ts exte n d s beyond th e fin a n c ia l sector. In v e s tm e n ts in in stru m e n ts w ith m a rk e t-d e te rm in e d in te re s t rates by the private n o n fin a n cia l se c to r have risen sig n ifica n tly, p a rtic u la rly since 1984, w hen bank d e p o s it rates began to be lib e ra liz e d (Table 4). The rising share of m a rk e t-p ric e d in s tru m e n ts in p o rt folios has u n d o u b te d ly in creased the im p o rta n c e of in te re s t ra te s in e x p e n d itu re d e c is io n s . M o re o ve r, p o te n tia l d is in te rm e d ia tio n betw een a d m in is te re d and m a r k e t-p r ic e d a s s e ts h a s w e a k e n e d th e B a n k o f J a p a n ’s a b ility to tra n s m it p o licy by a lte rin g spreads betw een in te rb a n k rates and (a d m in iste re d ) loan and 15C orporate issues of securities, which accounted for roughly 10 percent of the funds raised by the corporate sector before 1973, rose close to 15 percent over 1975-79, and in recent years have risen to more than a third of corporate fund raising. Table 2 Funds Interm ediation in Japan (Fiscal Year Average) Total funds s up plied (trillions of yen) C om position (p ercen tage of total) Funds raised by do m estic sectors Loans from do m estic banks Securities Governm ent bonds Foreign funds Funds s up plied to overseas m arket 1965-74 1975-84 1985-90 20.4 58.6 122.91 92.1 70.2 19.2 12.9 2.7 7.9 89.1 54.6 32.0 26.6 2.5 10.9 72 3 47.5 19.7 7.2 5.1 27.7 Sources: "Flow of Funds in Japan in 1990," Bank of Japan Research and S tatistics D epartm ent, S pecial Paper no. 204, July 1991; "Flow of Funds in Japan in 1989," Bank of Japan Research and S tatistics D epartm ent, S pecial Paper no. 191, A ugust 1990. Table 3 Major Japanese Money Markets (Trillions of Yen, End of Period Data) Interbank market D om estic interbank yen lia b ilities Euroyen inte rban k lia b ilitie s Open m arkets Bond gensaki CDs C om m ercial paper Total Memo D om estic interbank m arket as a share of total money m arkets (p ercen tage points) 32 FRBNY Q uarterly Review/Autumn 1991 1975 1980 1985 1989 6.7 _ 9.8 2.5 19 8 9.9 45.3 41.8 1.8 — — 4.5 2.4 — 4.6 9.7 — 6.3 21.1 13.1 8.5 19.2 44.0 127.6 78.9 51.0 45.0 35.5 d e p o sit rates. C om paring in te rb a n k interest rates w ith tw o ra te s s u b je c t to a d m in is tra tiv e c o n tr o l— tim e d e p o sit and loan ra te s — du ring three episodes of m o n e ta ry tig h te n in g p ro vides evid e nce of the reduced im por tance of this chan nel (C h a rt 2). In both 1973-74 and 1979-80, w ide d iffe re n tia ls opened betw een o vernight call rates and adm in iste re d loan and d eposit rates w hen p o licy tig h te n e d . However, in 1990, the m ost recent e p isode of tig h te n in g , spreads betw een th e se rates rem ained roug hly unchanged. R ecent s tru c tu re o f the m on e ta ry c o n tro l m ech an ism In response to these deve lop m e n ts, the Bank of Japan has g ra d u a lly m oved away from a control m echanism aim ed at re gulating the q u a n tity of bank credit. Instead, it has in cre a sin g ly sou gh t to a ffe ct expenditure d e c i s ions thro ugh o p e ra tio n s de sig ne d to affect m arket interest rates. The current policy control mechanism, outlined in Chart 3, shows a dramatic change from the mid-1970s. On the level of po licy instru m e n ts, the s h ift away from bank cre d it is reflected in the e lim in a tio n of co n tro ls that d ire ctly affe cte d banks’ a b ilitie s to extend credit. In p a rticular, w indow guidan ce , in the form of Bank of Japan in s tru ctio n s to in d ivid u a l banks regarding lending plans, was ended in 1982, and at about the sam e tim e, the active use of reserve requirem ents as a p o licy tool was d ro p p e d .16 As show n earlier, the use of in te re st rate i«A more lim ited form of w indow guidance, in which the Bank of Japan clarifie d its po licy orientation and discussed aggregate co n tro ls as a m eans of ra tio n in g cre d it has also slo w ly d im in is h e d , p a r t ic u la r ly fo llo w in g th e m a jo r p u s h to w a rd s d e re g u la tin g b a n k lo a n a nd d e p o s it ra te s begun in 1985. The Bank of Japan has replaced th e se in s tru m e n ts w ith a c tiv itie s o u ts id e th e in te rb a n k m a rk e t. It has u n d e rta ke n o p e ra tio n s in s h o rt-te rm g o ve rn m e n t bills (1981), CDs (1986), g e n sa ki (1987), and c o m m e rcia l p aper (1989). A lth o u g h o p e ra tio n s o u ts id e the in te rb a n k m arket have increased in fre q u e n cy in re ce n t years, the Bank of Japan co n tin u e s to rely la rg e ly on its lending p o licie s and o p e ra tio n s in in te rb a n k m a rke ts to a lte r reserves. A long w ith the reserve p rogress ratio, in te rb a n k in te r est rates rem ain the p rim a ry o p e ra tin g ta rg e t of the Bank of Japan. S ig n ific a n t ste p s have been ta ke n , how ever, to link in te rb a n k and o th e r m oney m a rke ts m ore closely, a d e ve lop m e n t th a t reflects the g re a te r im p o r ta n ce placed on fin a n c ia l p rice s in the m o n e ta ry co n tro l m echanism . In 1979, the B ank acted to a llo w in te rb a n k rates to adjust more ra p id ly to open m a rke t c o n d itio n s, and in s u b se q u e n t years, it co n tin u e d to reform its procedures for in te rv e n in g in in te rb a n k m a rke ts. W hen the Bank becam e co n ce rn e d th a t a ctio n s ta ke n to low er in te rb a n k in te re st rates d u rin g 1987-88 were not being tra n s m itte d to m oney m a rke ts, it im p le m e n te d a m a jo r Footnote 16 (continued) lending plans with individ ual banks, con tinued after 1982 and was finally abolished in 1991. Table 4 Financial Investments of the Domestic Nonfinancial Sector Total investm ents (trillio n s of yen) C om position (p ercen tage of total) A ssets w ith m arket-determ ined interest rates Bank deposits* Trust and insurance deposits D om estic securities Foreign credits Assets w ith regulated interest rates* 1975-79 Average 1984 1988 1990 43.9 62.5 106.0 113.8 30.0 — 15.8 12.6 1.6 70.0 50.7 4.2 23.2 13.4 9.9 49.3 86.0 50.7 32.8 - 5 .1 7.6 14.0 147.1 94.7 25.7 16.9 9.8 - 4 7 .1 Memo March 1984 Bank lia b ilities with m arket-determ ined interest rates (share of total liabilities) 13.5 S eptem ber 1989 50.3 Sources: “ Flow of Funds in Japan in 1990," Bank of Japan Research and S tatistics D epartm ent, S pecial Paper no. 204, July 1991; “ Flow of Funds in Japan in 1989,” Bank of Japan Research and S tatistics Departm ent, S pecial Paper no. 191, A ugust 1990. in c lu d e s unregulated tim e de posits, certific a te s of deposit, and money market certificates. in c lu d e s currency, dem and deposits, regulated tim e de posits, postal savings deposits, and trust fund bureau de posits. FRBNY Q uarterly Review/Autum n 1991 33 set of in te rb a n k m arke t reform s in N ovem ber 1988.17 T he se reform s involved s h iftin g in te rb a n k o p e ra tio n s to 17For de tails on the evolution of Bank of Japan operations in interbank m arkets, see Toshihiko Fukui, "Recent Developments of the Short-Term Money Market in Ja p a n ,” Bank of Japan Research and S tatistics D epartm ent, S pecial Paper no. 130, January 1986; and Nakao and Horii, "C hanges in the M onetary Control Te chniques.” Chart 2 Interest Rate Movements during Periods of Monetary Tightening Percent 14 1973-74 Call rate Loan rate Time deposit rate Percent 14 Percent 12 Source: Bank of Japan, E co n om ic Statistics M onthly. 34 FRBNY Q uarterly Review/Autumn 1991 s h o rte r m a tu ritie s and re p la cin g th e q u o ta tio n s yste m in the in te rb a n k m arket by an o ffe r-b id syste m to p rom ote g reater a rb itra g e betw een m arkets. The ch ange in J a p a n ’s m o n e ta ry co n tro l m e th ods over the past fifte e n years is m ost e v id e n t in the use of fin a n cial va ria b le s in the in te rm e d ia te sta g e of the p o l icy process. As e a rly as 1975, the B ank of Japan began its s h ift away from bank cre d it as an in te rm e d ia te ta rg e t and increased its e m p h a sis on the role of broad m oney in its p olicy o p e ra tio n s. In a sense, cre d it and m oney ta rg e ts had been e q u iv a le n t up u n til th is tim e b e cause of th e ir close re la tio n sh ip on bank b a la n ce sh e e ts. But w ith the la rg e -sca le flo ta tio n of g o ve rn m e n t bonds, s u b s ta n tia lly u n d e rw ritte n by banks, th e c h a n n e ls of m oney creation were no lo n g e r lim ite d to in cre a se s in le nd ing. M oney thus becam e view ed a b e tte r in d ic a to r of levels of aggregate e xp e n d itu re and assum ed a le a d in g role in the m o n e ta ry co n tro l m e ch a n ism . The Bank p ro b a b ly never a c tive ly e m ployed broad m oney (M2 + C Ds) as an in te rm e d ia te ta rg e t, however. Instead, broad m oney becam e the p rim a ry in d ic a to r am ong a group of fin a n c ia l v a ria b le s th a t p ro vid e d in fo r m ation on a c tiv ity and the s ta n ce of p o licy.18 Indeed, the Bank of Japan re fra in e d from se ttin g e x p lic it ta rg e ts for broad m oney and in ste a d chose to p u b lish q u a rte rly forecasts fo r M2 + C D s from 1978 onw ard. By the m id-1980s, fin a n c ia l lib e ra liz a tio n had begun to blur the b o u n d a rie s of s p e cific fin a n c ia l a sse ts in Japan. The w e a lth -h o ld in g p ro p e rtie s of b ank lia b ilitie s in the form of CDs or d e re g u la te d tim e d e p o s its were enhanced, w hile th e liq u id ity c h a ra c te ris tic s of s e c u ri ties packaged in the form of tru s t and in s u ra n c e fund accounts increased. T he rem oval of c o n tro ls on in te rn a tio n a l c a p ita l m o v e m e n ts fu r th e r e d th e s e tre n d s because in ve sto rs were able to tre a t a sse ts issue d in Japan or in foreign m a rke ts m ore in te rch a n g e a b ly. In recent years, the B ank of Japan has re sp o n de d to these d e ve lop m e n ts by g ra d u a lly reducing its e m ph asis on broad m oney in im p le m e n tin g policy. The d im in ish e d im p o rta n ce of broad m oney was h ig h lig h te d in 1987 w hen M2 + CDs grew above B ank of Japan fo re c a sts for three co n se cu tive q u a rte rs w ith o u t p rovoking a p o licy response. A lthough several v a ria b le s, in c lu d in g e xch a n g e rates and asset prices, have been em ployed along w ith broad m oney as key in te rm e d ia te v a ria b le s o ve r th e pa st decade, m arket in te re st rates have be co m e in c re a sin g ly 18For studies sup portin g this view, see M ichael M. H utchinson, "Japan's ’Money Focused’ M onetary Policy," Federal Reserve Bank of San Francisco Econom ic Review, Summ er 1986, pp. 33-46; Koichi H am ada and Fumio Hayashi, “ M onetary Policy in Postwar Jap an,” in A lb e rt Ando, Hidekazu Eguchi, Roger Farmer, and Yoshio Suzuki, eds., M onetary P olicy in Our Times (C am brid ge; MIT Press, 1985); and Shigehara, “ Ja p a n ’s E xperience w ith Use of M onetary Policy.” va rie ty of fa cto rs, a p o s s ib ility th a t can u n d e rcu t the a b ility of m o n e ta ry a u th o ritie s to in flu e n c e in te re s t rates in a pre d icta b le way. For s im ila r re a so n s, th e u s e fu ln e s s of m o n e ta ry aggregates in the m o n e ta ry c o n tro l m e ch a n ism may now be lim ited. The role of M2 + C D s as an in d ic a to r of a c tiv ity was la rg e ly tied to re s tric tio n s th a t m ade it the p rin cip a l m eans of liq u id ity in the J a p a n e se econom y. In this e n viro n m e n t, M2 + CDs te n d e d to re fle ct a c tiv ity fa irly c lo s e ly . A s th e w e a lth -h o ld in g p ro p e r tie s of M2 + CDs and liq u id ity c h a ra c te ris tic s of o th e r fin a n c ial assets have increased, all fin a n c ia l a g g re g a te s may h ave b e c o m e le s s a c c u ra te in d ic a to r s o f a c tiv ity because th e ir s h o rt-ru n b e h a vio r has be co m e se n sitive to m ovem ents in relative rates of return. Moreover, as the lines betw een fin a n c ia l a sse ts have becom e blurred, p olicy ch a n g e s a ffe ctin g g e n e ra l e co n o m ic c o n d itio n s or interest rates may have a w e a ke r link to any sp e cific a ggregate, p a rtic u la rly in the s h o rt term . In this se ctio n , we assess the B ank of J a p a n ’s a b ility to tra n s m it p o licie s to fin a n c ia l m a rke t v a ria b le s in the current lib e ra liz e d fin a n c ia l m a rke t e n v iro n m e n t. S p e c if ically, we exam ine the re la tio n s h ip betw een m ovem ents in in te rb a n k in te re st rates, the B a n k ’s m ain o p e ra tin g ta rg e t, and the in te rm e d ia te va ria b le s view ed as key in the m o n e ta ry co n tro l m e ch a n ism : m a rke t in te re s t rates and broad m oney M2 + CDs. O u r e a r lie r d is c u s s io n s u g g e s te d th a t by th e m id-1980s fin a n cial lib e ra liz a tio n had caused s ig n ific a n t changes in the fu n c tio n in g of Ja p a n e se fin a n c ia l m ar kets. Thus we fo cu s o u r a tte n tio n on the post-1984 e x p e rie n c e to a s s e s s how c lo s e ly B a n k o f J a p a n a ctions are being tra n s m itte d to in te re s t rates and m o n e ta ry aggregates in a lib e ra liz e d e n v iro n m e n t. In a d d i tion, this period is c o n tra s te d w ith 1974-84 to p rovide in sig h t into c h a n g e s in th e se re la tio n s h ip s over tim e. ce n tra l fo r p o licy o p e ra tio n s. M arket interest rates are an in d ic a to r of e co nom ic c o n d itio n s and help to tra n sm it in te rb a n k rate m ovem ents to real activity. C oncerns that a ction s in in te rb a n k m arkets were not stro n g ly c o n n ected to o th e r open m arket rates prom pted reform s in the in te rb a n k m arket in 1988. Moreover, in 1989 and 1990, the Bank of Japan c o n s is te n tly cited the rising level of m arket inte re st rates as a m otivation fo r tig h te n ing m o n e ta ry policy.19 M onetary control of interest rates and broad money: empirical evidence We have seen that fin a n cia l lib e ra liz a tio n has led to s ig n ifica n t ch ange s in the Bank of Ja p a n ’s op e ra tin g strategy. M a rke t-d e te rm in e d fin a n c ia l prices, interest rates in p articula r, play a more im p o rta n t role as both a ta rg e t and an in d ica tor of policy. In co ntrast, attem pts to c o n tro l bank credit or oth e r fin a n cial aggregates have g ra d u a lly d im in ished. W hile fin a n cial m arket changes have increased the a tte n tio n given to interest rates in policy fo rm u la tio n, they may also have m ade in te rp re ta tio n and control of in terest rates more com plex. In the 1970s, m arket s e g m e n ta tio n an d r e s tr ic tio n s on p o r tfo lio a c tiv itie s ensured th at ce ntral bank actions w ould result in a p re d icta b le pattern of su b s titu tio n betw een the in te r bank m arket and the ge nsaki m oney m arket. In the cu rre nt en viro n m e n t, ag en ts have a greater choice of m oney m a rket in stru m e n ts (both do m e stic and foreign) and can m ore ea sily move betw een these in stru m e n ts and long -te rm se cu ritie s. T hese linkages may produce a c lo se r co n n e ctio n am ong in te re st rates. N evertheless, they allow in te re st rates to be influenced by a w ider 19See Nakao and Horii, "C hanges in the M onetary Control Techniques," for a more de tailed discussion of the factors determ ining m onetary p o licy changes during the 1980s. Chart 3 Japan’s Monetary Control Mechanism: Late 1980s Instruments Operating Targets Bank of Japan lending Interbank market operations Open market operations — ► Bank reserves Interbank interest rates Intermediate Objectives — ► Market interest rates Broad money -------► Policy Goals GNP Inflation Discount rate FRBNY Q uarterly Review/Autum n 1991 35 E c o n o m e tric analysis O ur e co n o m e tric a na lysis, e xplained in gre a te r de tail in the ap p e n d ix, attem pts to in te g ra te sh o rt-te rm re la tio n ships am ong Japan e se in te re st rates and m o n e ta ry a ggregates w ith m odels governing th e ir lo n g er term behavior. T his e m p irica l s tra te g y is m otivated by the te n d e n cy of all the va ria b le s analyzed to d rift over tim e w ith o u t co nverging tow ard a u nique long-term level. A lth o u g h interest rates and m o n e ta ry aggregates may d rift, eco n o m ic th e o ry su g g e sts that com m on u n d e rly ing fa cto rs may d e te rm in e th e ir m ovem ent. The e xp e c ta tio n s th e o r y of th e te rm s tru c tu re , fo r e x a m p le , su g g e sts tha t long-te rm in te re st rates ap p ro xim a te ly equal an average of cu rre n t and expected future s h o rt term rates. Thus, if s h o rt-te rm interest rate changes are p e rsiste n t (because of p e rm a n e n t changes in in fla tio n a ry e xp e cta tio n s or real in te re st rates), th e se changes should be reflected across the term structure. S im ilarly, com m on fa cto rs m ay explain the evolution of broad m oney and the m o n e ta ry base. Interest rates may also be an im p o rta n t p a rt of th is re la tio n sh ip if the so u rce s of p e rs is te n t in te re s t rate c h a n g e s have a s y s te m a tic effe ct on broad m oney in d e pe n d e n t of changes in the m o n e ta ry base. T he firs t p a rt of o u r a n a ly s is se a rc h e s fo r links betw een Ja pane se in te re st rates or m o n e ta ry a g g re g ates and in te rb a n k in te re st rates. The resulting “ cointe g ra tin g ” regressio ns d e s c rib in g these links capture the lo n g -te rm response of va ria b le s to th o se p e rsiste n t ch a n g e s in m o n e ta ry p o lic y in d ic a te d by s u s ta in e d ch anges in the in te rb a n k call rate. U nfortunately, th e re g re ssio n s do not p ro vid e in fo rm a tion on th e se lin ka g e s at a h o rizon relevant to the w o rkin g s of m o n e ta ry policy. T hus, th e se co n d p a rt of the a n a lysis de ve lop s a m odel of the d y n a m ic response of in te re st rates and m o n e ta ry a g g re g a te s to ch a n g e s in th e c a ll ra te c o n s is te n t w ith th e s e c o in te g r a tin g re lationships. B efore tu rn in g to o u r s ta tis tic a l a n a ly s is , we present in Table 5 som e d e s c rip tiv e e vid e n ce on in te re s t rates, broad money, and e co n o m ic a ctiv ity .20 S p e cifica lly, the ta b le show s m ean levels of m o n th ly in te re s t ra te s and tw e lve -m o n th ch a n g e s in M2 + CDs, co n s u m e r price s, and in d u s tria l p ro d u ctio n , along w ith s ta n d a rd d e v ia tions fo r the p e rio d s 1975-84 and 1985-90. S u b sta n tia l d e c lin e s in both th e level and the v a ri a b ility of m oney m a rke t in te re s t ra te s are a p p a re n t since 1984. O ver 1985-90 m oney m a rke t in te re s t rates averaged roughly 5 to 51/2 p e rce n t, 200 b asis p oints below th e ir 1975-84 levels. T h e ir v a ria b ility, as m e a sured by the sta n d a rd d e v ia tio n , d e c lin e d by ro u g h ly o n e -th ird. 2°The data in this section are drawn from various issues of Bank of Japan, Econom ic S tatistics M onthly, and includ e the unconditio nal call rate (month end), the bond re purch ase— or g e n sa ki— rate (month end), the benchm ark governm ent bond rate (m onth end), the bank certificate of de posit rate (80-179 days, month average), the average rate on short-term bank loans (m onth end), the rate on one-year time deposits (month end), the m onetary base (month end, seasonally adjusted by the authors), and M2 + CDs (seasonally adjusted, month end). Table 5 Descriptive Statistics for Economic Activity and Interest Rates (Period Averages) 1975-84 Mean 1985-90 Standard Deviation Mean Standard Deviation Money market rates Interbank call rates G ensaki rates C ertifica te of deposit rates1 7.1 7.2 7.6 2.2 1.9 17 5.1 5.2 5.6 1.5 1.3 1.3 R egulated rates S hort-term loans S hort-term time deposits 6.8 4.1 1.2 09 5.1 2.5 1.1 0.8 Long-term governm ent bonds 7.9 11 5.4 1.2 10.5 26 10.0 1.6 3.7 5.6 6.6 3.5 4.5 1.4 Growth in broad money (M2 + C D s )T Econom ic activity Ind ustria l production growth* C onsum er price inflation* ■''Because c e rtific a te s of d e posit were not introduced until 1979, the values in the first two colum ns are averages for 1979-84. ♦Twelve-month percentage changes. 36 FRBNY Q uarterly Review/Autumn 1991 3.7 1.2 v : ' • T h e se d e c lin e s are c o n s is te n t w ith o v e ra ll m a c ro e conom ic d e velopm ents. Ja p a n e se c o n su m e r price in flation averaged less than V/2 perce n t over 1985-90, a drop of more than 4 perce n ta g e points from its 1975-84 average. Moreover, a sharp fall in the v a ria b ility of c o n s u m e r p rice in fla tio n , b road m oney g ro w th , and in d u stria l pro d u ctio n grow th since 1985 su g g e sts that e co no m ic a c tiv ity has becom e co n sid e ra b ly more stable in recent years. Interest rates on oth e r fin a n cia l assets have also d e clined but, in co n tra st to m oney m arket rates, exhibit no s ig n ifica n t change in th e ir variability. From 1975 to 1984, loan and d e p o sit rates as well as long-term bond yie ld s were c o n s id e ra b ly less va ria b le than m oney m ar ket rates. The lower va ria b ility in these rates probably reflected re strictio n s lim itin g th e ir responsiveness to m arket c o n d ition s. A lthou g h fin a n cia l lib e ra liz a tio n has u n d o u b te d ly allow ed these rates to a djust to changing m arket co n d ition s, overall e co n o m ic co n d itio n s appear to have becom e more stab le . As a result, the effe cts of lib e ra liza tio n are seen not in the increased v a ria b ility of th e se interest rates but rather in a co n ve rg e n ce in in te re st rate va ria b ility th ro u g h o u t the econom y. L o n g -te rm re la tio n s h ip s The results of three te sts fo r com m on trends, or cointe g ra tin g re la tion s, betw een the o vernight call m oney in te rest rate and various oth e r interest rates are pre sented in Table 6 along w ith p a ra m e te r e stim a te s for sp e cific e quations. T hese te sts d e te ct w h e th e r a single u n derlying fa cto r e xp lains the d rift in the regression va riab les (see appendix). O ve ra ll, th is e vid e n c e in d ic a te s th a t the lin ka g e s b e tw ee n th e B ank of J a p a n ’s o p e ra tin g ta rg e t and m oney m arket interest rates have been q uite strong th ro u g h o u t the 1974-91 period. The call rate appears to have been co in te g ra te d w ith both the g ensaki rate and the CD rate d uring 1974-84 and 1985-91. D uring the e a rlie r period, a 100 basis p oint increase in the call rate was associate d w ith a ro ughly equal change in the g ensaki rate. In the later period, the response of the g e nsaki rate was som ew hat sm aller, e stim ated at 82 basis points. In co n tra st, n e ith e r loan rates nor long-term g o ve rn m ent bond yields were co in te g ra te d w ith the call rate betw een 1974 and 1984. T his result is co n s is te n t w ith our e a rlie r c o n te n tio n tha t a d m in istra tive co n tro ls on loan rates and re strictio n s on the d evelopm ent of a se c o n d a ry m arket in long-te rm governm ent bonds may have p a rtia lly isolated th e se m arkets from in te rb a n k and s h o rt-te rm m oney m arkets. F in a n c ia l lib e ra liz a tio n d oes a p p e a r to have in te grated lo ng-term bond m arkets and m oney m arkets. B etw een 1974 and 1984 the call rate did not have a c o n s is te n t lo n g -ru n c o n n e c tio n to g o v e rn m e n t bond yields. A fte r 1984, however, stro n g e v id e n c e of a cointe g ra tin g relation betw een lo n g -te rm bond y ie ld s and call rates em e rg e s: lo n g -te rm bonds in cre a se by 69 basis p oints in response to a 100 basis p o in t rise in the call rate. There is little e vid e nce , however, th a t fin a n c ia l reform has tig h tly in te g ra te d m oney m a rke ts w ith loan m a rkets. A lthough the loan rate reacted m ore s tro n g ly to call rate changes a fte r 1984, bank loan rates were not c o in te grated w ith the call rate b etw een 1985 and 1991, s u g gesting th a t there has not been a c o n s is te n t lo n g -te rm relation betw een the rates. Table 6 Cointegration Relationships for Monthly Interest Rates Response to Call Rate of Gensaki R2 C ointegration tests ADF SW PP CD* R2 C ointegration tests ADF SW PP Long-term bond R2 C ointegration tests ADF SW PP Loan rate R2 C ointegration tests ADF SW PP 1974-84 1985-91* 1.00 .84 .82 .97 - 3.28* - 5 1 .6 7 * “ -3 2 .3 7 ’ “ .90 .97 -2 .9 7 - 2 9 .7 5 * ” -2 5 .2 2 * * .34 .65 -1 .7 8 - 1 0 .8 2 - 1 2 .4 2 -2 .7 1 -3 3 .9 0 * * * - 4 4 .2 0 * * * .85 .96 - 2 .4 2 -3 5 .2 6 * * * -4 3 .3 3 “ * .69 .83 - 3 .3 3 * -2 3 .7 7 “ * -2 0 .7 9 “ .49 .75 .73 .88 - 2 .4 0 -7 .8 7 - 5 .9 8 - 1 .3 3 - 9 .6 2 - 1 1 .1 4 Notes: ADF is the augm ented D ickey-Fuller sta tistic (using seven lags). PP is the Phillips-Perron Z„ sta tistic (using seven autovariance lags). SW is the Stock-W atson sta tistic (using seven lags). C ritical values for the D ickey-Fuller and P hillipsPerron statistics were o b tained from P.C.B. P hillips and Sam O uiliaris, "A sym ptotic P roperties of R esidual-B ased Tests for C ointe gra tion,” E co n o m e trica , vol. 58, no. 1 (January 1990), pp. 165-91 C ritical values for the Stock and Watson statistic are from James Stock and Mark Watson, “ Testing for Common Trends," Journal of the A m erican S tatistical A ssociation, vol. 83 (D ecem ber 1988), pp. 1097-1107. *Sample covers January 1985-May 1991, ^Earlier sam ple covers 1980-84. ‘ S ignificant at 10 pe rcent level. ‘ ‘ S ignificant at 5 pe rcent level. “ ‘ S ignificant at 1 pe rcent level. FRBNY Q uarterly Review/Autum n 1991 37 O ur m odel c o n n e c tin g m o n e ta ry p o licy actions to broad m oney is based on a standard view of the money su p p ly p ro c e s s .21 Bank of Japan o p e ra tio n s in in te rb a n k m arke ts are a ccom p an ie d by changes in both reserves available to the banking system and in te rb a n k interest rates. G iven un changed asset a llo ca tio n s by banks and d e p o sito rs, ch ange s in reserves can be expected to a lte r broad m oney (M2 + C Ds) in a p redictable manner. Interest rate m ovem ents can a lte r p o rtfo lio choices, th u s in flu e ncin g the m oney supply in d e p e n d e n tly of ch ange s in the m o n e ta ry base. H igher m arket interest rates, all else equal, raise the cost of holding bank reserves and co n se q u e n tly may increase the m oney m u ltip lie r (the ratio of broad m oney to the m o n eta ry base). At the sam e tim e, an increase in ce n tra l bank lending rates or in te rb a n k rates relative to m arket rates could increase dem and fo r reserves and thus low er the 21Money dem and is also im portant in money stock determ ination. Because we do not e x p licitly m odel money demand, our analysis should not be viewed as a full behavioral model for the determ ination of interest rates and the money stock. Table 7 Cointegrating Money Models Response of M2 + CDs to: In (base) Trend R2 C ointegration tests ADF SW PP 1974-84 1985-91f .554 .005 .993 -2 .0 9 1 -8 .1 7 4 -6 .2 1 2 .412 .005 .998 -3 .1 3 8 -3 6 .4 6 9 ** * -4 5 .2 2 7 *** Response of M2 + CDs to In (base) Gensaki rate Call rate Trend R2 Cointegration tests ADF SW PP 1.000* .009 -.0 1 9 .002 .997 1.000* .007 -.0 0 4 .000 .993 -2 .3 4 2 -3 9 .7 0 4 *** -4 9 .0 8 1 *** -2 .3 2 4 -4 5 .7 3 8 * ’ * -5 6 .6 5 0 *** Notes: R2 is the square of the correlation coefficient between actual and predicted In (M2) for the two regressions with base coe fficients equal to one. ADF is the augm ented Dickey-Fuller sta tis tic s (using seven lags). PP is the Phiilips-Perron Z„ statistic (using seven autovariance lags). SW is the StockWatson s ta tistic (using seven lags). C ritical values for the D ickey-Fuller and P hiilips-Perron statistics were ob tained from P hillips and O uiliaris, “A sym ptotic Properties of ResidualBased Tests." C ritical values for the Stock and Watson statistic are from Stock and Watson, “ Testing for Common Trends.” tS am ple covers January 1985-May 1991. ♦C onstrained to equal 1. ’ S ignificant at 10 pe rcent level. ’ •S ign ificant at 5 pe rcent level. “ ‘ S ignifican t at 1 pe rcent level. 38 FRBNY Q uarterly Review/Autumn 1991 m oney m ultiplier. We su g g e ste d e a rlie r th a t B ank of Japan a ctio n s a sso cia te d w ith h ig h e r in te rb a n k in te re st ra te s , in c lu d in g w in d o w g u id a n c e a n d c h a n g e s in reserve req u ire m e n ts, m ay have re in fo rce d a d e c lin e in the m oney m u ltip lie r in the p a s t .22 N on e th e le ss, in te re st rate e ffe cts on the m oney m u lti p lie r m ight be tra n sito ry, p a rtic u la rly b e ca u se re g u la to ry re stra in t on bank b e h a vio r was a p p lie d o n ly te m p o ra rily. T h u s , we fir s t m o d e l th e lo n g e r te rm b e h a v io r of M2 + CDs as a fu n c tio n of the m o n e ta ry base alon e, in clu d in g a tim e trend term to a llow fo r te c h n o lo g ic a l fa cto rs th a t may have a ltered the m oney m u ltip lie r over tim e. The co in te g ra tin g re g re ssio n s fo r th is m odel, p re sented in the u pper ha lf of Table 7, p ro vid e no e vid e nce of a c o in te g ra tin g re la tio n b e tw e e n th e b a s e a nd M2 + CDs before 1985. T h is result su g g e s ts th a t fa cto rs le ading to p e rs iste n t m ovem ents in th e m oney m u ltip lie r were an im p o rta n t d e te rm in a n t of th e lo n g -te rm b e h a v ior of b road m o n e y d u rin g th is p e rio d . In c o n tra s t, betw een 1985 and 1991, e vid e n ce of a c o in te g ra tin g regression is present, and th u s m oney base ch a nges, th rough a sta b le m u ltip lie r, a d e q u a te ly e xp la in th e lo n g term evolution of M2 + CDs. To assess w h e th e r the p e rs is te n t m ove m e n t in the m oney m u ltip lie r before 1985 was a sso cia te d w ith in te r est rate m ovem ents, we add the call and g e n sa ki rate to our regression m odel. In th is c o n te xt, the g e n sa ki rate captures a lte rn a tive bank in ve stm e n t o p p o rtu n itie s th at becam e available b e g inn in g in the se co n d h a lf of the 1970s. C a ll ra te m o v e m e n ts m e a s u re th e c o s t of re se rve s h o rtfa lls to b a n ks and a lso p ro x y fo r the effe cts of policy a ctio n s not related to ch a n g e s in the m o n e ta ry base. The co e ffic ie n t on the m o n e ta ry base is re stricte d to equal one in th is fra m e w o rk b e ca u se, by d e fin ition , base c h a n g e s are fu lly reflected in th e m oney su p p ly w hen the m u ltip lie r is u n ch a n g e d . Including in te re st rates in th e m odel y ie ld s strong evidence of c o in te g ra tio n fo r the 1974-84 p e rio d . M o re over, the p a ra m e te r e stim a te s are of th e c o rre c t sign and su g g e st a large e ffe c t of call rate ch a n g e s on broad money. S pecifically, w hen m a rke t ra te s and th e m o n e ta ry base are held c o n sta n t, a 100 basis p o in t increase in the call rate is a sso cia te d w ith a p e rm a n e n t d e clin e of nearly 2 p e rce n ta g e p o in ts in M2 + C Ds. N evertheless, th e s e call rate e ffe c ts d e c lin e s u b s ta n tia lly after 1984. Indeed, th e sm all size of the in te re st 22ln particular, Japanese banks were forced to constrain lending activities when po licy tigh tene d because loan rates were regulated and the Bank of Japan im posed qu antita tive restrictions on lending. These restrictions, tog ethe r with the B ank’s active use of reserve requirement changes to im plem ent m onetary policy, forced banks to increase their reserve-deposit ratios as po licy tigh tene d, an o u t com e that lowered the money m u ltip lier and broad money for a given m onetary base. rate c o e ffic ie n ts in the m oney supply e q u a tio n s su g g e sts th a t in te re st rate m ovem ents, in d e pe n d e n t of the m o n e ta ry base, may no lo n g er have any lasting effect on broad money. D yn a m ic re la tio n s The evid e nce of long-term lin ka g e s betw een call rates and o th e r fin a n cial va ria b le s does not, by itself, cla rify how m o n e ta ry p olicy cha n g e s are tra n sm itte d over a h o rizo n re le va n t fo r p o lic y m a ke rs. To a d d re ss th is q u e stio n , we e stim ate a d yn a m ic erro r-co rre ctio n m odel fo r Jap a n e se fin a n cial m a rke t variables. The m odel, p resented in detail in the a p p e n d ix, captures in a fa irly u n re stricte d way the o bse rve d tim e series re la tio n sh ip s am ong th e se va ria b le s by im posing the co n d ition that the d yn a m ic b eh avior co n ve rg e to the lo n g -te rm cointe g ra tin g regressio ns estim a te d above. A ssum ing th a t the Bank of Japan has c o n sid era b le control over the in te rb a n k call rate, we use th e m odel to assess the m o n e ta ry co n tro l m e ch a n ism by co m p a rin g the responses of in te re s t rates and broad m oney to an in itia l 100 basis p o in t increase in the call ra te .23 T hese responses, presented in C h a rts 4 and 5, can be in te r preted as the average re sp o n se of in te re s t ra te s and broad m oney to p o licy ch a n g e s over the sa m p le . The s im u la tio n s also tra ck s u b s e q u e n t ca ll rate m ovem ents g e n erated by the m odel; th e se m o ve m e n ts ca p tu re the te n d e n cy of Bank of Japan p o lic y s h ifts to o c c u r g ra d u ally as well as the ty p ic a l re sp o n se of call rates to c hanges in o th e r fin a n c ia l v a ria b le s. The evidence in C h a rt 4 su g g e s ts s o m e w h a t s tro n g e r tra n sm issio n of call rate sh o cks to in te re s t ra te s a fter ^A lth o u g h the results of this analysis are presented sep arately for interest rates and m onetary aggregates, they are derived from a single model that accounts for the interrelationship am ong these variables. Chart 4 Interest Rate Responses to a Call Rate Increase Percentage point change 2.0 ---------------------------------------------------------------------------------------1974-84 Percentage point change 2.0----------------------------- ___________ 1985-91 1 .5 - 1974-84 1 . 0 0 \ --------------Ratio of Gensaki to \ call rate changes 0 .7 5 - Months after shock Note: Chart shows the predicted response to an increase of 100 basis points in the call rate. FRBNY Q uarterly Review/Autum n 1991 39 1984, p rim a rily because of the increased re sp o n sive ness of lon g -te rm bond yie ld s. From 1975 to 1984 only the gen sa ki rate responded stro n g ly and im m e d ia te ly: a 100 basis point increase in the call rate prom pted an im m e d ia te and equal rise in the gensaki rate. In c o n tra st, th e im m ed iate responses of long-term bond yields and loan rates were q u ite sm all. Three m onths fo llow ing th e s h o c k o n ly a b o u t o n e -th ird of th e c u m u la tiv e increase in call rates had been passed th rough to loan rates, and less than o n e -fo u rth of this increase was tra n sm itte d to bond yie ld s. Over tim e, loan rates c o n tin u e d to rise, in p a rt re fle ctin g a d m in istra tive de cisio n s by the Bank of Japan. But even after a year only about o n e -th ird of the call rate increase was reflected in lo n g term bond yield s. D uring 1985-91, the g e nsaki rate responded more slo w ly to the call rate shock, rising a bout 40 basis p o in ts at the tim e of the shock. N onetheless, three m onths a fte r call rates increased, th e re sp o n se of the gensaki rate was clo se , in p ro p o rtio n a l te rm s, to both its e stim ated lo n g -te rm response and th a t o b se rve d in the e a rlie r p e rio d . L o n g -te rm bond y ie ld s , in c o n tra s t, reacted much more s tro n g ly to call rate s h o cks a fte r 1984. Three m onths a fte r the in itia l sh o ck, n e a rly 60 pe rce n t of the call rate sh o ck was passed th ro u g h to bond yie ld s, a response n e a rly three tim e s as great as d u rin g th e e a rlie r p e rio d . T h e lo a n ra te re s p o n s e show ed little change across th e tw o pe rio d s. S im u la tio n results fo r broad m oney and th e m oney m u ltip lie r in C h a rt 5 s u p p o rt the vie w th a t th e m oney m u ltip lie r was an im p o rta n t p a rt of the p o lic y tra n s m is sion m echanism before 1985. In the 1974-84 p eriod, increases in the call rate caused a sh a rp fall in the m oney m ultiplier. In the firs t th re e m onths fo llo w in g a call rate increase, the m u ltip lie r d e c lin e d by n e a rly 2 percent, in d ica tin g la rg e p o rtfo lio s h ifts by banks. Chart 5 Response of Broad Money and Money Multiplier to a Call Rate Increase Percentage point change 3 ------------------------------------1985-91 Percentage point change 3 1974-84 Call rate Call rate -•----- #- Broad money Money multiplier .1 _ _ _ _ _ Broad money Money multiplier -2 -2 ------ J_ _ _ L J_ _ _ L J_ _ _ L -3 Proportional change 0.5 1974-84 -3 I------Proportional change 0.5 --------------------------- Ratio of broad money to call rate changes ■ Ratio of multiplier to call rate changes ----- *----- •-1.5 J_ _ _ I_ _ _ I_ _ _ I_ _ _ L 2 3 4 5 J_ _ _ I 6 7 8 9 10 11 12 - J _ _ _ I_ _ _ L 2.0 1 2 Months after shock Note: Chart shows the predicted response to an increase of 100 basis points in the call rate. 40 FRBNY Q uarterly Review/Autumn 1991 3 4 5 6 7 8 Months after shock 9 10 11 12 The im m e d ia te e ffe ct of a call rate increase on broad m oney was quite sm a ll, but over tim e, broad money s te a d ily de clin e d . A t three m onths, broad m oney fell by about .4 p erce nt; a year a fte r the shock, broad money d e clin ed by more than 1 percent. D uring 1985-91, the response of broad m oney to a call rate increase has been roughly s im ila r to th a t seen in the e a rlie r period. N o ne th e le ss, the channel of tra n s m ission a p p ears to have ch anged dram atically. C all rate m ovem ents no lo n g e r a lte r the m oney m ultiplier, w hich rem ains roughly unchanged over the fo re ca st horizon. Thus, m o n e ta ry p o licy influ e nce s broad m oney through its e ffe ct on the supply of reserves rather than through its influence on bank asset allo ca tio n . P re d ic ta b ility o f d yn a m ic responses To assess m o n e ta ry control, we m ust co n s id e r not only the size of the response of fin a n cia l m arket variables to p o licy but also the p re d ic ta b ility of these responses. E vidence on p re d ic ta b ility of responses can be obtained by com puting the fo recast standard errors of our va ri ables at d iffe rent horizon s. T hese standard errors, pre sented in Table 8 , ind ica te the degree to w hich the actual responses of fin ancial variables to in te rb a n k rate m o v e m e n ts are lik e ly to fa ll n e a r th e e s tim a te d responses presented in C h a rts 4 and 5 .24 In co m p u tin g th e se sta n d a rd e rro rs, we have exclu d ed th e u n ce r ta in ty a ttrib u ta b le to flu c tu a tio n s in call rates. In our fram ew ork, call rate m ovem ents re p re se nt m o n e ta ry p o licy actions, and th e se e stim a te s sh o u ld ca p tu re the u n c e rta in ty in fin a n c ia l m a rke t v a ria b le s a risin g from fa cto rs oth e r than m o n e ta ry p o lic y .25 The estim a te d fo re ca st va ria n ce s fo r in te re s t rates present a m ixed picture. S ta n d a rd e rro rs fo r the g e nsaki rates d e clin e d s u b s ta n tia lly a fte r 1984, p o s s ib ly re fle ct ing the e m e rgence of m ore sta b le e co n o m ic c o n d itio n s. In c o n tra s t, the s ta n d a rd e rro rs fo r lo n g -te rm bond y ie ld s and lo a n in te re s t ra te s s h o w e d little o r no 24Note that these forecast standard errors only represent the uncertainty arising from u n pred ictable shocks affecting the system. In com puting the standard errors, we ignore un ce rta in ty due to im precision in our coe fficient estim ates. Thus, the forecast standard errors in Table 8 probab ly underestim ate the total un certa in ty surrounding these responses 25Despite this adjustm ent, a com parison of forecast variances for 1974-84 and 1985-91 is likely to be influenced by cha nging policy objectives as well as changes in the general degree of econom ic stability. Table 8 Predictability of Response to Cali Rate Increases Forecast S tandard Error* (P ercentage Points) Months after Shock 1974-84 Gensaki rate 3 6 12 .68 .92 1.39 .38 .55 .81 Long-term bond rate 3 6 12 .57 .72 .98 .63 .72 .92 Loan rate 3 6 12 .21 .37 .65 .20 .41 .80 M 2 + CDs 3 6 12 1.45 2.56 4.97 1.61 2.79 5.53 M onetary base 3 6 12 2.11 2.98 5.18 2.16 3.06 5.55 Money m u ltiplier 3 6 12 1.74 1.91 2.18 1.65 1.67 1.71 1985-91 Notes: The forecast standard errors are derived for three-, six-, and tw elve-period-ahead forecasts con ditiona l on a predeterm ined call rate path. Our ca lcu la tio n uses the un conditional forecast standard error and sub tracts the portion due to call rate shocks. tE xclu d e s fra ction of variance a ttributa ble to call rate changes. FRBNY Q uarterly Review/Autum n 1991 41 s y s te m a tic c h a n g e , w ith th e p o s s ib le e xc e p tio n of increased u n c e rta in ty a tte n d in g loan rate responses at lo n g er h orizon s. The sm all pre-1985 s ta n d a rd errors fo r bond yields and loan rates reflect the re strictio n s on rate m ove m ents in e ffe ct at the tim e. As lib e ra liz a tio n has p ro ce eded , g e nsa ki and o th e r rate forecast e rrors have converge d. Together w ith the d ecline in g ensaki fo re cast e rro rs, th is evide nce su g g e sts th a t the Bank of Japan has been able to in flu e nce the broad spectrum of m a rke t-d e te rm in e d in te re st rates w ith som ew hat greater ce rta in ty. T he fo recast standa rd erro rs fo r m o n e ta ry aggregates un a m b ig u o u sly point to gre a te r u n c e rta in ty fo r the e s ti m ated responses a fte r 1984. Both broad m oney and the m o n e ta ry base responded less p re d icta b ly to call rate chang es, d e sp ite evid e nce of a more sta b le e conom ic e n v iro n m e n t. T h is in c re a s e d u n c e rta in ty in b ro a d m oney re spo nse s m ay in p a rt reflect the d e clin in g use of m o n e ta ry a gg re gate s in p o licy de te rm in atio n . Conclusions We have argued th a t the su b sta n tia l lib e ra liz a tio n of J a p a n ’s fin a n cial system has profoundly affected both the fu n c tio n in g of the Ja p a n e se econom y and the c o n du ct of m o n e ta ry p o licy by the Bank of Japan. The tig h tly re stricte d fin a n c ia l system in the m id-1970s p ro m oted a m o n e ta ry po licy stra te g y that used bank credit to in flu ence activity. F in a n cia l lib e ra liz a tio n has, how ever, reduced the im p o rta n c e of banks in J a p a n ’s flow of fu n d s and e lim in a te d a n u m b e r of p o lic y to o ls th a t re s tric te d b a n k b ehavior. In re s p o n s e , th e B a n k of Japan has g ra d u a lly becom e more d e p e n d e n t on its a b ility to in flu e nce m a rke t in te re s t rates to tra n s m it its policies. O ur a n a lysis of the m o n e ta ry co n tro l m e ch a n ism s u g gests th a t in the c u rre n t lib e ra liz e d e n v iro n m e n t, th e Bank of Japan has been able to tra n s m it its p o lic ie s e ffe ctive ly th rough m a rke t in te re s t rates. In p a rticu la r, in te rb a n k in te re st rate m ovem ents, the key o p e ra tin g ta rg e ts of the Bank of Japan, have pro d u ce d stro n g and c o n s is te n t in te re st rate resp o n se s a cro ss th e Ja p a n e se term stru ctu re since 1984. T he in cre a se d re s p o n s iv e ness of lo n g -te rm bond yie ld s in re ce n t years is p a rtic u la rly notable. Moreover, som e e v id e n c e s u g g e s ts th a t the lin k a g e b e tw e e n p o lic y and in te re s t ra te s has becom e more p re d icta b le . In co n tra st, the lin ka g e b etw een m o n e ta ry p o lic y and broad m oney has p ro b a b ly w e a ke ne d . In th e past, p o l icy actions were la rg e ly tra n s m itte d to broad m oney through the m oney m u ltip lie r as the B ank of Japa n d ire ctly in flu e nce d banks’ p o rtfo lio d e c is io n s . Now, ho w ever, fin a n cial lib e ra liz a tio n has reduced th e B ank of J a p a n ’s leverage over bank behavior, and p o lic y ’s in flu e nce on b road m o n e y w o rks m ore c lo s e ly th ro u g h ch a n g e s in the m o n e ta ry base. A lth o u g h th e average response of broad m oney to p o licy ch a n g e s rem ains a bout the sam e, a g re a te r d egree of u n c e rta in ty a c co m panies the tra n s m is s io n of p o licy to broad money. Appendix: Cointegrating Money and Interest Rate Models This appendix expands on the arguments in the text for cointegration between interest rates and for cointegra tion in the money supply relation. It also presents the dynamic error correction models used to predict the responses of interest rates and monetary aggregates to monetary shocks. Cointegration To investigate the relation between Japanese interest rates and monetary aggregates, we utilize the cointegra tion methodology made popular by Engle and Granger.+ This methodology presupposes that a time series contSee R obert Engle and C.W.J. Granger, "C o-integration and Error C orrection: R epresentation, Estimation, and Testing," E conom etrica, vol. 35 (1987), pp. 251-76. 42 FRBNY Q uarterly Review/Autumn 1991 taining a unit root can only be explained over long peri ods by other series with a unit root. Series with a unit root (also called integrated series) are predicted well by their own lagged values. Typically, regression models for this type of series have substantial residual autocorrela tion because other variables cannot explain the unit root component in the outcome variable. Testing for a unit root in a time series, r„ is commonly carried out by testing whether the coefficient in a regres sion on r, i equals one. The literature uses tests based on the coefficient a in the regression A r, = a r, The coefficient will be zero if the series has a unit root and negative otherwise (unless the series is explosive). The augmented Dickey-Fuller test uses the t-statistic for a in the regression, adding lagged changes in r, to account Appendix: Cointegrating Money and Interest Models (continued) for possible stationary autocorrelation in rt. The StockWatson statistic, as used in this article, is based on filtering r, to eliminate stationary autocorrelation before computing the t-statistic. The Phillips-Perron statistic, Z,„ corrects for stationary autocorrelation by applying a nonparametric correction to 7a, where T is the sample size, in each case, the distribution of the test statistic is nonstandard and requires specially calculated critical values. Several series with unit roots may be linked through a cointegrating model— that is, a model whose residual does not contain a unit root. The regressors in this type of model explain the permanent or unit root component of the dependent variable. Although the variables in the regression may deviate from the regression line in the short run, they return to the regression relationship over time in the absence of additional shocks. In fact, the cointegrating model does not specify the dynamic adjust ment to the model in the long run. This adjustment is specified by auxiliary equations in the error correction model that give the dynamic behavior of the variables. The error correction models are a series of dynamic equations relating the current change in each variable in a cointegrated system to the lagged residual from the cointegrating model and to lagged changes in the vari ables. The coefficient of the lagged residual in our mod els measures the speed of adjustment to long-run equilibrium. Lagged changes appear as explanatory vari ables in the models to allow variables to have differing short-term effects. Thus, we carry out a two-step procedure in our empiri cal work: First, we test the time series for the presence of a unit root. If a unit root is found, we proceed to test for cointegration between those sets of series that could, in theory, be linked. The unit root tests are shown in Table A1 and the cointegration tests are discussed in the text of the article. Second, we use the cointegrating equations to formulate an error correction model for the dynamics in the model. This error correction model, examined in more detail beiow, is used to compute the impulse response functions reported in the text. Cointegration in interest rate and money models The expectations hypothesis of the term structure pro vides one model where short and long rates will be cointegrated.* We assume the existence of the following (approximate) relationship connecting short- and longer *See Thomas S argent, “A Note on Maximum Likelihood Estim ation of the Rational E xpectations Model of the Term S tructure," Journal of M onetary Econom ics, vol. 5 (1979), pp. 133-43; and John C am pbell and Robert Shiller, "C oin tegration and Tests of Present Value M o dels,” Journal o f P olitica l Economy, vol. 95 (1987), pp. 1062-88 term asset returns: (A.1) r'd.D) = (1/D) [r*(1,2) + f(2,3) + ... + f(D ~1,D )l where r'ft.D j is the yield to maturity, in D periods, of the longer term asset; r*(1,2) is the return from period one to two on the shorter term asset; and f( j, j+ l) is today’s forward rate between periods j and j+ 1 . The expecta tions hypothesis connects the forward rate from j io j+ 1 to the expected future spot rate from j to j + 1 as follows: (A.2) f(j,j+ 1) = E, f Q j + l ) + a(l), where a is a risk premium and E1 represents expecta tions at period one. Finally, we suppose that the short rate follows a simple random walk: (A.3) ^ 0 + 1,1+2) = r*(j,J+1) + e(j+1), where e is the unexpected component of the short rate. Since expected future short rates will be directly related to the current spot rate, E ^Q ) = the current long rate will follow: (A.4) i*(1,D) = r*(1,2) + (1/D) Ia (j). Table A1: Unit Root Tests (Monthly, January 1974 to May 1991) Series ADF Call rate Gensaki rate Long-term bond rate CD ratef Loan rate - 3 36 - 3 .7 3 * - 1 .6 5 - 2 .3 4 -2 .5 5 In (M2) In (Base) - 2 .2 6 - .2 5 In (M2)* In (Base)* -2 .8 1 -2 .2 1 ' SW PP - 1 4 01* - 18.38** - 4 45 - 8 .2 6 -9 .0 1 - 8 .7 2 - 13.09 - 4 .5 4 - 8 .1 0 -6 .6 2 -.6 2 -.1 4 - 9 .6 9 -1 8 .3 7 * -.4 9 -.4 4 - 5 .0 0 -1 5 .5 4 Notes: ADF is the augm ented D ickey-Fuller s ta tistic (using seven lags). PP is the P hillips-Perron Zn sta tistic (using seven autovariance lags). SW is the Stock-W atson sta tistic (using seven lags). C ritical values for the D ickey-Fuller and P hillipsPerron statistics were ob tained from P hillips and O uiliaris, “Asym ptotic Properties of R esidual-B ased Tests.” C ritical values for the Stock and Watson sta tistic are from S tock and Watson, "Testing for Comm on Trends." t First sam ple covers 1980-84. ♦Includes time trend. ‘ S ignificant at 10 pe rce n t level. “ S ignificant at 5 pe rcent level. “ ‘ S ignificant at 1 pe rcent level. FRBNY Q uarterly Review/Autum n 1991 Appendix: Cointegrating Money and interest Models (continued) The short rate may have a unit root if it is influenced by either inflation or the real interest rate and if changes in these variables tend to be permanent. Under these con ditions, equation A.4 implies that the long rate should be cointegrated with the short rate. Of course, if the risk premium has a unit root, then the long rate will be cointegrated with the (measurable) short rate and the (not directly measurable) risk premium term, and we would not expect the short rate-long rate pair to be cointegrated by themselves.® ^Because we do not restrict the term spread to be stationary, our interest rate m odels are more general than those strictly im plie d by the expectations hypothesis. A lthough some of our coe fficient estim ates in the cointegrating interest rate models seem far enough from one to cast doubt on the expectations hypothesis, our assets have different issuers and potentially quite different risk cha racteristics, co m p lic a tin g the risk term s in our earlier form ulation. In our money model, we assume that the broad money aggregate is connected to base money through a money multiplier, M2 = m Base, where M2 is the broad aggre gate, Base is base money, and m is the money multiplier. Broad money and base money would be cointegrated if the influences on bank and depositor asset allocation typically only have transitory effects on the money multi plier. To obtain a cointegrating relationship between the base and broad money during 1974-84, we find it neces sary to allow for both trend and interest rate effects. These modifications suggest that the multiplier has a unit root arising from trends in asset choice and permanent interest rate effects. Dynamic models The cointegrating models in the text are incorporated into dynamic models for the call, gensaki, and ten-year Table A2: Error Correction Models D ependent Variables Ind epe nde nt Variables Call C hanges Gensaki C hanges Bond C hanges January 1974 to D ecem ber 1984 Sum of 56(.28) lagg ed call changes lagg ed gensaki changes - ,06(4)9) .3 1 (1 5 ) lagg ed bond changes lag g e d loan changes .28( 55) 4.73(4.69) la g g e d M2 changes -1 .5 8 (3 .3 6 ) lagg ed base changes .3 5 (2 1 ) —1.11 (.43) .57( 22) 1.54(.73) 2.37(0.93) 3.86(4.75) 08(.08) .01 (.06) .02(.21) - .07(.24) .67(3.17) - 2 36(2.09) 24(.09) 1.12(2.63) .0 1 (1 4 ) 1.43(3.87) - .1 3 (0 6 ) Residual (gensaki) R esidual (money) R2 January 1985 to May 1991 Sum of lagg ed call changes lagg ed gensaki changes lagg ed bond changes lagg ed loan changes lagg ed M2 changes lagg ed base changes Residual (gensaki) Residual (long-term bond) Residual (money) R2 .22 .30 .12 - .1 8 ( 1 0 ) .39(.26) ,06(,08) .1 4 (3 6 ) 2.32(2.39) -.9 3 (1 .4 3 ) 02 (.17) - ,23(.30) . 1 7 (3 8 ) ,53(.46) 3.14(4.28) -2 .8 0 (2 .3 4 ) .1 9 (1 8 ) -2 5 (1 1 ) .1 3 (0 6 ) -1 .8 3 (1 .5 8 ) -r.2 4 (2 1 ) - ,1 8 (.1 5 ) .45 .44 1 0 (0 4 ) ,05(,03) 11 (.03) 58(.07) — 1.06(.69) .3 2 (4 6 ) — ~ .72 (.31 ) .89(.27) .1 0 (1 1 ) .7 6 (4 0 ) 4.84(3.77) -2 .8 7 (2 .2 6 ) -2 .6 9 (2 .4 9 ) Loan C hanges .87 .03( 05) .0 2 (0 7 ) ,09(.03) .8 2 (0 6 ) 22( 49) .. 25( 27) M2 C hanges .002(.002) .000(.001) -.0 0 1 (.002) - .008(.006) ,90(.08) —,04(.05) .01( 004) .001 (.003) .001 (.005) - ,02 (.01) ,33(.18) ,16(.24) -.0 6 (.0 4 ) .20(.08) .01 - .0 0 0 ( 0 0 3 ) .004(.006) .0 0 5 (0 0 3 ) - .02(.01) 1.08(.14) - .21(.06) .26 -,0 0 2 (,0 0 7 ) 006(.013) 006(.006) .0 0 9 (0 1 8 ) 18(.25) ,48(.32) — - .1 0 (0 7 ) .17 Base C hanges .91 .26 .6 0 (1 5 ) .40 Notes: S tandard errors are shown in parentheses. The residual for the gensaki rate is obtained from the c o in te g ra tin g m odel con n e ctin g the gensaki and call rates; the residual for the long-term bond rate is obtained from the co in te g ra tin g m odel co n necting the long-term governm ent bond and gensaki rates; the residual for money is the residual from the cointegra ting m odel for money supply. 44 FRBNY Q uarterly Review/Autumn 1991 Appendix: Cointegrating Money and Interest Rate Models (continued) government bond rates as well as the monetary base and the M2 + CDs aggregate. Since we do not detect a long term relationship between the loan rate and other inter est rates, our cointegrating equations do not include the loan models presented in Table 6 of the text. The coin tegrating, or long-run, interest rate models are: (A.5) G = a0 + a,C + LJ — a 2 + a3G + /== 1 €lJ, ln(M2) = k0+ In(Base) + k,G + k2C + k3t + eM2. Our estimates of these models are given in text Tables 6 and 7. The dynamic equations have a general error correction form that relates the current change in each variable to lagged changes in all of the variables and to the residuals from the cointegrating coregressions.+t Esti mates of these dynamic models are shown in Table A2. We typically include four lags of the dependent variable and one lag of the other variables.§§ln most cases, we If The main text presents a c o integra ting model connecting the governm ent bond yield to the call rate. Our m odeling stra tegy uses the equation A .6 connecting the bond yield and the gensaki rate as the long-term relation for the bond yield. The correspo ndin g equations are: 1974-84 LJ = 275 + 28G. R2 = .51 1985-91 LJ * 1.01 + 84G, R2 = .86. There is strong evidence of cointegra tion in the second period and essentially none in the first. ttF or details, see Engle and Granger, "C o-integ ration and Testing," and Jam es Stock, “A sym ptotic Properties of Least Squares Estim ators of C ointe gra ting Vectors," Econom etrica, 1987, pp. 1035-56 SfOur cho ice of lag structure is m otivated by the autocorrelations of first differences of the data. These generally seem consistent with fou rth-ord er autoregressive m odels. We in clu d e more lags of other interest rates in the loan equation to allow for possible effects over several periods. AC(f) = —5ieG(f —1) —82e ^ (f-1 )-8 3eM2(t-1 ) 4 + X P,AC(f - i ) + y-[ AG(t —1) + ^ ALJ(t -1 ) + 01ALoan(?-1) + (,1AlnM2(t~'\) + <i>,AlnBase(t-l). where C represents the call rate, G is the gensaki rate, and LJ is the Japanese benchmark long-term govern ment bond yield.!l Our cointegrating equation for broad money supply relates broad money, M2, to the base, Base, the gensaki rate, G, the call rate, C, and a time trend t, restricting the coefficient on the base so that the interest rates and time trend affect the money multiplier: (A.7) (A.8) eG and (A.6) only include residuals from cointegrating equations that contain the dependent variable. The general form of the model is illustrated by the call rate equation below: The lagged residuals in this equation ensure that the short-run behavior in the dynamic model will converge to the long-run behavior embodied in equations A.5, A.6, and A.7. The lagged changes of the variables allow the short-run impacts of interest rate shocks to differ from the long-run behavior in the cointegrating equations. Analogous to the call rate equation are the equations for the gensaki rate, the Japanese long-term government bond yield, and bank loan rate given below: (A.9) AG(f) = -S 1eG( f- 1 ) - 5 2€ ^ (f-1 )-S 3eM2(t-1 ) 4 + M C (f~ 1 ) 2 "y, AG(f - /) + ALJ{t -1 ) /'= 1 + 0, ALoan(t -1 ) + £AlnM2(t -1 ) + <J>,AlnBase(t -1), (A.10) - S1c<3(f—1) —82€ ^ (f-1) 4 + p1AC (f-1) + 7 lA G (f-1)+ £ 4/,ALJ( t- i) i- 1 + 0, ALoan{t-1 ) + £, AlnM2(t -1 ) + <}>,AlnBase{t -1), ALJ(t) = and (A.1!) ALoan{t)= 3 X /'=0 3 P,AC(f-/)+ X 3 3 •+ £ %ALJ( t - i) + 2 i= 1 + £1 AlnM2(t -1 ) i= y A G (t-i) 1 ®ALoan{t-i) /'=1 + 4), MnBase(t -1). These equations allow us to estimate the dynamic response of Japanese interest rates to shocks in the call rate (our proxy for Japanese monetary policy actions). The dynamic error correction equations for the mone tary base, Base, and broad money, M2, have the follow FRBNY Q uarterly Review/Autum n 1991 45 J|§ f ji | §* A n n o n H iv P n in to n rn tin n Money M Appendix: Cointegrating and Interest Rate Models (continued) ing forms, in which the lagged residual is obtained from equation A.7: (A.12) AlnM2(t)= -5 e M2(f-1 ) + 0, AC(t -1 ) + , AG(f -1 ) + 4>,ALJ(t -1 ) + d>,ALoan(t -1 ) 7 4 + I 1 -/) + <{>!MnBase(t -1 ) I—I and (A.13) AlnBase{t) = - BeM2(f -1 ) + 01A C (f-1) + 7 tA G (t-1) + \\>,ALJ( t - 1) + (M Z-oanff- 1 ) siai® 4 + [,,& lnM 2(t-1)+ ' i <$>,UnBase(t-i). /= 1 46 FRBNY Q uarterly Review/Autumn 1991 To simulate the models and to compute forecast stan dard errors, we have to impose a structure on the current disturbances in the error correction equations. We use an ordering of the disturbances (call, base, gensaki, long-term bond, M2 + CDs, loan) in which current shocks to each variable in the list affect contemporaneous shocks in variables listed later. Our ordering allows the current call rate shocks to affect shocks in all the other variables in the system. When we analyze predictability of policy responses, we remove the component attribut able to call rate shocks from the forecast standard errors to exclude uncertainty related to policy changes. Although these standard errors correctly measure uncer tainty when monetary policy is designed to minimize the variance of a single variable, the interpretation of the errors may be more difficult when monetary policy has multiple objectives. ............ , ‘ ..............r .ll..- Expected Inflation and Real Interest Rates Based on Index-linked Bond Prices: The U.K. Experience by Gabriel de Kock Recently some analysts have suggested that the Trea sury finance part of the federal deficit by floating indexed bonds.1 One of the claims made for this strat egy is that it would yield significant monetary policy benefits. In particular, the prices of indexed bonds could offer timely and accurate market measures of expected inflation and ex ante real interest rates. As such they could provide the Federal Reserve System with valu able information about market perceptions of, and reac tion to, its policies. The argument has also been made that a real-time market measure of expected inflation might provide the Federal Reserve System with a valu able indicator of the future course of inflation and offer the public a ready means of monitoring the Fed, thereby encouraging public interest in better policies. A market measure of expected inflation may be useful in the formulation of policy even if is not a good gauge of the actual future inflation performance of the econ omy. It may be a poor indicator of future inflation because private sector inflation expectations are, in fact, not realized. Even in this case, however, the asso ciated real interest rate should nevertheless be a good measure of ex ante real interest rates faced by the private sector. That is, at the macroeconomic level, indexed bond prices should contain useful information about real economic activity. The potential value of a real-time market measure of expected inflation to policy makers can only be assessed indirectly because no “true” alternative mea sure of private sector inflation expectations exists. This article evaluates the usefulness of a market measure of expected inflation by applying two closely related tests: ’ See, for e xam p le, R obert H etzel, "A Better Way to Fight Inflation,” The W all S tre e t J o u rn a l, April 25, 1991. first, whether the market measure of expected inflation is a good indicator of inflation developments; and sec ond, how well the market measure captures private sector inflation expectations even if the expectations do not reflect actual inflation performance. The second of these tests is based on two proposi tions. The first proposition is the familiar Fisher hypoth esis that nominal interest rates should equal expected inflation plus the ex ante real interest rate; the second, a prediction shared by all standard dynamic mac roeconomic models, is that true real interest rates should provide information about future economic activ ity.2 If the market-expected real interest rate, measured as the difference between a long-term nominal rate and the market measure of expected inflation, has no signifi cant effect on future real economic activity, it seems likely that it is also a poor measure of the true real interest rate. Under these circumstances, the market measure of expected inflation would then seem likely to be a poor measure of “true” private sector inflation expectations. If a real-time measure of inflation expectations neither anticipates future inflation developments nor conveys useful information about real economic activity, it is probably of only limited use to policy makers. Drawing on this framework, this article examines the U.K. experience with indexed gilts (IGs) to assess whether indexed bond prices convey information useful in form ulating and m onitoring m o n etary policy. 2ln theory, higher real interest rates do not necessarily lead to lower real econom ic activity (G N P ); in fact, the two variab les m ay be positively a sso ciated with e ac h other. In the context of s tan d a rd m acroeconom etric m odels, however, real interest rates and econom ic activity are, ceteris paribus, negatively related to e ac h other. FRBNY Quarterly Review/Autumn 1991 47 More specifically, the article evaluates whether the expected inflation rate derived from the prices of indexed and nominal bonds predicts future inflation and whether the corresponding expected real interest rate provides information about real economic activity not obtainable from more traditional information variables and measures of policy stance. Although several coun tries have experimented with indexation since World War II, the experience of the United Kingdom, where marketable index-linked gilts have been issued since March 1981, is likely to be the most relevant to the United States. While the market for index-linked gilts does provide a real time measure— accurate or not— of expected infla tion and ex ante real interest rates, this information does not appear to be of much practical value in for mulating and evaluating monetary policy. This general interpretation of the data derives from two specific con clusions: First, the expected inflation rate embodied in nominal bond yields is no better than simple measures of inflation expectations based on past inflation alone. It is a biased predictor of the future level of inflation, although it does provide minimal information about acceleration and deceleration of inflation. Second, indexed gilt prices do not seem to provide information about future movements in real economic activity, sug gesting that the real interest rate on indexed gilts is unlikely to be a good measure of ex ante real interest rates faced by the private sector in the markets for goods and services. The expected inflation rate embod ied in U.K. bond yields, therefore, appears to be a poor measure of true inflation expectations, given our empiri cal results. By contrast, lagged inflation and nominal interest rates often used to derive real interest rates that may measure monetary policy do have predictive content for U.K. real GNP growth. The first section of the article summarizes the devel opment of the U.K. indexed gilt market. More specifi cally, it focuses on the particular circumstances surrounding the introduction of IGs in the United King dom and the main features of the IG market at present. There follows a brief discussion of the potential mone tary policy role of indexed bonds and a review of the decomposition of nominal yields into expected real interest rate and expected inflation components. The article then examines the information about future infla tion provided by indexed bonds and the ability of indexed gilt prices to predict developments in real eco nomic activity. The evolution of the U.K. market for index-linked gilts The Conservative government introduced IGs in 1981 as part of its anti-inflation program. Three reasons were http://fraser.stlouisfed.org/ 48 FRBNY Quarterly Review/Autumn 1991 Federal Reserve Bank of St. Louis given for the move: (1) the introduction of IGs would improve the Bank of England’s control over monetary aggregates, (2) indexation would result in substantial savings to the Treasury if, as anticipated, inflation declined significantly as a result of the government’s policies, and (3) indexing government debt would signal the government’s determination to reduce inflation. These reasons reflected policy concerns specific to the United Kingdom in the early eighties, although the third is sometimes viewed as relevant to current U.S. policy, if only on a theoretical level.3 Issuing IGs was expected to bring about closer con trol over the monetary aggregates by ameliorating con straints on monetary policy imposed by the distinctive structure of the U.K. gilts market. Because market makers, considered essential to the smooth functioning of the market, were weakly capitalized, the authorities felt obliged to minimize fluctuations in gilt prices that would threaten the market makers’ survival. Thus, the Bank of England was constrained to stabilize nominal interest rates, a policy that entailed loss of control over monetary aggregates.4 Most notably, in the late seven ties and early eighties, market expectations of rising inflation forced the Bank of England to follow a destabilizing expansionary policy to prevent gilts prices from falling too steeply. Index-linked gilts, which could be sold in times of market expectations of rising infla tion, enabled the authorities to reestablish control over monetary aggregates. In this way, inflation expectations could be kept in check, thereby mitigating fluctuations in conventional gilts prices. The second argument for issuing indexed bonds, that the real cost to the government of issuing index-linked debt would be lower than that of borrowing on conven tional terms, is valid if the government expects an infla tion rate lower than the market expectation of inflation embedded in nominal yields (assuming that the tax system is neutral with respect to inflation). These condi tions were clearly fulfilled in the United Kingdom in early 1981: long-term bond rates were close to 14 per cent and retail prices were still rising rapidly, but the government expected its firm anti-inflation policies to pay off in the near future. However, under the indexation scheme envisaged, the tax system would not be neutral with respect to inflation, because the inflation compo nent of nominal rates would be fully taxable while only 3See C harles A. E. G o odh art, M o ne y, In fo rm a tio n a n d U n c e rta in ty (C am brid ge: M .I.T. Press, 1989), for a discussion of the policy d e b a te surrounding the introduction of in dexed gilts. «The Bank of England could not use op en m arket o p eration s at the short end of the m arket b e c a u s e the stock of Treasury bills outstanding was very small by U .S. stan dard s. At the e nd of M arch 1980, for exam ple, Treasury bills a c c o u n te d for only 2 .9 p e rce n t of m arket holdings of governm ent deb t. part of the inflation compensation on index-linked gilts would be taxed. Thus, although government interest outlays would be lower with indexed gilts, conventional gilts could be expected to produce much more tax revenue. Initial calculations suggested that inflation would have to decline very rapidly before the reduction in outlays brought about by indexation IGs would exceed the loss in revenue entailed. The government’s third reason for introducing indexed bonds— to enhance the credibility of its anti-inflation program— derives from the fact that index-linking reduces the benefits of unanticipated inflation. Investors will be justly skeptical of announced anti-inflationary policies if the government at the same time issues nominally denominated debt, because the government may always be tempted to resort to unanticipated infla tion to reduce the real value of its debt. By contrast, unanticipated inflation does not promise any capital gains to the government if its debt is indexed. Conse quently, by issuing indexed debt, the government could enhance its credibility.5 The advantages of IGs were partly offset by a number of possible disadvantages. First, there was concern that issuing an attractive long-term asset could have a nega tive impact on equity prices and corporate financing opportunities. Second, it was feared that foreign demand for IGs might put upward pressure on the pound, which was already overvalued as a result of tight monetary policies and the discovery of North Sea oil. Third, since capital gains were not yet indexed for tax purposes, issuing indexed debt would entail the taxa tion of purely inflationary capital gains on IGs— a step that could in turn stimulate political pressure for the indexation of taxes. Finally, the United Kingdom faced strong political pressure from other OECD governments concerned that any form of indexation would fuel OPEC pressure to index-link oil prices. Indexed gilts were issued consistently throughout the 1980s at coupon rates mostly between 3 and 4 percent (compared with 2 percent for the first two issues). As early as end-March 1985, IGs made up 6.5 percent of total market holdings of U.K. public debt; by March 1990, the total amount of IGs outstanding was about 5Note, however, that indexation could also have an adverse im pact on exp ectatio ns (and thereby m ake it more difficult to reduce inflation) if it was interpreted as an effort by the m onetary authorities to dec re a s e the political cost of inflation before giving up the battle against inflation altogether. But such an adverse im pact would be more likely if indexation covered a w ide range of co n tra cts — som ething the C onservative governm ent had taken great pains to avoid. Issuing index-linked gilts probably also enhanced the go vernm ent’s credibility over tim e bec a u s e it effected im m ediate cosm etic im provem ent in the Public Sector Borrowing Requirement: In d e x a tio n was im plem ented in a way that pushed com pensation for inflationary d e p reciation of principal into the future, w hereas the governm ent would have had to pay higher nominal interest rates im m ed iately had it issued conventional bonds. £17.5 billion, or 10.9 percent of market holdings of British government debt and 19.3 percent of the value of gilts outstanding (Table 1).6 The stock of IGs outstand ing is made up of thirteen issues with maturities varying from two to thirty-three years. Long-dated issues make up the bulk of the value of IGs outstanding; only 7.3 percent of the amount outstanding are of maturities shorter than five years, and 16.5 percent of maturities shorter than ten years (Table 2). Anecdotal evidence suggests that holdings of IGs remain concentrated and that the number of customers remains small.7 The most important holders of IGs are pension funds, followed by insurance companies, while individual investors are largely confined to the short end of the market. These features match those of the gilt market as a whole. The IG market is thin in comparison with the market for conventional gilts. Although turnover varies, it only amounted to 2.9 percent of total gilt turnover in 1990 and 3.1 percent for the first four months of 1991 (Table 2). The demand for new issues of IGs has been disappointing when the ex post real yields on conventional gilts exceed those on IGs, as they did during the rapid decline of inflation in 1982-83 and 1985-86. (The issuance of IGs also resulted in substan tial savings to the Treasury during this period for the same reason.) The potential role of indexed bonds in monetary policy Proponents of issuing indexed bonds in the United States have emphasized monetary policy benefits that depend critically on the informational role of indexed bond prices. This consideration figures importantly in the academic literature on indexed bonds although it did not arise in the U.K. policy debate.8 Advocates argue 6By the end of 1990 this p e rc e n ta g e had risen to 2 0 .5 p ercen t b e cause under the s ch em e of indexation used in the U nited K ingdom , the value of the IG s ou tstand ing rises in line with inflation. N onm a rk e ta b le national savings c ertific ate s or “granny bonds," which have been issued since 1975, have had a lim ited im pact, m aking up less than 2 p e rce n t of m arket holdings of British governm ent d e b t as of e n d -M a rc h 1990. 7The Bank of England does not com pile s e p a ra te statistics on holdings of IGs and conventional gilts by type of institution. At the end of M arch 1990, pension funds and in surance co m p a n ie s held 5 7 .5 percen t of U.K. governm ent d e b t ou tstand ing , while individuals and private trusts held about 3 8 p e rce n t (“The N et D eb t of the Public Sector: E n d -M arc h 1 9 9 0 ,” B a n k o f E n g la n d Q u a rte rly B u lle tin , N ovem ber 1990, pp. 519 -2 6 ). 8A theoretical literature on the relative e fficiency of op en m arket operations in indexed and nom inal bonds d ates b ack to Jam es Tobin, "An Essay on the P rinciples of Public D eb t M a n a g e m e n t,” reprinted in M a c ro e c o n o m ic s , vol. 1 of E s s a y s in E c o n o m ic s (M arkham Publishing C o., 1971). Tobin a rg u e d that op en m arket operations in in dexed bonds will affect real activity more strongly and with greater certa in ty than will op en m arket op eration s in FRBNY Quarterly Review/Autumn 1991 49 Table 1 Com position of U.K. National Debt March 1985 Billions of Pounds M arket ho ld ings S terling m arketable debt G overnm ent stock Index-linked Other Treasury bills S terling nonm arketable de bt National savings'^ Index-linked Other Other Foreign currency de bt O fficial holdings M arch 1990 Percentage B illions of Pounds P ercentage 146.7 100.0 160.0 100.0 114.4 78.0 117.0 73.2 9.5 103.7 1.2 6.5 71.5 0.8 17.5 90.5 9.0 10.9 56.6 5.7 29.3 20.0 36.5 22.8 3.6 18.8 6.9 2.4 12.8 6.8 3.0 26.1 7.4 1.9 16.3 4.6 2.9 2.0 6.5 4.0 11.6 32.5 Source: Bank of England. t N ational savings includ e a variety of no n-negotiable savings instrum ents issued by the governm ent. th a t indexed bond p rice s w ould provide p olicy m akers and the pub lic w ith info rm a tio n on inflation e xpectations and ex ante real in te re st rates on a re al-tim e basis. In this view, the Federal R eserve System w ould gain valu able info rm a tio n about m arket p e rce p tio n s of, and reac tio n to, its po licies. F u rtherm ore, this in form ation could provide p o licy m akers w ith a good pre d icto r of inflation and a m easure of the im pact of m o n e ta ry p o licy on both in fla tio n and real eco n o m ic activity. H etzel has argued th a t the ready a va ila b ility to policy m akers and the pub lic of an in d ica to r of the in fla tio n a ry co n se q u e n ce s of m o n e ta ry p o licie s w ould have three benefits. F irst, it w ould increase pu b lic u n d e rsta n d in g of, and su p p o rt for, a n ti-in fla tio n a ry policie s. S econd, it w ould serve as a b a ro m e te r of Fed cre d ib ility and co n se q u e n tly increase in ce n tive s for the Fed to c o m m it itse lf to a n ti-in fla tio n a ry p o licie s. Finally, by exposing the true co n se q u e n ce s of p o licie s th a t trade o ff in flation for s h o rt-te rm outp u t g ains, it w ould stren g th e n the Fed’s e ffo rt to focus a tte n tio n on its lo n g -te rm p rice s ta b ility o b je c tiv e s . Table 2 Features of U.K. Market for Index-linked Gilts M aturity C om position of IGs O utstanding M illions of Pounds1 Less than one year One to five years Five to ten years Over ten years Total P ercentage 865 638 1,920 17,365 20,788 4.2 3.1 9.2 83,5 100.0 C ontributio n of IGs to M onthly G ilt Turnover Total Turnover_____IG Turnovef____ (B illions (B illions (Percentage of Pounds) of Pounds) of Total) 1990 average January-April 1991 average 75,445.5 85,702.9 2,196.6 2,630.6 2.9 3.1 Source: Bank of England. tln c lu d e s indexation of p rin cip a l up to the b e ginnin g of 1991. Footnote 8 (continued) nominal bonds because indexed bonds are likely to be a closer substitute for equity than nominal bonds. Tobin has been criticized by Stanley Fischer, “The Demand for Indexed Bonds," Journal of P olitical Econom y, vol. 83, no. 3 (June 1975), pp. 509-34, and by Paul Beckerman, "Index-linked Government Bonds and the Efficiency of Monetary Policy," Journal o f M acroeconom ics, vol. 2, no. 4 (Fall 1980), pp. 307-31. Fischer suggested that open market operations in nominal bonds that are complements for equity in private portfolios may have a more pronounced impact on Tobin’s q, and thus on real activity, than open market operations in indexed bonds that serve as substitutes for equity. Beckerman has pointed out that Tobin's conclusion requires a set of potentially inconsistent assumptions. http://fraser.stlouisfed.org/ 50 FRBNY Q uarterly Review/Autumn 1991 Federal Reserve Bank of St. Louis However, the b e n e fits cite d by H etzel are o n ly lik e ly to m a te ria lize if the m a rke t m easure of e xp e cte d in fla tio n is a re liable in d ic a to r of the e ffe cts of m o n e ta ry p o lic ie s and m a cro e co no m ic d is tu rb a n c e s on in fla tio n .9 9Earlier advocates of indexation— for example, Alicia H. Munnell and Joseph B. Grolnic ("Should the U.S. Government Issue Index Bonds," New E ngland E conom ic Review, September-October 1986, pp. 3-22)— have emphasized the provision of index-linked government liabilities that could be used to back indexed pension contracts. To be sure, a market-based decomposition of nominal interest rates into their expected inflation and expected real interest rate components may be useful to policy makers, even if the market measure of expected infla tion is not a good predictor of inflation. This would be especially true if the market measure of expected infla tion were a good gauge of private sector inflation expec tations. If so, it would provide a reliable measure of ex ante real interest rates and thus convey information on private decisions and future economic developments. More generally, indexed bond prices could offer policy makers and private agents up-to-date information about the sources of macroeconomic disturbances. In fact, Boschen, using a simple model, has shown that a mar ket for indexed bonds could reduce the magnitude of business cycle fluctuations by allowing private agents to distinguish real and nominal disturbances more accurately.10 The empirical analysis in this section evaluates whether the U.K. market for index-linked gilts actually conveys the policy-relevant information about future inflation and real economic activity that advocates of indexed bonds have attributed to it. Data on IG and conventional gilt prices are used to construct a monthly series of expected inflation rates and expected real interest rates spanning the period from March 1982 to March 1991. As detailed below, these data indicate that the derived measure of expected inflation, termed the IG measure, is a poor predictor of inflation and that the IG market does not provide information about future real economic activity. Inflation expectations and expected real interest rates derived from IG prices The data on expected real interest rates and expected inflation are constructed by using an indexed bond's price to decompose the yield on a nominal bond into expected real interest rate and expected inflation com ponents. The calculation assumes that the expected real yields of indexed and conventional bonds of the same maturity must be equal and consequently that the IG measure of the inflation rate expected by investors to prevail over the rem aining lifetim e o f the bonds can be estimated as the difference between the redemption yields on the nominal and indexed bonds. In practice, the calculation and interpretation of the expected infla tion rate and expected real rate are somewhat more involved, because IGs are not fully indexed, investors may be risk averse, and the indexed and conventional bonds may differ in liquidity and tax treatment. More detailed information on the nature of indexed gilts and 10S ee John F. B oschen, “The Inform ation C ontent of In dexed B onds," J o u r n a l o f M o ne y, C re d it, a n d B a n k in g , vol. 18, no. 1 (February 1986), pp. 7 6 -8 7 . the calculation and interpretation of expected inflation and real interest rates is given in the box. Data on the IG measure of expected inflation and the corresponding long-term real interest rate are derived by decomposing the nominal yield on a conventional bond maturing in 1996. This calculation yields the long est data series because the first IGs issued mature in 1996. The top panel of Chart 1 illustrates the decom position of this long-term yield into its expected real yield and expected inflation components. Note that there is a break in the series in 1986.11 For reference, the lower panel of Chart 1 shows the nominal yield on the 1996 bond along with the yield on ten-year gilts. The yield on the 1996 bond moves closely with the yield on ten-year gilts, confirming that the 1996 bond (as well as the decomposition of its yield) is representative of the long-term government bond market in the United Kingdom. Chart 1 suggests that changes in expected inflation account for the bulk of nominal interest rate move ments. This result is the counterpart of the striking stability of the real interest rate measure, which varies between 2.36 percent and 4.37 percent per annum. The underlying stability of the real yield on IGs is confirmed by the Bank of England’s calculations: although based on the assumption of a fixed 5 percent inflation rate over the remaining lifetime of the IG, the Bank’s esti mate of the real yield varies over a similar range.12 Chart 1 also illustrates the response of the IG mea sure of expected inflation and the corresponding real 11The data for the period from O c to b er 1986 to M arch 1991 p ertain to a 10 percent coupon bond m aturing in N ovem ber 1996 and those for the period from M arch 1982 to S e p te m b e r 1986 to a 14 p ercen t coupon bond m aturing in July 1996. The decom p o s itio n of the nominal yield is bas e d on price d a ta for a 2 p e rce n t in d e xe d gilt m aturing in S e p te m b e r 1996. In both cas e s the m aturity m atch is probably close enough not to affect the results. The decom position is som ew hat sensitive to the p a rtic u la r m atc h e d -m a tu rity conventional bond used, presum ably b e c a u s e the different bonds are not equally liquid. The yields on the 14 p e rce n t bond and the 10 percent bond differ by 2 9 basis points, on ave ra g e , in the months for which overlapping ob servations are available. Our results nevertheless in dic a te that only the le v e ls of c a lc u la te d e x p e cte d real yields and inflation rates are a ffec te d , not their m ovem ents over tim e. For e xam p le, the correlation coefficient of the two exp e cte d inflation rates c a lc u la te d for the overlapping observations is 0 .9 9 . For the pu rp oses of C h a rt 1 and the em p irical analysis discussed below, the e arlie r observations w ere a d ju ste d by the m ean difference c a lc u la te d from the overlappin g observations. Specifically, the real rate in creased abou t 6 basis points and the e xp e cte d inflation rate d e c lin e d by abou t 3 5 basis points. 12The Bank of E n g la n d ’s real interest rate m easure is som ew hat higher, on average, than the m easure rep orted here (3 .6 4 p ercen t com pared with 3 .5 4 p e rce n t) and som ew hat less volatile; its s tandard deviation is abou t 4 0 .5 basis points, com p a re d with 4 2 .6 basis points for the m easure used here. The correlation coefficient of the two m easures is 0 .9 3 . In “Sources of Fluctuation," G aske docum ents that in the early part of the sam ple the co-m ovem ents betw een the Bank of England's series and m acroecono m ic variab les are quite different from those of an ex ante real rate m easure like the one used in this article. FRBNY Quarterly Review/Autumn 1991 51 Box: Extracting Ex Ante Real Yields and Expected Inflation from Indexed Gilts Prices Features of U.S. index-linked gilts The form of index-linking adopted in the United Kingdom may be called principal value indexation. The value of the principal of an IG is linked to the retail price index (RPI), and coupon payments, payable every six months, are calculated as a fixed percentage of this inflation-adjusted principal. Holders of IGs are not fully protected against inflation, and hence the real return on an IG is uncertain, because the principal— and consequently interest pay ments— are indexed to the RPI with an eight-month lag. That is, the value of the bond for the purpose of calculat ing the coupon payment for a given six-month interest period exceeds its initial face value by the increase in the RPI over the period starting eight months before the issue date and ending eight months before the date on which the coupon payment is made. For example, the principal in period t of a £100 bond issued on date I and paying a 2 percent coupon is £100x(RPIt _8/RPI,_8), and the coupon payment on date t will be £ 1 x(R P It_a/ RPIi „ 8), where time is measured in months. Note also that, because of the lag in indexation, an IG is a pure nominal bond during the last eight months of its lifetime. The eight-month lag is needed to ensure that the rate of interest accrual in money terms for any six-month period is known before the start of that period so that pur chasers can compensate sellers for interest accrued since the last coupon payment preceding the transaction. The period for which interest is due accounts for six of the eight months; the normal lag in availability of the RPI data for the seventh month; and the need to avoid prob lems that could arise if the publication lag exceeded one month, for the eighth months Eligibility to take up the initial offerings of IGs in 1981 and early 1982 was restricted to domestic tax-exempt institutions (pension funds, life insurance companies tak ing pension business, and charitable societies) in order to forestall potential tax problems and to avoid repercus sions for the exchange rate. These restrictions on the ownership of IGs became redundant and were removed when the government introduced indexation of capital gains for tax purposes in March 1982. Since that time, IGs have enjoyed a significant tax advantage relative to conventional gilts. While holders of conventional gilts are taxed at the income tax rate on nominal interest earn ings— a rate that consists in part of compensation for tFor a discussio n of institutional features of the indexed gilt market, see Patrick P hillips, Inside the New G ilt-e dged M arket, 2d ed. (C am bridge, England: W oodhead-Faulkner, 1987). FRBNY Q uarterly Review/Autumn 1991 depreciation of principal— holders of IGs pay no taxes at all on inflationary increases in the nominal value of the IGs. Consequently, an anticipated increase in inflation will tend to depress conventional gilts prices relative to IG prices by more than necessary to equate pretax nominal yields on conventionals and IGs. Calculation method The imperfect indexation of IGs makes it impossible to calculate an expected real redemption yield on the IG without making an assumption about inflation over the bond’s remaining lifetime. Nevertheless, as long as investors are risk neutral, the prices of an IG and a nominal gilt of matched maturity can still be used to decompose the nominal yield on the conventional gilt into an expected real rate and an expected inflation rate. The procedure used to calculate the expected real yield and the expected inflation rate derives from work by Arak and Kreicher, Woodward, and Gaske.* It can be explained by a simple example that captures the salient features of the U.K. gilt market. Consider a maturitymatched pair of nominal and indexed bonds with face values Fn and F (at issue) maturing in period T. Let P" and P't denote the (nominal) prices of the nominal and indexed bonds, respectively, at the beginning of period t, with time measured in months. Coupon payments are made every six months. The first payment after period t occurs in period t+ j (j«6), and the last payment coin cides with redemption in period T. The coupon payment on the nominal bond is denoted by Cn. If the IG was issued in period I, its redemption value will be FV= F® x (RPIT_8/RPI, „ 8), and the nominal coupon paid in period t + j will be CUj = O x (RPI,+j_8/RPIi-e)> where C® is the face coupon (C1 = F® x c = face value x coupon rate) on the index-linked bond. Finally, let it denote the periodt annual yield to maturity on the nominal bond, -nf the annual inflation rate expected to prevail from period t to period T, and rt the annual expected real interest rate from period t to period T. Then i, is the solution to *See M arcelle Arak and Lawrence Kreicher, "The Real Rate of Interest: Inferences from the New U.K. Indexed G ilts ,” International Econom ic Review, vol. 26, no. 2 (June), pp. 399-407; G. Thomas W oodward, "C om m ent: ‘The Real Rate of Interest: Inferences from the New U.K Indexed G ilts,'" International E conom ic Review, vol. 29, no. 3 (August), pp. 565-68, and Mary Ellen Gaske, "S ources of Fluctuations in E xpected Long-term Real Rates: E vidence Extracted from U.K. Indexed Bond R ates,” U npu blished Ph.D. D issertation, U niversity of M aryland. Box: Extracting Ex Ante Real Yields and Expected Inflation from Indexed Gilts Prices (continued) K (A.1) Ptn =: C" J (1+ it/1 2 )-« *+i> + {1 + it/12)-(T-«)Fn, k= 0 where K * [ T - ( t + j)]/6. The decomposition of i, into irf and r, must satisfy K (A.2) Pj= J [(1 + r,/12)(1 +irf/12)]-U+6k)C[+t+6k ] k= 0 + [(1 + r,/12)(1 +irf/12)]-(T-'>FV and (A.3)(1 + r,/12)(1 + Trf/12) = (1 + it/12). Since the expected nominal coupon on the indexed bond in period t+ j + 6k is Ci +j +6k = (1 + 'jrf/12)6kCj+j • (1 + irf/12)6k X (RPIt+j. 8/R PI,.8) x O and the expected nominal redemption value of the indexed bond is FV= (1+'rrf/12)T- t' 8 x (RPI,/RPI,_8) X P, equation A.2 can be simplified to (A.4) P; = (R PIt+j. ft/R P tl . 8)[(1 +n f/1 2 )(1 + rt/1 2 )]-iO K + (RPIt/R P I,.a)(1 + irf/1 2 )-8{ J (1 +r,) <6k+i>Ci + (1 +r,)~<T-,)F'j. k= 0 The gilt prices provided by the Bank of England are clean prices: that is, they do not include accrued interest. For these calculations, accrued interest was added to the clean prices in proportion to the number of months that had elapsed since the previous interest payment. Limitations of the IG measures of the expected inflation and real interest rates The decomposition of the nominal long-term bond rate into its expected inflation and real interest rate compo nents will be strictly correct if two requirements are met: (1) investors are risk neutral, and (2) nominal and indexed-linked bonds are identical in all respects other than indexation (specifically: risk, maturity, liquidity, and tax treatment). If investors are not risk neutral, interpret ing the IG measure of expected inflation and the corre sponding expected real yield is complicated by the existence of risk premia. Suppose, for expository pur poses, that the indexed bond is fully indexed so that the ex ante real interest rate on the IG can be determined from its price alone. In this case, the expected real yield on the nominal bond will differ from that on the indexed bond by an inflation risk premium. Consequently, the estimate of expected inflation will be contaminated by an inflation risk premium that may vary systematically with expected inflation. It is difficult to predict the sign of this correlation on a priori grounds or to separate the con taminated measure of expected inflation into its two com ponents because standard models of asset pricing under uncertainty fit real-world data very poorly. A further prob lem arises because holders of IGs are not fully compen sated for inflation; the expected real return on the IG will also contain an inflation risk premium if investors are not risk neutral; This premium is likely to be negligible unless the IG is close to maturity, because the investor’s exposure to inflation depends on the change in the RPI during the last eight months of the IG’s lifetime. For a long-dated IG, this change is likely to have only a small impact on the average ex post real return over its life time. Thus, it would be safe to assume that the measure of expected inflation is much more likely to be contami nated by a risk premium than is the expected real interest rate. If the two bonds used for the calculations differ in their liquidity— say the indexed bond is less liquid— the mea sure of expected inflation will be further contaminated by a liquidity premium. As noted in the text, IG turnover is small relative to the turnover of conventional gilts, and thus IGs are presumably less liquid overall than conven tional gilts. Nevertheless, because this general state ment may not necessarily apply to a particular maturitymatched pair of index-linked and conventional gilts, it is difficult to determine whether a specific IG pays a liqui dity premium. Finally, although index-linked gilts enjoy tax advantages compared with conventional gilts, the procedure for decomposing nominal yields ignores tax effects for two reasons. First, the difference in tax treat ment may not be of much practical import because of the dominant role that tax-free institutions play in the gilt market. Second, Gaske found that taking tax effects into account did not change results qualitatively. FRBNY Q uarterly Review/Autum n 1991 in te re st rate to ch an ge s in m o n e ta ry policy. B ritish m on e ta ry po licy was tig h te n e d very sharply in the m iddle of 1988 and rem ained re strictive until S e p te m b e r 1990. T his tig h te n in g is reflected in the thre e -m o n th in te rb a n k rate, w h ich rose by m ore than 700 basis points from May 1988 to the end of 1989 and rem ained in the n e ig h b o rh o o d o f 15 p e rc e n t u n til S e p te m b e r 1990 (C h a rt 1, low er panel). The term structure, inverted since m id-1988, also show s the effects of m o n e ta ry tig h tn e s s . T he sharp slo w in g of the U.K. econom y in 1990 te n d s to confirm th a t high nom inal rates did in fact reflect m o n e ta ry tig h tn e s s ra th e r than a mere run-up of 54 FRBNY Q uarterly Review/Autumn 1991 interest rates in a n tic ip a tio n of a c c e le ra tin g in fla tio n ; it also su g g e sts th a t real rates p ro b a b ly rose from 1988 to 1990. However, th e d e c o m p o s itio n of th e lo n g -te rm bond yie ld in d ica tes th a t th is m o n e ta ry tig h te n in g did not raise expected real lo n g -te rm in te re s t rates by m uch or, more im plausibly, did not low er e xp e cte d in fla tio n at a ll .13 In fact, the d e c o m p o s itio n la rg e ly a ttrib u te s the 13The real rate did rise som ewhat from m id-1988 to m id-1989 and again from February 1990 to S eptem ber 1990. However, the average real rate during the tw enty-nine months of po licy tigh tening was actually 30 basis points lower than the average real rate over the preceding tw enty-nine months. s te a dy rise in lon g-te rm g o vernm ent bond yie ld s from e a rly 1988 to m id-1990 to an increase in expected in fla tio n. It is d iffic u lt not only to reconcile this pattern of real and nom inal interest rate m ovem ents w ith the c o n ve n tio n a l view that m o n e ta ry p o licy affects real e c o nom ic a c tiv ity th rough lo n g -te rm real in te re st rates, but also to believe that private se c to r long-term inflation e x p e cta tio n s were not a d ju ste d dow nw ards in the face of a ve ry resolute tig h te n in g of p o licy .14 These stylize d facts s u g g e st that, at least fo r the 1988-90 period, the IG m easure was a poor m easure of expected inflation and tha t the co rre sp o nd in g real interest rate did not a c c u ra te ly reflect ex ante real in te re st rates faced by the private sector. 1996, w h ile the a ctual in fla tio n rate is the p e rce n ta g e change in the RPI over the past tw elve m onths. C h a rt 2 g e n e ra lly c o n tra d ic ts a s s e rtio n s th a t the IG m easure of expected in flation sim p ly m im ics the b e h a vio r of actual in fla tio n .15 The IG m easure of e xp e cte d in fla tio n has rem ained above the actual rate fo r m ost of th e sam ple period and is also ra th e r less v o la tile than a ctual in fla tion. It has, however, been fa irly c lo se to th e a ctual rate from late 1989 onw ards. A lthough the IG m easure of exp e cte d in fla tio n a p p lie s to an interval th a t is lo n g e r th a n the s h o rt- to m edium term horizon of p rim a ry co n ce rn to p o lic y m akers, it may n e ve rth e le ss p rovide a good fo re ca st of in flation over a h orizon of im m e d ia te p o lic y in te re st. To assess th is p o ssib ility, the IG m easure of e xp e cte d fu tu re in fla tion was com pared w ith th re e sim p le m e a su re s based on past in fla tio n: the average rates of RPI in fla tio n over the past tw elve, tw enty-four, and th irty -s ix m o n th s .16 Each m easure of exp e cte d in fla tio n was e valuated as a The IG m e asure o f e x p e c te d in fla tio n as a p re d ic to r o f in fla tio n C h a rt 2 show s the IG m easure of expected inflation along w ith actual inflation m easured by the tw e lve m onth perce n ta g e cha ng e s in the retail price index (RPI). N ote that the expected inflation rate on a p a rtic u lar date is the average annual rate from that date to late 1sSee, for exam ple, Anthony Harris, “ Lessons from the Indexed D ecad e,” Financial Times, A pril 29, 1991. 14E xpectations of rising inflation could c o incid e with a m onetary p o licy tigh tening if the tigh tening occurred at the same time as an exogenous increase in dem and. 16A measure based on a sho rter period of past inflation was not used because the RPI is not available on a seasonally ad ju sted basis. Chart 2 Actual versus Expected Inflation Percent \ \ w * . \ \ \ \ A V \ \ \ i\ l\ i \ 1 1 \ • A , \ \ i r. * V 7 * v\ ^ » ft1 ’ « * j K 7 " V - \ I A V \ ^ i \ \ Expected inflation i I V VV .// * \ 11 y \ / V* f rr \ \ I I VJ Actual inflsition Ll l I i i I i i, l i i 1982 illllllllll Ll l u l l 1 111 1983 1984 m I i i I i i 1985 I h I I 1 I I I I I I LL I I l - l l l l i l I L -11 I n 1 I I I l l 1986 1987 1988 II I I I In 1989 ll 1 I l l l l l l l l l l 1990 1, 11 1991 Sources: Bank of England data and author’s calculations. Notes: Actual inflation is measured by the percentage change in the retail price index over the preceding twelve months. Expected inflation is the measure derived from prices of index-linked gilts. FRBNY Q uarterly R eview/Autum n 1991 55 fo re ca st of inflation over three horizons: tw elve, tw entyfour, and th irty -s ix m onths. The results of th is exercise, repo rted in Table 3, rely on tw o yardsticks of forecast a ccu ra cy: the root m ean squared forecast error, w hich m easures average p redictive accuracy over the forecast pe riod, and the regression co e ffic ie n t of actual inflation on the m easure of expected inflation, w hich m easures b ia s— th a t is, the te n d e n cy to over- or u n d e rp re d ict persisten tly. The average rate of inflation over the past tw elve m onths is the m ost a ccurate predictor, on aver age, of the level of in flation over all three horizons co nsid ered (it has the sm a lle s t root mean squared fo re cast error). A lth ough by no m eans unbiased, it has the sm a lle st bias; the regression co e fficie n t of actual in fla tio n on this m easure is clo s e r to unity at the one- and tw o -ye a r h orizon s than are th o se on the o th e r m ea sures. The p erfo rm ance of the IG m easure of expected in fla tio n is s ig n ifica n tly b e tte r (over all three horizons) than th a t of the average in flation rate over the past th irty -s ix m onths and s im ila r to that of the average in fla tio n rate over the past tw e n ty-fo u r m onths. Note, however, th a t the IG m easure is o nly m o derately in fe rio r to the best of the auto re g re ssive m easures (the tw e lv e m onth m easure). For exam ple, the root m ean squared fo re cast errors ind ica te th a t a bout tw o -th ird s of actual inflatio n rates over a o n e -ye a r horizon w ill lie no fu rth e r than 2.42 p erce ntag e p o in ts from the rate predicted on the basis of the past tw elve m onths’ inflation, and no fu rth e r than 2.79 pe rce n ta g e points from the rate pre d icte d by the IG m easure. It should com e as no su rp rise that the IG m easure fares no b e tte r than recent in fla tio n in fo re ca stin g fu ture inflation. As C h a rt 2 show s, the IG m easure is a biased p re d icto r of in fla tio n, e xce e d in g the a ctual in fla tio n rate over m ost of the sam ple p e rio d .17 A p la u s ib le in te rp re ta tion of the upw ard bias in the IG m easure of in flation e xp e cta tio n s in the e a rly p a rt of the sa m p le is th a t the U.K. m o n e ta ry a u th o ritie s were not c re d ib le in the early eig h ties. A fte r a long p e riod of fa irly high in fla tio n, several years of low in fla tio n m ight have been n e c e s sa ry to c o n vin ce m a rke t p a rtic ip a n ts th a t th e m o n e ta ry a u th o ritie s w o u ld m a in ta in n o n in fla tio n a r y p o lic ie s . A lternatively, the bias may sim p ly reveal th a t the IG m easure is flawed. One m ight ask w h e th e r the IG m easure, if p u rged of bias, w ould p re d ict in fla tio n over a p o lic y -re le v a n t h o ri zon more a ccu ra te ly than w ould naive m e asures based on past inflation. U sing the m o n th ly ch a n g e in the IG m easure, rather than its level, to fo re ca st fu tu re in flation 17The poor forecasting perform ance of the IG measure relative to autoregressive measures is also to be exp ected from a purely statistical view point. The RPI inflation rate is nonstationary; that is. changes in the inflation rate tend to be perm anent and consequently the inflation rate does not tend to return to its longrun average after a change. U nder these con ditions, inflation over the recent past will generally be the best sim ple p re d icto r of future inflation. Note, however, that using past inflation as a p re d icto r of (the level of) inflation over longer horizons results in large expectations errors. That is, the variance of the forecast error, conditional on inform ation available at the tim e the forecast is form ed, is proportional to the length of the forecast horizon. Similarly, if the IG measure were an accu rate p re d icto r of future inflation, it would tend to move very closely with current inflation. Thus, from a purely statistica l view point, it m ight be considered surprising that the IG measure does not respond one-to-one to changes in actual inflation. Table 3 Comparison of Indexed Gilt Measure and Naive Autoregressive Measures of Expected Inflation Root Mean Squared Forecast Errors (Percentage per Year) Forecast Horizon Measure of E xpected Inflation IG measure Inflation over past twelve months Inflation over past tw enty-four months Inflation over past thirty-six months One Year Two Years Three Years 2.79 2.42 2.81 3.58 2.73 2.41 2.77 3.57 2.78 2.52 2.77 3.57 Regression C oefficient of Actual on Expected Inflation Forecast Horizon Measure of E xpected Inflation IG measure Inflation over past twelve months Inflation over past tw enty-four months Inflation over past thirty-six months One Year - 0 .0 7 0.26 -0 .0 2 - 0 .1 8 Two Years - 0 .3 7 0.00 - 0 .2 4 - 0 .2 0 Three Years - 0 .4 4 -0 .3 2 - 0 .2 8 - 0 .1 6 Note: Sam ple pe riods are as follows: M arch 1982 to July 1990 for one-year forecasts, M arch 1982 to July 1989 for tw o-year forecasts, and M arch 1982 to July 1988 for three-year forecasts. http://fraser.stlouisfed.org/ 56 FRBNY Q uarterly Review/Autumn 1991 Federal Reserve Bank of St. Louis w ill e lim in a te bias th a t rem ains co n sta n t over tim e. But if the bias is in fact due to c re d ib ility problem s, it has pro bab ly d ecreased over tim e, and co n se q u e n tly som e bias is like ly to rem ain in the data. N e vertheless, unless one know s the process w hereby m arket p a rtic ip a n ts change th e ir view s on the c re d ib ility of the m onetary a u th o ritie s , any m e th o d of e lim in a tin g bias w ill be im perfect. The resu lts in Table 4 suggest th a t w hen purged of bias in th is manner, the IG m easure does m a rg in a lly b e tte r than sim p le autoregressive m easures in p re dicting futu re a cce le ra tio n and d e ce lera tio n of inflation. The ta ble co m pares the perfo rm a n ce of the m o nth ly ch ange in the IG m easure in forecasting the change in the RPI infla tion rate over the fo llo w in g tw elve m onths w ith th a t of fo u r naive autoregressive m easures: the ch a nge s in the RPI in fla tio n rate over the preceding on e -, th re e -, s ix-, and tw e lv e -m o n th p e rio d s .18 The m onthly cha nge in the IG m easure is not only the m ost a ccurate predictor, on average, of the RPI inflation rate (it has the sm a lle st root m ean squared forecast error), but it is also a p p re cia b ly clo s e r than the naive m ea sures to offe ring an unbiased fo re ca st of changes in the RPI in flation rate (the regression co e ffic ie n t of actual on p redicted ch ange s is the c lo s e s t to u n ity ).19 It should be 18The tw elve-m onth forecast horizon was chosen somewhat arbitrarily for illustrative purposes. A lthough clearly of interest to policy makers, this horizon is not necessarily the most relevant. 19C om paring the a b ility of alternative measures of inflation expectations to predict changes in inflation is also advisable on purely statistica l grounds. Because the inflation rate is nonstationary, the change in the inflation rate contains all the new inform ation pe rtainin g to the future course of inflation. Table 4 Forecast Performance for Changes in Retail Price Index Inflation: Comparison of Indexed Gilt and Naive Measures Criterion of Forecast Accuracy Measure Root Mean Squared Error (Percentage per Year) Regression Coefficient of Actual on Predicted Inflation 2.34 0.69 2.37 0.32 2.57 0.03 2.95 -0 .0 1 3.53 0.03 One-month change in IG Measure One-month change in RPI inflation Three-month change in RPI inflation Six-month change in RPI inflation Twelve-month change in RPI inflation Notes: Changes in RPI inflation are measured as twelve-month changes in the twelve-month percentage change in the RPI. Sample period is April 1982 to July 1990. em phasized, however, th a t the d iffe re n ce in the fo re casting p e rfo rm a n ce of th e IG m easure and th e ch ange in the RPI in flation rate over th e p re ce d in g m onth is so sm all as to be of no p ra c tic a l s ig n ific a n c e . O ver lo n g er data sam ples or d iffe re n t tim e p e rio d s, th e ra n kin g of the tw o m easures could e a sily be reversed. The preceding c o m p a ris o n s of the IG m easure and naive a u to re g re s s iv e m e a s u re s of e x p e c te d in fla tio n were de sig ne d to show w h e th e r th e IG m easure is a be tte r p re d icto r of in fla tio n than are s im p le a lte rn a tive s. The p redictive value of the IG m easure of expected inflation can also be asse sse d by d e te rm in in g w h e th e r it p rovides in fo rm a tio n th a t im proves th e fo re ca stin g a b ility of an a u to re g re ssive m odel based on past in fla tion. The te st results re p o rte d in Table 5 p ro vid e e v i dence th a t the a d d itio n a l in fo rm a tio n c o n trib u te d by the IG m easure, w h ile s ta tis tic a lly s ig n ific a n t, is too m ar ginal to be of p ra ctica l use. The te s t e va lu ate s w h e th e r th e IG m e a s u re o f e x p e c te d in fla tio n ca n p re d ic t changes in the RPI in fla tio n rate, m easured as the m onthly p e rce n ta g e ch a n g e in the RPI, once th irte e n la g g ed c h a n g e s in th e in fla tio n rate (a n d s e a s o n a l dum m ies) have been ta ke n into a cco u n t in fo rm in g the pre d ictio n s. The firs t fo u r lagged va lu e s of m o n th ly changes in the IG m easure of e xp e cte d in fla tio n are jo in tly s ta tis tic a lly s ig n ific a n t at the 2.5 p e rc e n t level, and e ig h t lagged values are jo in tly s ig n ific a n t at the 10 p e rce n t level. However, the a d d itio n of e ig h t lagged changes in the IG m easure o nly ra ise s th e a d ju ste d R2 of the fore ca stin g e q u a tio n from 0.8 to 0.82, to o sm all an im provem ent to be of p ra ctica l im p o rt. T h e se results are not se n sitive to the n u m b e r of lagged c h a n g e s in the inflation rate in the re g re ssio n , a lth o u g h the nu m b er included (th irte e n ) is so m e w h a t a rb itra ry. In a d d ition , th e lim ite d p r e d ic tiv e v a lu e o f th e IG m e a s u re Table 5 Contribution of Indexed Gilt Measure in Predicting Monthly Changes in Inflation Lags of IG Measure of Expected Inflation None One to four One to eight One to thirteen R2 M arginal S ignificance* (F-Test) 0.80 0.82 0.82 0.81 0.024 0.094 0.146 — Notes: In the baseline regression, the m onthly cha nge in the RPI is regressed on thirteen own lags and a set of seasonal dum m ies. Sample period is May 1983 to A pril 1991. TMeasures the highest sig nifica nce level at w hich one can reject the null hypothesis that the num ber of lagg ed changes in the IG measure of exp e cte d inflation does not co n trib u te to forecasting the change in RPI inflation over the next m onth. The confidence level in the null hypothesis is given by 1m arginal significance. FRBNY Q uarterly R eview/Autum n 1991 57 may sim p ly reflect the degree of p re d ic ta b ility of the RPI th a t derive s from purely m e ch a n ica l aspects of its c a l cu la tio n . For exam ple, ch a n g e s in the banks’ base in te r est rates have a fore se e a b le e ffe ct on m ortg a g e interest rates b ecause variable rate m ortg a g e s are the rule in th e U n ite d K in g d o m ; c o n s e q u e n tly , th e s e c h a n g e s a ffe ct the RPI p re d icta b ly th ro u g h the e ffe ct of m o rt gage paym ents on the co st of housing. E x p e c te d in fla tio n a n d e x p e c te d real in te re s t rates as p re d ic to rs o f re a l e c o n o m ic a c tiv ity T he IG m a rket may convey in form ation useful in the fo rm u la tio n of m o n e ta ry p o licy even though the IG m e a sure of expected infla tio n is no more su cce ssfu l in fo re ca stin g in flation than are sim ple m easures based on past in flation. The IG m easure may be an im p e rfe ct p re d icto r of future inflation sim p ly because it fa ith fu lly reflects the private s e c to r’s unrealized e xp e cta tio n s of in fla tio n. In this case, the IG m arket may provide policy m akers w ith a reliable m easure of ex ante real interest rates. If private se cto r sp e n d in g is in te re st-se n sitive , the m arket may also yield in form ation about private se cto r spe nding plans and near-term e conom ic d e ve l opm e nts. In sum , the IG m easure of expected real lo n g term in te re st rates may be a p o te n tia lly valuable in d ic a to r of m acroe co nom ic d evelopm ents. O ne way to te st the accu ra cy of the IG m easure in gaug ing private se cto r in fla tio n e xp e cta tio n s is to d e te r m ine w h e th e r the IG m easure of ex a nte real lo n g -te rm in te re st rates p rovides in fo rm a tio n a b o u t fu tu re real e c o n o m ic a ctiv ity . C o m p a re d w ith m ore d ire c t te s ts such as co rre la tin g the IG m easure w ith su rv e y m e a sures of in flation e x p e cta tio n s, th is te s t has a d is a d v a n tag e : it can only p rovide in fo rm a tio n a b o u t th e a c cu ra cy of the IG m easure as a ya rd s tic k of private s e c to r in flation e xp e cta tio n s to the exte n t th a t private se c to r d e cisio n s are in te re s t-s e n s itiv e . T h is s h o rtc o m in g is also an im p o rta n t a d vantage, however, b e ca u se th e te st m easures the a ccu ra cy of the IG m easure in te rm s of a goal variable of d ire ct in te re st to p o licy m a ke rs. F u rth e r more, the te st ca p tu re s any p o te n tia l le a d in g in d ic a to r role of indexed bond p rice s and as such is of in te re st to p olicy m akers. To d e te rm in e w h e th e r th e indexed g ilt m a rke t p ro vides in fo rm a tio n a b o u t fu tu re real e co n o m ic a ctivity, real GNP grow th is regressed on la g g ed G N P grow th and various nom inal in te re st rates and in fla tio n rate m easures, in clu d in g th o se o b ta in e d from in d e x -lin ke d g ilts. The results from th e se re g re ssio n s are p ro b a b ly best a ppreciated in the co n te x t of the s ty liz e d facts e sta b lishe d by s im ila r re g re ssio n s on U.S. data. In particular, a lth o u g h U.S. real GNP grow th is ty p ic a lly very d iffic u lt to p redict, E strella and H a rd o u ve lis and S tock and W atson have found th a t s h o rt-te rm nom inal in te re st rates and m easures of the slo p e of the term stru ctu re convey in fo rm a tio n a b o u t fu tu re m o ve m e nts in Table 6 Predictive Value of Nominal interest Rates for Real GNP Growth in the United States and the United Kingdom United States Four Lags of S tatistic F LR Real GNP Growth Ten-Year Governm ent Bond Yield Six-M onth C om m ercial Paper Rate 0.99 4.27 0.27 3.61*** 15.04*** 6.48*** 26.00*** United Kingdom Four Lags of S tatistic F LR R* Real GNP Growth 3.45** 14.38*** 0.12 Ten-Year Governm ent Bond Yield 1.20 5.35 Three-M onth Interbank Rate 2.44* 10.65** Notes: Sam ple period for the United States is 1954-11 to 1991-1: for the United K ingdom , 1965-1 to 1991-11. F and LR are the F sta tistic and likelihood ratio s ta tistic for testing the null hypothesis that a pa rticu la r variable has no explanatory power for future real GNP grow th in a regression inc lu d in g the listed variables as regressors. One asterisk denotes significa nce at the 10 pe rcent level; two, sig n ifica n ce at the 5 pe rcent level; and three, sign ific a n c e af the 1 percent level. Interest rates are measured as qu a rte rly averages of m onth-end observations. GNP growth is measured as q u a rte rly pe rce n ta g e cha nges seasonally ad ju sted at an annual rate. 58 FRBNY Q uarterly Review/Autumn 1991 real o u tp u t .20 Table 6 provides ben ch m a rk regression results illu s tra tin g the p red ictive value of nom inal sh o rtand lo ng-term in te re st rates fo r U.S. real GNP grow th. It also show s that nom inal in te re st rates are less in fo r m ative in the U nited K ingdom than in the U nited S tates. R esults fo r the U nited K ingdom spa n n in g the period s ince the in ce p tion of the in d e x-lin ke d m arket are pre se n ted in Table 7. The ta b le illu stra te s how adding d iffe re n t va ria b le s in an eq u a tio n to forecast real GNP g row th a ffects the a d justed R2 of the equation. Note th a t the co m p o n e n ts of the real in te re st rate are added separately, in p a rt because the hyp o th e sis that the nom inal rate and the inflation rate have co e fficie n ts equal but o p p o site in sign is g e n e ra lly rejected by the data. The regression results s u p p o rt three sp e cific co n clu sio n s: First, U.K. real G NP grow th, like its U.S. c o u n te rp a rt, is hard to predict, but in c o n tra st to the U.S. e xpe rien ce , long- and s h o rt-te rm nom inal interest rates (and by im p lica tio n m easures of the slope of the term structure) do not fore ca st future m ovem ents in real GNP (co lu m ns 1 and 2). S econd, the d e co m p o sitio n of lo n g -term nom inal rates based on indexed g ilt prices provides no s ig n ifica n t in fo rm a tio n about future real GNP grow th in the U nited K ingdom . (The ad ju ste d R2 of the fore ca stin g eq ua tio n a ctu a lly falls w hen the IG m ea sure of expected inflation is added to the forecasting e q u a tio n .) Finally, backw a rd -lo o kin g m easures of sh o rta n d lo n g -te r m re a l in te r e s t ra te s , o fte n u se d as 20See A rturo Estrella and Gikas Hardouvelis, “ Possible Roles of the Yield Curve in M onetary Policy,” Interm ediate Targets and Indicators for M onetary Policy, Federal Reserve Bank of New York, New York, 1990; and James H. Stock and Mark W. Watson, “ The Business C ycle Properties of Selected U.S. Economic Time Series, 1959-1988,” National Bureau of E conom ic Research, Working Paper no. 3376. Table 7 Predictive Value for Future Real GNP Growth: Comparison of Indexed Gilt and Other Variables Regression Number Four lags of Real GNP growth Long-term government bond yield Three-month interbank rate IG measure of expected inflation Twelve-month change in RPI R2 Standard error (1) X (2) X (3) X (4) X X X X x X X X 0.16 2.69 0.17 2.67 Note: Sample period is 1983-11 to 1990-IV. 0.14 2.72 X 0.55 1.97 in d ica tors of the s ta n ce of m o n e ta ry policy, do yield s ig n ific a n t in fo rm a tio n a b o u t fu tu re real e co n o m ic a c tiv ity. Four lags of the th re e -m o n th in te rb a n k rate to g e th e r w ith fo u r lags of the RPI in fla tio n rate (the p e rce n ta g e change in the RPI over the p re ce d in g tw elve m onths) raise the ad ju ste d R2 of a re g re ssio n of real GNP grow th on four lagged values of real G NP grow th from 0.16 to 0.55 (colum n 4). The results re p o rte d here show th a t the p rice s of in d e x-lin ke d g ilts do not convey p o lic y -re le v a n t in fo rm a tio n a b o u t fu tu re tre n d s in e c o n o m ic a c tiv ity . If we a ccept th a t private s e c to r d e c is io n s are s e n s itiv e to real in te re st rate m ovem ents, the re su lts im p ly th a t the IG m easure of ex ante real lo n g -te rm in te re s t rates does not a c cu ra te ly reveal real in te re s t ra te s fa ce d by the private sector. T he lim ita tio n s of the IG m easure of expected in flation as a gauge of private s e c to r in fla tio n e xp e cta tio n s co u ld be due to any of th re e ca u se s. F irst, lim ite d p a rtic ip a tio n in the U.K. indexed g ilt m a rke t may have m ade the IG real in te re s t rate relevant to o n ly a v e ry sm a ll p a rt of th e p riv a te se cto r. S e c o n d , th e expected rate of in fla tio n and the c o rre s p o n d in g real in te re st rate in the bond m a rke t m ay not be relevant to the m a jo rity of p a rtic ip a n ts in th e g o o d s and fa c to r m arkets. Finally, the p oor p e rfo rm a n c e of the IG m e a sure of expected in fla tio n m ay d e rive from ta x d is to r tio n s or the fa ct th a t m a rke t p a rtic ip a n ts are risk averse. Two caveats to th e se c o n c lu s io n s d e s e rv e m en tio n, however. First, the p re d ictive pow er of IG p rice s w ill depend on the m o n e ta ry p o lic y rule fo llo w e d by the au th o ritie s. If the B ank of E ngland were s ta b iliz in g real interest rates, one w ould not e xp e ct the IG m easure of th e real in te re s t ra te to have p re d ic te d re a l G N P c h a n g e s . S e c o n d , th e c o n c lu s io n s a re te n ta tiv e , because the U.K e xp e rie n ce w ith indexed b onds is co m p a ra tively sh o rt. The a d d itio n of o n ly a few ye a rs’ data may ve ry w ell lead to c o n c lu s io n s m ore favor able to the p o sitio n held by p ro p o n e n ts of indexed bonds. Conclusion T his a rtic le has used data from th e U.K. m a rke t for in d e x-lin ke d g ilts to assess the a lle g e d p o lic y b e n e fits of indexed bonds. It has been su g g e ste d th a t a re a l tim e m arket m easure of exp e cte d in fla tio n (and the co rre sp o nd in g ex ante real in te re st rate) d e rive d from indexed bond p rice s co u ld p rovide the F ederal R eserve S ystem w ith va lu a ble in fo rm a tio n a bout m a rke t p e rc e p tio n s of, and reaction to, its p o licie s, and convey in fo r m ation about fu tu re in fla tio n and real e c o n o m ic d e v e l o p m e n ts . T h e e v id e n c e p re s e n te d in th is a rtic le , however, su g g e sts th a t th e p rice s of in d e x -lin k e d g ilts may not convey m uch in fo rm a tio n a b o u t fu tu re in fla tio n and real e co n o m ic activity. For th is reason, a u th o ritie s FRBNY Q uarterly R eview/Autum n 1991 59 may question whether a real-time market measure of expected inflation can shed light on private sector reac tions to monetary policy. The ability of the IG measure of expected inflation to anticipate future inflation developments appears to be, at best, mixed. It is a biased predictor of future inflation, fares no better than simple inflation expectations mea sures based on past inflation, and does not add appre ciably to the predictive power of a more sophisticated backward-looking model of inflation expectations. U.K. indexed bond prices also do not seem to convey policy-relevant information about future real economic 60 FRBNY Quarterly Review/Autumn 1991 activity. This finding is consistent with the IG measure’s being an imperfect gauge of private sector inflation expectations and with the failure of the corresponding real interest rate to reflect accurately ex ante real inter est rates faced by the private sector. The behavior of the IG measures of expected inflation and the real long term interest rate during the period of restrictive mone tary policy from mid-1988 to late 1990 further supports this judgment. In sum, these results suggest that the U.K. IG measure of inflation expectations seems to offer only limited, if any, information for the conduct of mone tary policy. Tracking the Economy with the Purchasing Managers’ Index by Ethan S. Harris In the last several years the purchasing managers’ index has emerged as a key indicator of manufacturing activity. This “index” consists of five separate indexes measuring monthly changes in manufacturing output, employment, new orders, inventories, and vendor deliv eries, together with a composite index that gives a weighted average of the other five. Financial markets are now quite sensitive to the index, and news reports on the economy regularly feature it. The index receives such close attention for several reasons: it is the first broad indicator released each month, it covers the cyclically sensitive manufacturing sector (Chart 1), and the data are easy to interpret and are virtually never revised. Despite the index’s popular appeal and market-moving power, some skepticism about the utility of this indicator is warranted. It is not constructed with the scientific sampling and statistical methods that underlie most official macroeconomic series (see appendix). A qualitative measure of activity, it reports whether busi ness has increased or decreased but makes no assess ment of the strength of the change. Most important, the index has not been rigorously tested: although there is ample evidence that the index tracks the general ups and downs of the economy, analysts have not demon strated that the purchasing managers’ data yield infor mation on the economy beyond that already provided by other indicators. This article analyzes the strengths and weaknesses of the index as a forecasting tool. It begins by explain ing how the index is constructed. The next section presents the basic correlations between the five compo nent indexes and the economic aggregates they are supposed to track. The remainder of the article investi gates the predictive power of the purchasing managers’ data: Do the indexes lag or lead economic activity? Do they foreshadow turning points in the business cycle? Can the indexes improve on the forecasts of simple economic models or on consensus forecasts? Our results give mixed support for the purchasing managers’ index. One shortcoming is the index’s ten dency to pick up activity in the weeks preceding the month it is supposed to measure. Another limitation is that none of the components explains more than half of the monthly variation in the corresponding official statis tics. Furthermore, the index is not a reliable leading indicator: it sends too many false signals and its lead time is too erratic to be of use in anticipating cyclical swings. Nevertheless, the index does add significantly to the explanatory power of simple econometric models and consensus forecasts. And it could be even more useful to forecasters if the sampling and statistical methodology were improved. Thus, although the index has some important limitations, with careful application it can be useful in forecasting economic activity. Description About the middle of each month the National Associa tion of Purchasing Managers (NAPM) surveys roughly 300 association members representing twenty-one manufacturing industries in all fifty states. The survey asks each purchasing manager how the current level of five key economic indicators— production, new orders, employment, inventories, and vendor delivery time— FRBNY Quarterly Review/Autumn 1991 61 c o m p a re s w ith th e p re v io u s m o n th ’s le v e l .1 T h e respon se s are sim p ly “ higher,” “ lower,” or “ the sa m e .” T he unw eighted p erce n ta g e of firm s in each ca te g o ry is then ta b u la te d and a d iffu s io n index is c o n stru cte d by su m m in g the p e rc e n ta g e of p o sitive re sp o n se s and o n e -h a lf of those responding “ the s a m e .”2 A reading above 50 percent in a d iffu s io n index m eans that more firm s are expan ding a c tiv ity than co n tra ctin g activity. Finally, th e se data are se a so n a lly adjusted and co m bined into a sin gle w e ig h ted com p o site index. A lth o u g h the surve y has been published since 1931 (w ith an in te rru p tio n fo r W orld War II), several of its m ore so p h istica te d features were only introduced in recent years. The data were o rig in a lly p u blished in raw, se a so n a lly u na djuste d form ; in the early 1980s, w ith help from the C om m erce D epartm ent, the asso cia tion began p ub lish in g se a so n a lly adjusted d iffu sio n indexes. T he sam p le size has also been increased to a lm ost 300 1The survey also includes questions on com m odity prices and buying policy. In the last several years new export orders and im ports have been added. 2The NAPM survey treats vendor de live ry time somewhat differently. The responses for this in dica tor are “ slower,” “ faster,” and “ no cha n g e ." The diffusion index for vendor deliveries is the sum of the percentage reporting slower delivery time and half the percentage reporting no change. 62 FRBNY Q uarterly Review/Autumn 1991 from about 225. S ince th e su m m e r of 1989, w hen fin a n cia l m a rke ts b e ca m e in c re a s in g ly in te re s te d in th e index, the survey has been released e a rlie r and at the sam e tim e each m o n th . It now u s u a lly “ b e a ts ” th e em p lo ym e n t re p o rt by several days and th u s c a p tu re s m axim um a tte n tio n in the m a rk e t .3 The index as a measure of economic activity The NAPM co m p o n e n t indexes have c o u n te rp a rts in official data p u b lish e d by the fe d e ra l g o ve rn m en t. S ince the indexes are m easures of the d iffu s io n of the e c o nom ic activity, they sh ould have ro u g h ly a lin e a r re la tio n s h ip to the g ro w th in co rre s p o n d in g g o ve rn m e n t d a ta .4 In oth e r w ords, if a h ig h e r p ro p o rtio n of firm s are re p o rtin g e xp a n d e d a c tiv ity , th e n we w o u ld e xp e ct hig h er grow th in a g g re g a te activity. Table 1 presents evid e nce of how c lo s e ly the NAPM data track the econom y. The ta b le show s the re sults of regressing the p e rce n t ch a n g e in the o fficia l d a ta on the co rre sp o nd in g c o m p o n e n t of the index. T he firs t three colum ns present e stim a te s using m o n th ly d a ta fo r the 1959-91 p e rio d . As th e t-s ta tis tic s (in p a re n th e s e s ) show, the NAPM c o e ffic ie n t is s ig n ific a n t in all of the equations. The overall fit (R -square), however, is g e n e r ally m odest; the indexes e xplain less than h a lf of the m onthly variation in grow th fo r all v a ria b le s . T he w e a k est results are for new o rd e rs and p rice s, tw o h ig h ly vo la tile se rie s; the best re su lts are fo r e m p lo ym e n t. The fo u rth and fifth co lu m n s of Table 1 show the overall fit w hen q u a rte rly and ann u a l d a ta are used. A lthough this tim e a g g re g a tio n g e n e ra lly im proves the fit, the NAPM data still leave a good p o rtio n of the variation in grow th u n e xp la in e d . T he last tw o c o lu m n s of Table 1 show th e im p lie d b re a k -e v e n p o in t fo r the indexes. T h e o re tica lly, w hen a g g re g a te grow th is zero, a d iffu sio n index sh ould average out to a b o u t 50 per cent, w ith equal nu m b e rs of firm s re p o rtin g h ig h e r and lower activity. The regression e stim a te s s u g g e st th at using 50 p e rce n t as the break-even p o in t can be m is leading. For exam ple, in the regression of in d u stria l production on the c o m p o s ite index, e s tim a te s for the 1980-91 period show a break-even rate of ju s t 46.9 p e rc e n t .5 in te re s t in the index has also drawn attention to som e of the regional purchasing m anagers’ surveys. The C hicag o index is closely watched, in part because it is released before the national index. ♦Specifically, if (1) firm s have id en tical but no nsynchronous cycles and (2) growth is evenly d istrib u te d am ong large and sm all firms along a rectangular d istrib ution , then there will be an exact linear relationship between the proportion of firm s exp andin g and the rate of growth of aggregate activity. Regression tests found no evidence of significant nonlinearities. 5Recent experience illustrates the d a nger of using 50 pe rcent as the break-even point. From May 1989 to A pril 1990 the com posite NAPM index dip p e d below 50 percent, averaging 47.6 percent, If The index as a leading indicator pe rce n t as the e co n o m y slip s in to re ce ssio n . E m p irical w ork by C ox and Torda show s th a t the c o m p o s ite index "reached its c y c lic a l p eak a b o u t 1 1 1/2 m onths before the on se t of the seven po stw a r re ce ssio n s” and th a t “ the lead tim e of the c o m p o s ite index of le a d in g e co n o m ic in d ica tors is sim ilar, a b o u t 12 m o n th s .”6 C ox and Torda also fin d th a t th e c o m p o s ite in d e x g e n e ra lly le a ds cyclica l recoveries. U nfortunately, average lead tim e is a p o o r c rite rio n for ju d g in g a leading indicator. To be u se fu l, a le a d in g in d ica to r m ust p re d ict tu rn in g p o in ts w ith a re la tive ly regular lead of at least a few m onths. It m u st a lso give a re la tive ly sm all n u m b e r of fa lse s ig n a ls . Here we te st the predictive pow er of tw o ty p e s of m o ve m e n t in the c o m p o site index: tu rn in g p o in ts in the index and p e rio d s w h e n th e in d e x c ro s s e s v a rio u s “ b r e a k - e v e n ” o r “ th re s h o ld ” points. N e ith e r NAPM sig n a l re lia b ly p re d icts b u s in e s s cycle tu rn in g points. As C h a rt 2 show s, the index ofte n tu rn s dow n long before a b u sin ess cycle peak, re fle ctin g the slow ing of grow th fo llo w in g the in itia l c y c lic a l recovery. Even if we ignore th is in itia l peak, the index has m u ltip le peaks in the co u rse of each e x p a n sio n , and th e peak Tracing the gen eral m ovem ents in e conom ic a c tiv ity is not a ve ry rig o ro u s te st of an indicator. M uch of the in te re st in the purchasing m an a g e rs’ index am ong b u s i ness e co n o m ists stem s from its a lleged a b ility to signal c h a nges in eco n o m ic trends. The tre m e n d o u s attention the index now receives sta rte d in the sum m er of 1989 w hen the index, fa llin g below 50 percent, appeared to presage a recession. C le a rly the in d e x’s e arly release m akes it a “ tim e ly in d ic a to r” ; the more d iffic u lt question is w h e th e r it in fact a n tic ip a te s a c tivity in the m onths ahead. D oes it lead a ctiv ity or m easure c o n te m p o ra n eou s a c tiv ity ? And are b u sin ess eco n o m ists correct in a s s u m in g th a t it g iv e s a re lia b le w a rn in g of re cession? The purch asin g m anag e rs’ index, like all d iffu sio n indexes, has leading in d ic a to r q u a litie s. C h a rt 2 show s the re la tio n sh ip betw een the co m p o site index and the grow th in m anu fa ctu rin g o u tp u t over the bu sin ess cycle. The index peaks w hen grow th is highest, d e clin e s to 50 pe rcent as grow th levels off, and then fa lls below 50 Footnote 5 (continued) 50 percent is the break-even point, this drop in the index implies about a 2 pe rcent de clin e in m anufacturing output. In fact, as the regression estim ates predict, output showed no change over this period. 6W illiam A. Cox and Theodore S. Torda, "Survey By Purchasing M anagers Can Provide Signal On End Of R ecession,” Business A m erica , July 14, 1980, p. 21. Table 1 “ B reak-even” Regressions for Manufacturing Growth Predictive Power (R2) Break-even Point Constant Slope Monthly Q uarterly Annual 1959-91 1980-91 - 3 .6 2 (11.7) - 2 .4 4 (17 9) 0.070 (12,9) .300 .666 .582 51.5 49.3 0.050 (18.4) 465 .772 .681 49.1 47.4 New orders* - 3 .2 2 (4 5 ) .074 .558 .588 51.4 49.3 M aterials inventories - 2 .2 0 (10.2) 0.063 (4.8) 0.049 (1 1 2 ) .246 .512 .239 44.7 47.0 C apacity utilization* 69.24 (89.2) .426 .439 .435 — — Crude produce r prices - 1 .8 9 (3.4) 0.236 (16.9) 0.034 (4.1) .041 .169 .506 56.3 56.1 Industrial production - 3 .3 0 (10.3) 0.067 (11.4) .250 .556 .669 49.0 46.9 Real GNP - 3 .6 0 (7.4) 0.081 (9 1 ) .393 .713 44.4 44.5 Series E xplained Industrial production Payroll em ploym ent With the C om posite Index Notes: Regression c oe fficients are based on the January 1959-M ay 1991 sam ple. Except in the ca p a city utilizatio n equation, the d e pend ent variable enters as a sim ple percentage change. A bsolute t-values are in parentheses. f Sample starts in 1967 and the de p e n d e n t variable is deflated using the im plicit deflator for shipm ents. *The inde pen den t variable is vendor deliveries, lagg ed three months. FRBNY Q uarterly Review/Autum n 1991 63 th a t fin a lly “ c o rre c tly ” sig n a ls recession can o ccu r a n y w here from zero to tw e n ty m onths before the onset of re ce ssion s. The index is ju s t as e rratic in predicting cyclica l trou ghs, botto m in g out anyw here from zero to tw elve m onths before the e co n o m y -w id e tro u g h . If the 50 perce n t th re sh o ld is used ra th e r than the in d e x’s peak, e q u a lly vexing p ro b le m s e m e rg e (Table 2). The index u su a lly drops below 50 before c y c lic a l peaks, Chart 2 The Purchasing Managers’ Index and the Growth in Manufacturing Output Percent 80 ------- Percent 30 r NAPM composite index ^ r ! Manufacturing output 11 Scale---► 20 ii i l i i i l i 1959 60 m 61 1 1 1 1 ! 111 li i ii 111 m l i i l l 62 63 64 65 66 67 11 iliiliiiiiiilllllilit 68 69 70 71 72 73 74 ll i 11h ; iII ll 11111111111 i I ■H , 11 -LL 75 76 77 78 79 80 81 82 83 84 85 i l l 1.11 .1, 86 87 88 -30 89 90 91 Notes: Chart shows quarterly averages of monthly data. Manufacturing output is from the industrial production index. Shaded areas denote recessions. Table 2 Does the Composite Index Signal Business Cycle Turning Points? Lead ( + ) or Lag ( - ) Time in Months NAPM Threshold Peak 50.0 49.0 NAPM Threshold 44.5 O ctober 1949 50.0 N ovem ber 1948 +8 +8 July 1953 +2 +2 -1 May 1954 A ugust 1957 +5 +5 -2 A pril 1958 -2 A pril 1960 + 1 + 1 - 1 February 1961 D ecem ber 1969 -1 - 1 -9 N ovem ber 1973 -1 0 -1 0 -1 1 +5 +2 July 1981 0 July 1990 + 14 January 1980 Average False alarm s 0 Through 49.0 44.5 + 1 + 1 +2 0 0 +2 -1 -2 -2 -1 November 1970 -3 -3 -1 March 1975 -5 -5 -3 -2 July 1980 -2 -2 -1 0 -2 November 1982 -3 -3 -2 0 -3 + 2.7 + 0.8 -3 .4 Average -2 .0 -1 .9 -0 .5 4 4 1 0 0 1 False alarms -1 Notes: In keeping w ith the leading indica tor literature, the com posite index is assum ed to signal a turning point when it crosses the threshold value for three or more consecutive months. The signal is dated from the first m onth the threshold is crossed. S ignals reversed for at least three m onths before a c y c lic a l turning point occu rs are considered “ false alarm s." 64 FRBNY Q uarterly Review/Autumn 1991 but the lead time is quite variable. In the 1973-75 recession, the index did not signal recession until almost a year after the onset of the downturn; by con trast, in the most recent recession, the index stumbled along at just below 50 percent for more than a year before the economy turned down. Even worse, it falsely predicted four business cycle peaks, with several sig nals lasting as much as a year. Its record for cyclical troughs is equally dismal: the index usually surpassed the 50 percent break-even point two to five months after the economy had moved out of recession. Similar problems arise when threshold values below 50 percent are used. The findings in Table 1 suggest a 49 percent break-even value for industrial production and a 44.5 value for real GNP. Using 49 percent rather than 50 percent as the break-even value has virtually no impact on the timing of the signal. Using 44.5 percent changes the results, but not for the better. At cyclical peaks, the index falls below 44.5 percent after the turning point in all but one downturn, with an average lag of three months. At cyclical troughs, the signal is a little more timely, but again it usually fails to anticipate the recovery. The only advantage of the 44.5 percent threshold is that it produces only two false signals in the postwar period. Thus the composite index has two problems as a leading indicator. First, because business cycles do not follow a smooth growth pattern, the index often peaks during the initial recovery and then reaches several mini-peaks in the course of an expansion. Second, because growth usually does not flatten out gradually at the peak of the business cycle, the composite index may not dip below 50 percent until after the recession starts. Furthermore, as noted in the appendix, the NAPM data may lag economic activity by about half a month because respondents have incomplete data on the current month when they fill out the survey. There fore, as a cyclical indicator, the index is better used to confirm recent turning points than to anticipate them. Three horse races Clearly the composite index and its components have important limitations as stand-alone indicators of the strength of the economy. This section tests how the indexes perform in comparison with alternative forecast ing tools. In particular, the analysis explores how the indexes stack up against economic models and consen sus forecasts in explaining the growth in nonfarm payroll employment, industrial production, and real GNP. The results confirm that the indexes are poor stand-alone predictors, but they also demonstrate that the indexes provide helpful incremental information to forecasters. In other words, the indexes represent an imperfect but useful addition to our knowledge of cur rent economic conditions. Forecasting nonfarm em ploym ent growth In the last several years the employment report has become the most important economic indicator for datawatchers.7 Recognizing this, the purchasing managers’ survey committee has pushed up the release date for the index so that it now usually precedes the employ ment report. Not surprisingly, the composite index and its employment component are viewed as vital informa tion in the payroll employment guessing game. The explanatory power of the index is tested against two standards. First, the predictions of a simple eco nomic model of employment growth are compared with those of the NAPM data. Second, the performance of the NAPM data is measured against that of the consen sus forecast reported by Money Market News Service. The informal economic model used here is con structed from variables available to forecasters before the purchasing managers’ data are released. These include several interest rate spread variables identified in work by Bernanke, Estrella and Hardouvelis, and others as reliable predictors of economic activity.8 Spe cifically, the model includes the six-month commercial paper rate, the spread between the commercial paper and Treasury bill rates, the spread between corporate BAA bonds and ten-year Treasuries, and the difference between ten-year and three-month Treasury rates. Also included are several “real” variables watched by payroll forecasters: domestic auto sales, initial claims for unemployment insurance, and the index of leading eco nomic indicators. All told, this ad hoc economic model has eight explanatory variables. The four interest rate variables are entered contemporaneously and with six lags, autos and claims enter currently and with a lag, and both the index of leading indicators and the depen dent variable enter with six lags. Adding the NAPM employment index to this model yields a rigorous test of its incremental explanatory power.9 Table 3 compares the explanatory power of the eco nomic model, the NAPM employment index, and the full 7The m arkets a p p e a r to have a “flavor of the m onth" a p p ro a c h to econom ic indicators, with m erc h a n d is e trad e, c onsum er prices, producer prices, m oney growth, and the em p loym ent report eac h getting top billing at various tim es. O verall, however, em ploym ent seem s to be the most consistent leader. 8Ben S. B ernanke, “On the Predictive Power of Interest R ates and Interest Rate S preads," Federal R eserve B ank of Boston N e w E n g la n d E c o n o m ic R evie w , N o v em b e r-D e c e m b er 1990, pp. 51-68; Arturo Estrella and Gikas A. H ardouvelis, “The Term Structure as a Predictor of Real Econom ic Activity," Federal R eserve B ank of New York, R esearch P aper no. 8 9 0 7 , M ay 1989. ®Each m odel was also tested using the N APM com posite index and using m anufacturing em ploym ent as the d e p e n d e n t variab le, and the results were very similar. FRBNY Quarterly Review/Autumn 1991 65 m odel (th at is, the eco n o m ic m odel com bined w ith the NAPM index) over tw o sam ple periods, one of extended d u ra tio n (1959-91) and the o th e r lim ited to recent years (1980-91). For each m odel the table show s the c o e ffi cie n t on the e m plo ym e n t index w ith its t-s ta tis tic and the overall fit of the m odel as m easured by the adjusted R -s q u a re . S e v e ra l re s u lts are n o te w o rth y . F irs t, alth o u g h both the eco n o m ic m odel and the purchasing m an a g e rs’ index are h ig h ly s ig n ifica n t, the econom ic m od el e x p la in s so m e w h a t m ore of th e va ria tio n in e m p lo ym e n t grow th. T h is finding should not be s u rp ris ing, however, because the NAPM data m easure only grow th in the m anu factu rin g sector, w hile the econom ic m odel has a rich array of e xp la na to ry va riables. S e c ond, and m ore im p o rta n t, w hen the NAPM variable is added to the econ om ic m odel, th is variable co n tin u e s to be h ig h ly sig n ifica n t. In fact, the adjusted R -squares su g g e st th a t the best m odel com bines the NAPM data and the econom ic m o d e l .10 Even stro n g e r su p p o rt fo r the NAPM index com es from c o m p a rin g it w ith th e co n s e n s u s fo re c a s t fo r payroll em plo ym e n t grow th issued by M oney M arket News S e rvice . T his in fo rm a l su rve y of data w atchers is taken ju s t before the NAPM and e m ploym ent data are released. The sam ple is lim ite d to the period since 1985 because of the difficulty in obtaining earlier data. Table 4 10ldeally, it w ould make sense to m odify the estim ation in two ways: (1) sim plify the model by d roppin g the less significa nt lags on each variable and (2) use unrevised data for the independent variables (to d u p lic a te what is available to forecasters). Our purpose here, however, is to stack the od ds against the NAPM index as much as possible rather than to devise an optim al model. Furthermore, prelim inary tests show that the results are not sensitive to either of these changes. Table 3 Explaining the Percent Change in Nonfarm Payroll Employment Sample: 1959-91 Model NAPM* NAPM index 0.022 .423 (16.9) Economic model* Full model§ E2 — .429 0.021 (7.3) .506 show s the results of th is c o m p a ris o n . A g a in , both the NAPM index and the c o n s e n s u s e xplain a large p o rtio n of the variation in e m p lo ym e n t, but th e best re su lts are ob tained w hen the c o n se n su s and N APM are c o m b in e d in the sam e equ a tio n . T h is fin d in g su g g e s ts th a t payroll fo re ca ste rs sh o u ld m o d ify th e ir fo re ca st in lig h t of the NAPM release. For exam ple, all else e q u a l, a 1 p e rc e n t age p oint increase in the NAPM index sh o u ld in d u ce a 10,000 upward revision in exp e cte d p a yro ll e m p lo ym e n t grow th. So fa r we have fo cu se d on the in -s a m p le fit of the various e m p lo ym e n t m odels. T he u ltim a te te s t of th e se equ a tio n s, however, is how th e y perform o u t of sam p le. For each m odel a se rie s of o n e -m o n th -a h e a d fo re ca sts is ca lcu la te d by using data from 1959 to 1984 and then extending the sam ple forw ard one m onth at a tim e. Table 5 show s the relative size of the p re d ictio n erro rs fo r each of the m odels. As w ith the in -s a m p le tests, adding e ith e r the c o m p o s ite or e m p lo y m e n t index to the oth e r m odels reduces the average p re d ictio n errors. The b e s f re s u lt c o m b in e s a s im p le a u to re g re s s iv e Table 4 Explaining Employment Growth with the NAPM Employment Index and the Consensus Model Model NAPM index Consensus model NAPM com bined with consensus m odel Constant NAPM - 0 .7 7 5 (6.0) -0 .0 3 5 (1-6) 0.020 (7.3) - 0 .4 2 9 (4.0) 0.009 (3-7) C onsensus w _ .408 — 1.173 (10.7) .600 0.917 (7.5) .658 Notes: Sample period is January 1985 to May 1991. The de pend ent variable is the pe rcentage cha nge in total nonfarm em ploym ent. C onsensus data are converted from cha nge to pe rcentage change. A bsolute t-values are in parentheses. Sample: 1980-91 NAPM* R2 0.024 .522 (12.2) — .600 0.019 .646 (3.6) *Values are coefficients on the NAPM index, with absolute tvalues in parentheses. ♦Includes the com m ercial paper rate, three interest rate spread variables, auto sales, initial claims, the index of leading indica tors, and lags of the dependent variable, in c lu d e s both the NAPM employment index and all of the econom ic variables. 66 FRBNY Q uarterly Review/Autumn 1991 Table 5 Out-of-Sam ple Prediction Errors for Payroll Employment Growth Model W ithout NAPM _______ With NAPM C om posite Em ploym ent — .157 .145 Autoregressive m odel .143 .141 .136 Econom ic m odel .181 .158 .146 NAPM index only Notes: Table shows the root mean square error for the January 19 85-M ay 1991 period. The “ autoregressive m o del" sim ply uses six lags on the d e p e n d e n t variable. m odel w ith the co m p o site in d e x .11 11Note that the econom ic model alone does the worst job of predicting out of sam ple for this period. This result is consistent with Bernanke's argum ent that a structural shift in the relationship between the spread variables and econom ic activity occurred in the 1980s. Table 6 Explaining Industrial Production Growth Model NAPM index Consensus model Hours Economic model NAPM index plus consensus model NAPM index plus hours NAPM index plus econom ic model Constant NAPM Consensus -2 .7 2 6 (8 4 ) R2 0.055 (9 0 ) — — .371 — 0.926 (12.8) — .544 — — 0,507 (11.4) 488 — — — .618 — .544 0.059 (1.3) 0.027 (4.6) 1.478 (1.6) -0 .4 0 1 (0.9) Hours 0.829 (7.2) 0.178 (4.4) 0.009 (1.1) 0.004 (5.1) -0 .0 7 3 (0.1) 0.030 (2.1) — — 0.401 (8.8) — .568 .631 Notes: Sample period is January 1980 to May 1991. The depen dent variable is the percentage change in the industrial production index. The consensus is from Money Market News Service. Absolute t-values are in parentheses. Table 7 The NAPM Composite Index and Real GNP Growth Model Constant NAPM NAPM index -1 4 .4 6 0 (5.5) 0.546 (0.8) 0.326 (6 6 ) Consensus model Economic model NAPM index plus consensus model NAPM index plus econom ic model Consensus - R2 .361 — 0.832 (4.6) .217 — — .658 -1 2 .0 0 0 (4-2) 0.261 (4.5) 0.363 (1.9) .378 0.506 (0.1) 0.152 (1.7) 11.346 (4.6) — .669 Notes: Sample period is 1970-1 to 1989-11. The dependent vari able is annualized one-quarter growth in real GNP. Absolute t-values are in parentheses. F o re ca stin g in d u s tria l p ro d u c tio n a n d re a l GNP The NAPM data are also useful in fo re ca stin g in d u stria l production and real GNP. Table 6 co m p a re s the e x p la n ato ry pow er of fo u r m o d e ls of in d u s tria l p ro d u c tio n : the NAPM p ro d u ctio n index, the grow th in em p lo ye e hours, the M oney M arket c o n s e n s u s fo re ca st, and an e c o nom ic m odel using the sam e va ria b le s d is c u s s e d in the previous se ctio n . The t-s ta tis tic s on th e N APM c o e ffi cie n ts su g g e st th a t the index adds s ig n ific a n tly to the eco n o m ic m odel and the s im p le em p lo ye e hours m odel, but th a t it is not a useful a d d itio n to the c o n s e n s u s forecast. T h is finding sh o u ld not be a su rp ris e , however, since the NAPM d ata are a va ilable to fo re c a s te rs before the co n se n su s su rve y is ta ke n and th e re fo re sh o u ld already be in co rp o ra te d into the co n s e n s u s fo re ca st. Table 7 show s the re su lts of the final h o rse race. It com pares the pow er of the co m p o s ite N APM index, the eco n o m ic m odel, and a c o n s e n s u s fo re c a s t to p re d ict grow th in real G N P For the e co n o m ic m odel the v a ri ables used are the sam e as th o se in th e e m p lo y m e n t and in d u stria l p ro d u ctio n e q u a tio n s, but each v a ria b le enters co n te m p o ra n e o u s ly and w ith tw o lags. The c o n se n su s data , c o m p ile d by th e A m e ric a n S ta tis tic a l A s s o c ia tio n an d th e N a tio n a l B u re a u of E c o n o m ic R esearch, are o n e -q u a rte r-a h e a d fo re ca sts, ta ke n in the m iddle of the p receding quarter. A g a in , the re su lts of th e c o m p a ris o n are g e n e ra lly s u p p o rtiv e of th e NAPM data. The NAPM index p re d icts real G N P grow th b e tte r than the c o n se n su s fo re ca st, a lth o u g h w orse than the e co n o m ic m odel. The re la tive ly w eak p e rfo rm ance of the co n se n su s is easy to e xp la in : the NAPM and e co n o m ic m odels use u p -to -d a te in fo rm a tio n , w h ile the co n se n su s is based o nly on in fo rm a tio n a vailable before each quarter. A m ore im p o rta n t re su lt is th a t the NAPM index c o n tin u e s to be s ig n ific a n t w hen added to the oth e r m odels (a lth o u g h it is o n ly m a rg in a lly s ig n ifi cant w hen co m b in e d w ith the e c o n o m ic m odel). Conclusion D espite its grow ing p o p u la rity, the NAPM index has undergone ve ry little c ritic a l scrutiny. O u r re su lts s u g gest th a t the index is a flaw ed but still useful indicator. It is a poor leading in d ic a to r and, on its ow n, can be a m is le a d in g m e a su re of s h o rt-ru n m o v e m e n ts in th e econom y. In c o m b in a tio n w ith o th e r data, however, it is very helpful in p re d ictin g co n te m p o ra n e o u s m a n u fa ctu r ing activity. In sum , the index d e se rve s at le a st p a rt of its reputation as a key e co n o m ic indicator. FRBNY Q uarterly Review/Autum n 1991 67 Appendix: The Design of the NAPM Data Set W ith one notable exception, the NAPM data have received high praise in the literature.* Hoagland and Taylor, for example, argue that the survey data “are available sooner, are more reliable, and are much more cost effective than government information.”* Klein and Moore cite the early release of the data as an important advantage; they recommend that the inventory index be substituted for the official inventory data to improve the timing of the index of leading indicators.5 Despite this strong support, the NAPM data need improvement in at least three important areas. Sampling bias Unlike the surveys underlying official statistics, the pur chasing managers’ survey does not use a scientific sam ple. The NAPM data are drawn from hand-picked members of larger, older firms rather than from a proba bility sample. No attempt is made to account for industry growth through the increase in the number of firms. Furthermore, newer, fast-growing firms are added to the sample only after they have become established in the business, while declining firms remain in the sample until they go out of business. In official statistics, both of these downward biases are eliminated through adjustments and rebenchmarking. The sampling design has additional problems. The sample is small, comprising less than 1 percent of the association’s membership. Because of nonresponses and the entry and exit of members, firms answering the survey questionnaire can vary from sample to sample. No attempt is made to correct for this variation by linking companies that respond in both the current and previous month— a procedure followed in the official statistics. Finally, the data are never revised, implying that errors are never corrected and late responses are never incor porated into the data. These sampling problems may explain the apparent downward bias in the indexes. Theoretically, when aggretThe exce ption is Feliks Tamm, “An A genda for Inventories Input to the Leading C om posite Index," in Kajal Lahiri and Geoffrey H. Moore, eds., L eading Econom ic Indicators, (C am bridge: C am b ridg e University Press, 1991), pp. 429-60. Tamm points out a variety of flaws in the NAPM inventory data. Some of his con cerns are discussed here. *John H. H oagland and B arbara E. Taylor, “ Purchasing B usiness Surveys: Uses and Im provements," Freedom o f C hoice: Presentations from the 72nd Annual International Purchasing C onference (O radell, N.J.: National A ssociation of P urchasing M anagem ent, 1987), p. 1. §Philip A Klein and Geoffrey H. Moore, "N .A .P M . Business Survey Data: Their Value as Leading Ind icators," Journal of P urchasing a n d M aterials M anagem ent, W inter 1988, pp. 32-40. FRBNY Q uarterly Review/Autumn 1991 gate activity is unchanged, the indexes should read about 50 percent, with equal numbers of firms reporting higher and lower activity. In fact, as Table 1 in the text shows, the break-even values tend to be well below 50 percent. The results for the inventory index are particu larly troubling. Not only is the break-even point well below 50 percent, but the index also averages only 47.8 percent over the entire postwar period. This finding implies that the level of inventories held by manufactur ing firms has had a downward trend. Government statis tics, on the other hand, show inflation-adjusted materials and supplies for manufacturers roughly doubling over this period.11 Backward-looking data An important attribute of the NAPM data— its tim e liness— is also one of its biggest shortcomings. Since the results are released just after the end of each month, the questionnaire must be answered in the middle of the month. As a result, when respondents try to compare the “current” month with the “previous” month, they may in fact be comparing their impression of the last few weeks (including part of the previous month) with their recollec tions of the weeks before that interval. As the table below shows, the timing of the responses means that in some cases the NAPM data are more closely correlated with lagged activity than with current activity. Subjective responses Survey respondents may not accurately assess whether conditions are “better” or “worse ” Their answers may reflect what should be or what is projected rather than what is. The low average reading for inventories, for !i C om paring the NAPM indexes for em ploym ent and output with the official diffusion indexes for m anufacturing em ploym ent and industrial productio n confirm s this bias. Regression estim ates for the 1980-91 period show that the break-even values for both officia l diffu sion series are closer to 50 percent. Correlation of NAPM Indexes and M anufacturing Data Official Series Lead C ontem porary Lag Industrial production .410 .547 .614* Employment .569 .682 New orders .211 .272 .720* .426* M aterials inventories .481 .496- .476 Notes: The sam ple period is January 1959 to May 1991. The asterisk in dica tes peak correlation. Appendix: The Design of the NAPM Data Set (continued) example, may reflect the constant concern about exces sive stocks rather than actual inventory management. Wishful responses are particularly likely since the sam ple is taken before the full results for the month are known, and many of the questions refer to areas of the firm not under the direct purview of the purchasing manager. The response that economic activity is “the same” is equally problematic. Over time an average of more than half the responses is “the same.” For example, from January 1990 to June 1991 the percentage of “same” responses was: new orders (46.5), production (53.9), inventories (53.1), vendor deliveries (82.3), and employ ment (64.4). Such stability at the firm level seems quite unlikely in an unstable period for the economy as a whole. Apparently, “the same” is a catch-all assessment meaning “don’t know" and “no response” as well as “no change." Improving the data In a real sense the NAPM data set is an uncut gem. By using modern sampling and statistical techniques, the association could greatly improve the accuracy of the data. A probability sample should replace the handpicked sample; respondents should be linked from one sample to the next; efforts should be made to reduce the number of “sam e” responses and to ensure that responses reflect actual activity; and respondents should be encouraged to report on the current month’s activity only. In addition, correctly accounting for inventories and adjusting for lags and leads in the components would improve the composite index.++ Of course, the NAPM data neither could nor should mimic the official statistics: this would require delays in its release and would put an impossible burden on the respondents. The purchasing managers’ association has made some efforts to refine the data. Nevertheless, with the index increasingly in the spotlight, further modernization is warranted. ttThe inventory index should enter the com posite index as a first difference rather than a level since it m easures a stock rather than a flow. In a forthcom ing paper, Mark Flaherty and the author present an alternative com posite index that has an im proved track record in pre d ictin g industrial productio n, real GNP, and the index of c o in cid e n t indica tors. FRBNY Q uarterly Review/Autum n 1991 69 Treasury and Federal Reserve Foreign Exchange Operations May-July 1991 The dollar rose significantly in June and early July, only to ease back during the next few weeks and end the May-July reporting period with little net change. Over the three months as a whole, the dollar rose about 2 percent against the mark, about 1 percent against the yen, and just under 1 percent on a trade-weighted basis.1 Shifting assessments of the strength of economic recovery in the United States were important in stim ulating movements of the dollar exchange rate during the period. In addition, political turbulence in Eastern Europe helped support the dollar against the mark through most of the period, while intervention and evi dence of international cooperation around the time of the Group of Seven (G-7) summit meeting in July was seen in the market as limiting the prospect of a continu ing dollar rise. The U.S. monetary authorities intervened on two occasions to signal an interest in resisting the rise of the dollar, selling a total of $150 million against marks as part of their cooperation with other central banks. The U.S. monetary authorities also engaged in offmarket transactions with foreign monetary authorities, selling $8,548.5 million equivalent of their foreign cur rency reserves for dollars. A report p resen ted by Sam Y. Cross, Executive Vice President in ch a rg e of the Foreign G roup at the Federal R eserve Bank of New York and M a n a g e r of Foreign O p eration s for the System O pen M arket A ccount. Vivek M oorthy was prim arily responsible for preparation of the report. ’ The trad e -w e ig h te d basis is as m easured by the Federal Reserve Board index. 70 FRBNY Quarterly Review/Autumn 1991 The dollar fluctuates without clear direction in May As the period opened, sentiment toward the dollar was favorable but market participants appeared uncertain whether the dollar could extend the sharp rise that it had experienced during the preceding two months. The U.S. discount rate cut of 50 basis points to 5.5 percent on April 30 had been unexpected, and that move gener ated some downward pressure on dollar rates on May 1. The U.S. employment data for April, released on May 3, were stronger than expected, but on inspection, other details of the report revealed areas of continuing weak ness. In that environment, the dollar traded in a narrow range for the first half of May. Then, late in European trading on Friday, May 17, Sweden’s Riksbank announced that it would link the Swedish krona to the ECU, replacing its trade-weighted basket of currencies, in which the dollar carried the largest weight, with a basket composed entirely of Euro pean Community currencies. Within a few. hours of the announcement the dollar moved up by about four pfen nigs against the mark as Swedish and other Scandina vian entities rushed to adjust the currency composition of their liabilities to that of the ECU by purchasing dollars to repay dollar-denominated liabilities. With Swedish interest rates relatively high, Swedish entities had borrowed extensively abroad, partly to finance domestic operations, confident that they were largely shielded from exchange rate risk because the Swedish authorities would limit the movement of the krona’s exchange rate relative to the trade-weighted basket to only a couple of percentage points. With the change in the krona’s peg, the exchange risk these companies would henceforth face on their dollar liabilities was p e rceived to be m uch h ig h er than before, and they had an in ce ntive to replace d o lla r-d e n o m in a te d lia b ilitie s w ith tho se of E uropean cu rre n cie s more heavily repre se n ted in the ECU. W ith U.S. m arkets still open a fte r the Sw edish a n n o uncem e n t, and w ith the m ark rela tive ly w id e ly trade d in the U.S. m arket, the pressures re sulting from the May 17 e xch a n g e -m a rke t o p e ra tio n s were co n ce n tra te d in d o lla r/m a rk tra n s a c tio n s , re su lting in the sharp rise of the d o lla r a g a in s t the m ark. U nder th e se c ircu m sta n ce s, there was som e in te rv e n tio n ; the U.S. a u th o ritie s sold $50 m illio n on th a t Friday. A fte r the w eekend, w ith p ressures c o n tin u in g , there was in te r vention by a nu m b e r of fo re ig n ce n tra l banks. Soon th e re a fte r th e m a rk e ts s e ttle d dow n and th e d o lla r traded in a narrow range fo r the rest of th e m onth. The dollar advances during June and early July Chart 1 D uring ea rly June, a slew of U.S. e c o n o m ic in d ica to rs were released th a t were g e n e ra lly m uch m ore favo ra b le than expected, and m a rke t o b s e rv e rs began to ta lk a bout the p o s s ib ility th a t the U.S. re co ve ry m ig h t be m ore vig o ro u s than p re vio u sly a n tic ip a te d . In response, e xp e cta tio n s of a fu rth e r d e clin e in U.S. in te re s t rates faded and the d o lla r s ta rte d to rise. In p a rticu la r, on June 7, w hen it was re p o rte d th a t May e m p lo y m e n t rose well above e xp e cta tio n s, the d o lla r rose a lm o st tw o pfennigs on the day to clo se at DM 1.7720. D evelopm ents in G e rm a n y d u rin g Ju n e a lso te n d e d to strengthen the dollar. News of G e rm a n y ’s firs t tra de d e ficit since 1981, e vid e nce th a t in fla tio n was h ig h er than pre vio u sly a n tic ip a te d — even before the im p o s itio n of a c o n su m p tio n tax th a t w ould raise all m a jo r price indexes fo r the u pcom ing m o n th s — and w hat som e saw as the reluctance of the B u n d e sb a n k to ra ise o fficia l in te re st rates all w eighed on the m ark. By late June, the After generally rising through June and early July, the dollar eased towards the end of the period. 1991 German marks per U.S. dollar 1.9 Japanese yen per U.S. dollar 150 Chart 2 Three-Month Eurorate Differentials Foreign Rate minus U.S. Rate Percentage points 1.8 German mark — Scale / / 1.7 A v V \ / V 1.4 L i „■ r J/ / ii / \ ! f \ German - - . s ' '" X Japanese yen 130 m 11 11 m I. I I I I I I J / /\ / r / \—' \ 1 ,1 1 I J J 125 A 1991 Notes: The top chart shows the percentage change of the weekly average of daily closing rates from May 1991 through July 1991. The bottom chart shows the weekly average closing rates for the German mark a nd the Japanese yen from January 1991 through July 1991. r\ ^ 135 / / l...i 1 1 1 s.^ ^ 140 r // A "' Ll i i i.. 1i i l 1 l Jan Feb V* Japan i i Mar i 1i i i i 1.1111-i 11111 Apr May Jun i i I Jul 1991 FRBNY Q uarterly Review/Autum n 1991 71 d o lla r had risen w ell above DM 1.80 in in tra d a y tra d in g . W ith respect to the yen, the d o lla r showed its greatest s tre ngth of the re p ortin g period during the first half of June, breaking above the ¥ 1 4 2 level three tim es. The dollar-ye n exchange rate reflected not only the more buoyant o u tlo o k for the U.S. econom y but also concerns in Jap anese fin a n cial m a rke ts a bout p o ssib le problem s w ith banking and sto ck m arket practices. As the d o lla r m oved up to levels not seen fo r more than a year, m arket p a rtic ip a n ts becam e w ary of the p o s s ib ility th a t som e action to curb the d o lla r’s rise m ig ht be decid ed upon at the G -7 m eeting of finance m in iste rs and ce n tra l bankers, scheduled to be held in London on June 23. As a result, the m arket becam e less co nce rn ed abou t the upside risk fo r the dollar, and the curren cy traded in a narrow range as the m eeting a pproached. In the event, the G -7 m in iste rs and gove rn o rs issued a co m m u niq ue tha t “ reaffirm ed th e ir c o m m itm e nt to c o o p e ra te c lo s e ly , ta k in g a c c o u n t of th e n eed fo r o rd e rly m a rke ts, if n e c e s s a ry th ro u g h a p p ro p ria te ly c o n ce rte d action in foreign exchange m a rke ts.” M arket p a rtic ip a n ts did not in itia lly construe the G-7 statem ent as a firm co m m itm e n t to in te rve n e to re sist th e d o lla r’s rise. But co m m e n ts fo llo w in g the m e e tin g by several o f f ic ia ls , in c lu d in g J a p a n e s e F in a n c e M in is t e r H a s h im o to , F rench F in a n c e M in is te r B e re g e v o y and U.S. Treasury U nder S e c re ta ry M u lfo rd , re in fo rce d the fe e lin g th a t the p o s s ib ility of in te rv e n tio n was being a ctively co n sid ere d . R um ors a b o u t o ff-m a rk e t tra n s a c tio n s b e tw e e n th e B u n d e s b a n k a n d th e F e d e ra l R eserve, w hich were la te r co n firm e d by th e a u th o ritie s (see below), were also ta ke n as in d ica tio n s th a t p re p a rations to co n ta in the d o lla r’s rise were underway. T hereafter, the d o lla r rem ained w ell below its e a rlie r highs a g a in st the yen. The release on June 25 of data in d ica tin g a la rg e r than exp e cte d rise in U.S. d u ra b le goods o rd e rs fo r May te m p o ra rily s u p p o rte d the d o lla r a gainst all cu rre n cie s. But the sp re a d in g ta lk of new fin a n cial sca n d a ls in Japan was by th is tim e having o ff se ttin g e ffe cts on the yen. On th e one hand, m arket p a rtic ip a n ts cam e to e xp e ct th a t the a u th o ritie s m ight move more q u ic k ly than o th e rw ise to low er in te re st rates as Ja p a n e se share p rice s co n tin u e d to d e c lin e . In fact, the Bank of Japan a n n o u n ce d a o n e -h a lf p e rc e n t age p oint cut in its d is c o u n t rate, to 5.5 p e rce n t, on Chart 3 Data released during the period first supported the dollar and later led it to ease. The employment report for May was much stronger than anticipated while that for June was much weaker. Thousands of jobs Thousands of jobs 200 200 --------------------D ifference betw een A ctual and E xpected C hanges 150 100 50 -50 100 -150 Apr 1991 May 1991 Jun Notes: The left panel shows the reported monthly changes in nonfarm payroll employment as of end-July. The right panel shows the actual minus expected monthly changes in payroll employment: the actual changes used in computing these differences are the preliminary numbers for April, May, and June 1991 reported on May 3, June 7, and July 5, 1991, respectively, while the corresponding expected changes are based on surveys of market expectations. http://fraser.stlouisfed.org/ 72 FRBNY Q uarterly Review/Autumn 1991 Federal Reserve Bank of St. Louis July 1. On the oth e r hand, m arket p a rtic ip a n ts viewed the adverse im pact of the sto ck m a rk e t’s d ecline on Japanese banks’ ca p ita l ratios as increasing the lik e lihood that m ajor Ja pane se inve sto rs w ould be re p a tri ating overseas funds to invest in new subo rd in a te d debt in stru m e n ts th a t these banks w ould be issuing to shore up th e ir c a p ita l p o sitio ns. These o ffse ttin g d e ve lop m ents helped to keep the dollar-yen exchange rate re la tively steady, tra d in g around ¥ 1 3 8 fo r the rem ainder of the th re e -m o n th re p ortin g period. The m ark, however, cam e u nder fu rth e r se llin g pres sure at the end of June and ea rly July. The d o lla r in itia lly stre n g th e n e d a g a in st the m ark in response to the b e tte r than expected U.S. data fo r May durable goods orde rs released on June 25. Two days later, w hen, in response to a co n tro ve rsial G erm an co u rt ruling, G erm an o fficia ls re p o rte d ly suggested th a t a w ith h o ld in g tax on in ve stm e n t incom e m ight be rein stated, the dolla r broke de cisive ly through the DM 1.80 level. The idea th at such a ta x — ve ry u n p o p u la r w hen it was im posed in 1989 and q u ickly w ith d ra w n — m ight a g a in be u n d e r c o n s id e r a tio n had an im m e d ia te d e p re ssing im pact both on the m ark and m a rk -d e n o m i nated assets. The DAX index of share p rices slum ped 2.5 percent the fo llo w in g day, and the d o lla r continued to rise in the follow ing days to reach its high ag a inst the m ark for the period and the year of DM 1.8427 in Table 1 Federal Reserve Reciprocal Currency Arrangements in M illions of Dollars Amount of Facility Institution A ustrian National Bank National Bank of Belgium Bank of C anada National Bank of Denm ark Bank of England Bank of France D eutsche B undesbank Bank of Italy Bank of Japan Bank of M exico N etherlands Bank Bank of Norway Bank of Sweden Swiss National Bank Bank for International Settlem ents Dollars against Swiss francs Dollars against other authorized European currencies Total July 31, 1991 250 1,000 2,000 250 3,000 2,000 6,000 3,000 5,000 700 500 250 300 4,000 600 1,250 30,100 European tra d in g on J u ly 5. The dollar gives back most of its gains during the rest of July Just as the d o lla r was reaching its highs of the period a g a inst the m ark, m a rke t c o n fid e n c e in the U.S. re co v ery began to weaken. U.S. econom ic data released during the m onth no lo n g er p rovided u n a m b ig u o u s e vid e n ce of e c o n o m ic re c o v e ry . T h e re le a s e on J u ly 5 o f th e e m p lo ym e n t re p o rt fo r June, in p a rticu la r, show ed an unexpected drop in em p lo ym e n t. S im ultaneously, e x p e c ta tio n s began to grow th a t the B undesbank w ould tig h te n m o n e ta ry p o lic y and p ursue a more a ggressive m o n e ta ry sta n ce th a n p re v io u s ly supposed. By then, the release of p rice fig u re s for several G erm an sta te s th a t s u g g e s te d a sh a rp a c ce le r ation in prices fo r “ c o re ” item s was seen as g ivin g the B undesbank more reason fo r an e a rly p o lic y tig h te n in g move. M arket p a rtic ip a n ts a p peared to be u n c e rta in only about the e xte n t and tim in g of such a m ove— w h e th e r it w ould com e before or a fte r the su c c e s s io n of Dr. S ch le sin g e r to the P residency of the B u n d e sb a n k at the end of July. A g a in st this b a ckground, the d o lla r’s rise a g a in s t the m ark sta lle d , and the e xchange rate flu c tu a te d w ith o u t d ire ctio n ju s t above DM 1.80. However, on Ju ly 11, w hen the B u ndesbank did not raise o ffic ia l in te re s t rates at its b iw eekly m eeting and w hen a sh a rp drop in U.S. w eekly u ne m p lo ym e n t in su ra n ce cla im s was re p o rte d , the d o l lar jum ped back up to a lm o st DM 1.84. E arly the next m orning, as the d o lla r co n tin u e d to move higher, foreign ce n tra l banks co n d u cte d several rounds of in te rv e n tio n , se llin g d o lla rs a g a in st both m arks and o th e r cu rre n cie s. A fte r the New York m a rke t ope n e d , the U.S. m o n e ta ry a u th o r itie s a ls o p a r tic ip a te d , s e llin g $ 1 0 0 m illio n a g a inst m arks. The w id e sp re a d p a rtic ip a tio n of ce n tra l banks in the co n c e rte d in te rv e n tio n , ahead of th e G-7 sum m it m eeting the next w eek, and the fa ct th a t the ce n tra l banks co n tin u e d to o p e ra te th ro u g h o u t the day suggested to m arket p a rtic ip a n ts th a t the ce n tra l banks were united in th e ir in te n tio n to curb the d o lla r’s rise. As a result, the d o lla r d e clin e d by a b o u t five p fe n n ig s during the day to clo se in New York at DM 1.7893. T his episode of in te rv e n tio n , to g e th e r w ith an in c re a s in g ly u n ce rta in U.S. e co n o m ic sce n a rio , set the to n e for tra d in g for the rest of the m onth. T he d o lla r again reached the DM 1.80 level on tw o o c c a s io n s the next w eek in response to strong in d u s tria l p ro d u ctio n data and C h a irm a n G re e n s p a n ’s s ta te m e n t in his s e m i annual H um phrey-H aw kins te s tim o n y th a t a recovery was under way, but it fa ile d to move higher. The co m m u n iq u e released on Ju ly 17 a fte r th e G -7 sum m it m eeting re ite ra te d s u p p o rt fo r close c o o p e ra tion in foreign e xchange m a rke ts, m o n e ta ry and fiscal FRBNY Q uarterly Review/Autum n 1991 73 p o licie s to fo s te r low real in te re st rates, and S oviet eco n o m ic and p o litica l tra n sfo rm a tio n . W hile the co m m u nique had little im m e d ia te im pact on exchange rates, it co n trib u te d to an atm osphere in w hich the fe a r of co n ce rte d in te rve n tio n rem ained. In th a t en viro n m e n t, the d o lla r did not stre n g th e n even w hen u nexpectedly favorable housing s ta rts data were released later that day. D uring the rest of July, new U.S data releases brought into q u e stio n the vig o r and even the s u s ta in a b ility of eco n o m ic recovery. S e n tim e n t also spread am ong U.S. fin a n cial m a rket a n a lysts th a t a sig n ifica n t d e clin e in U.S. inflation, both actual and prospective, w ould be reflected in a d e clin e in long bond yie ld s. Moreover, sta te m e n ts by a va rie ty of U.S. officials, in clu d in g som e FOM C m em bers, a b o u t the need to respond if M2 grow th rem ained w eak revived m arket e xp e cta tio n s that U.S. s h o rt rates m ight still decline. As a result, the d o lla r fell below DM 1.80 d u ring the third w eek of July and to levels around DM 1.75 fo r the rest of the period. The d o lla r closed the M ay-July reporting period about 2 p e rce n t higher ag a in st the m ark and 1 pe rce n t higher a g a inst the yen. The d o lla r rose about 5.5 percent a g a inst the S w iss fra n c as expe cta tio n s grew th a t the m o n e ta ry a u th o ritie s in S w itze rla n d w ould not follow th o se in G erm any by tig h te n in g m o n e ta ry policy. The d o lla r eased very slig h tly a g a inst the C anadian d o lla r as th e m a rke t cam e to b e lie ve th a t th e C a n adia n a u th o ritie s w ould be more restrained about easing m o n e ta ry po licy than w ould the U.S. m onetary a u th o ritie s. D u rin g th e re p o rtin g p e rio d , th e U .S . m o n e ta ry a u th o ritie s c o n d u c te d o ff-m a rk e t o p e ra tio n s d ire c tly w ith foreign m o n e ta ry a u th o ritie s to adjust the foreign cu rre n cy reserve assets of both the Exchange S ta b iliz a tion Fund (ESF) and the Federal R eserve. The U.S. a u th o ritie s exchanged $ 8 ,5 4 8 .5 m illion eq u ivale n t of foreig n cu rre n cie s for d o lla rs: • On June 25, the U.S. a u th o ritie s purchased a total of $ 5 ,5 4 8 .5 m illio n a g a in s t G e rm a n m a rk s fro m th e D eutsche B u nde sb a n k in spot and forw ard tra n sa ctio n s. The U .S. and G e rm a n a u th o ritie s agreed th a t th e ir respective h o ld in g s of G erm an m arks and d o lla rs were in excess of cu rre n t needs and that it was to th e ir m utual advantage to reduce those holdings. For each of the se tra n sa ctio n s, 60 p e rce n t of the purchase was to be executed fo r the a cco u n t of the Federal R eserve and 40 p e rce n t for the acco u n t of the ESF. A spot tra n s a c tio n of $ 2 ,2 3 0 .5 m illio n s e ttle d on June 27. A forw ard tra n sa ctio n of $55 6.2 m illio n se ttle d on Ju ly 29. The rem aining forw ard tra n s a c tio n s are to be se ttle d during the re m a in d e r of the ca le n d a r year. • On Ju ly 16, the U.S. a u th o ritie s purchased a total of http://fraser.stlouisfed.org/ 74 FRBNY Q uarterly Review/Autumn 1991 Federal Reserve Bank of St. Louis $3,000 m illio n a g a in s t a n o th e r fo re ig n c u rre n c y in spo t and fo rw a rd tr a n s a c tio n s w ith a fo re ig n m o n e ta ry a u th o rity. T he d o lla rs p u rc h a s e d w ere s p lit e q u a lly betw een the ESF and the Federal R eserve. A spot tra n sa ctio n of $1,000 m illio n se ttle d on Ju ly 18. Forw ard tra n s a c tio n s to ta lin g $ 2 ,0 0 0 m illio n are to be se ttle d during the next q u a rte rly re p o rtin g p eriod. In a d d ition , in July, the ESF sold a to ta l of $130.2 m illio n e q u ivale n t of m arks a g a in s t SD R s. T he ESF also purchased a to ta l of $ 2 3 0 .4 m illio n a g a in s t s a le s of SDR s in tra n s a c tio n s w ith fo re ig n m o n e ta ry a u th o ritie s in need of SDR s e ith e r fo r p a ym e n t of IM F c h a rg e s or fo r repurchases. Both the sa le s and p u rch a se s of S D R s were a rranged by the IMF. P rim arily because of its a c q u is itio n of d o lla rs in the fo re ig n c u rre n c y e x c h a n g e s a n d S D R tr a n s a c tio n s d e scrib e d above, the ESF was able, a fte r the end of the re p orting period, to re p u rch ase $ 2 ,5 0 0 m illio n e q u iv a lent of foreign cu rre n cy w arehoused w ith th e F ederal R eserve. These re p u rch ase s reduced the a m o u n t of ESF foreign cu rre n cy b a la nce s w a re h o u se d w ith the F ederal R e se rve fro m $ 4 ,5 0 0 m illio n e q u iv a le n t to $2,000 m illio n e q u iva le n t as of th e end of A ug u st. D uring the M ay-July p e rio d , th e Federal R e se rve re a l ized profits of $147.5 m illio n from th e o ff-m a rk e t foreign c u rre n c y e x c h a n g e s d e s c rib e d above. T he T re a su ry realized profits of $ 60.3 m illio n , of w h ich $18.8 m illion was from the o ff-m a rk e t fo re ig n c u rre n c y and SDR e xchanges and $41.5 m illio n was from th e renew al of ce rta in w arehousing o p e ra tio n s. The Federal R eserve and the ESF re g u la rly in vest Table 2 Net Profits ( + ) or Losses ( - ) on United States Treasury and Federal Reserve Foreign Exchange Operations In M illions of Dollars Valuation profits and losses on outstanding assets and lia b ilities as of April 30. 1991 Realized A pril 30, 1991 to July 31, 1991 Valuation profits and losses on outstanding assets and liab ilitie s as of July 31, 1991 Federal Reserve U.S. Treasury E xchange S tabilizatio n Fund + 2,316.3 + 570.6 + 147.5 + 60.3 + 1,919.9 + 321.4 Note: Data are on a value-date basis. their foreign currency balances in a variety of instru ments that yield market-related rates of return and that have a high degree of quality and liquidity. A portion of the balances are invested in securities issued by for eign governments. As of the end of July, holdings of such securities by the Federal Reserve amounted to $7,807.7 million equivalent, and holdings by the Trea sury amounted to the equivalent of $7,540.2 million valued at end-of-period exchange rates. FRBNY Quarterly Review/Autumn 1991 75 RECENT FRBNY UNPUBLISHED RESEARCH PAPERS* 9119. H ickok, Susan, Jam es Orr, M.A. Akhtar, and K u rt C. W u lfe ku h ler. “ The U ruguay Round of GATT Trade N e g o tia tio n s.” June 1991. 9120. Kasm an, B ruce, and A nthony R o drigues. “ F in a n cia l R eform and M o n e ta ry C o n tro l in Japan.” M ay 1991. 9121. H ickok, Susan, Juann H. Hung, and K urt C. W ulfekuhler. “An E xpanded, C o in te gra te d M odel of U.S. Trade.” June 1991. 9122. Lown, C ara S. “An In d ica to r of Future Inflation E xtra cte d from the S te e p ness of the In te re st Rate Y ield C urve along Its E ntire L e n g th .” W ith Je ffre y A. Frankel. June 1991. 9123. G aske, M. E llen. “ S ources of F lu ctu a tio n s in Long-Term E xp e cte d Real R ates of Interest: E vidence from th e U.K. Indexed B ond M a rk e t.” J u ly 1991. 9124. H arris, Ethan S. “ Tracking the E conom y w ith the P urchasing M a n a g e rs’ Index.” A u g u st 1991. 9125. S kile s, M arilyn E. “ S ta b iliz a tio n and F inancial S e cto r R eform in M e xico .” A u g u st 1991. 9126. C h a rre tte , S u sa n . “ C a p ita l F lig h t from D e b to r N a tio n s W h e n L a b o r Is M o b ile .” S e ptem ber 1991. 9172. S a n ta m a ria , M a rco . “ P riv a tiz in g N ovem ber 1991. 9 128. W izm an, T h ie rry A. “ R eturns on C a p ita l A sse ts and V a ria tio ns in E conom ic G row th and V o la tility: A M odel of B ayesian L e a rn in g .” N ovem ber 1991. W ith C onnel R. Fullenkam p. 9 129. Akhtar, M .A., and E than S. H arris. “ The S up p ly-S id e C o n se q u e n ce s of U.S. F iscal Policy in the 1980s.” N ovem ber 1991. S o cia l S e c u rity : T he C h ile a n C a s e .” fS in g le co p ie s of these papers are available upon request. W rite R esearch P apers, Room 901, R esearch Function, Federal Reserve Bank of New York, 33 Liberty S treet, New York, N.Y. 10045. FRBNY Q uarterly Review/Autumn 1991 Single-copy subscriptions to the Quarterly Review (ISSN 0147-6580) are free. Multiple copies are available for an annual cost of $12 for each additional subscription. Checks should be made payable in U.S. dollars to the Federal Reserve Bank of New York and sent to the Public Information Departm ent, 33 Liberty Street, New York, N.Y. 10045 (212-720-6150). Single and multiple copies for U.S. and for other Western Hemisphere subscribers are sent via third- and fourth-class mail, respectively. All copies for Eastern Hemisphere subscribers are airlifted to Amsterdam and then forwarded via surface mail. Multiple-copy subscriptions are packaged in envelopes containing no more than ten copies each. Q uarterly Review subscribers also receive the Bank’s Annual Report. Quarterly Review articles may be reproduced for educational or training purposes, provid ing they are reprinted in full and include credit to the author, the publication, and the Bank. Library of Congress Card Number: 77-646559 FRBNY Quarterly Review/Autumn 1991