The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Vol. 28, No. 2 ECONOMIC REVIEW 1992 Quarter 2 Intervention and the Bid-Ask Spread in G-3 Foreign Exchange Rates 2 by William P. Osterberg An Ebbing Tide Lowers All Boats: Monetary Policy, Inflation, and Social Justice 14 by David Altig Sluggish Deposit Rates: Endogenous Institutions and Aggregate Fluctuations by Joseph G. Haubrich FEDERAL RESERVE BANK OF CLEVELAND 23 1992 Quarter 2 Vol. 28, No. 2 Intervention and the Bid-Ask Spread in G-3 Foreign Exchange Rates 2 by William P. Osterberg Recent research suggests that central-bank intervention may influence the volatility of foreign exchange rates or impair the efficiency of such markets. Using official daily intervention data for Germany, Japan, and the United States, the author tests for whether the anticipation of inter vention explains wider bid—ask spreads. No evidence is found for such a relationship in the spot and forward rates of marks/dollars and yen/ dollars. Rather, it appears that narrower spreads are associated with periods of purported intervention and that spreads are narrower if, con ditional on the occurrence of intervention, the market is likely to have expected intervention. An Ebbing Tide Lowers All Boats: Monetary Policy, Inflation, and Social Justice 14 by David Altig Some economists argue that, because low-income individuals are un duly burdened by unemployment and not much affected by inflation in the short run, fairness dictates expansionary monetary policy in times of sluggish economic activity. However, individuals with low incomes are likely to be hurt in the long run if such policies lead to higher infla tion. This paper argues that the same social justice criterion that justi fies the call for the Fed to “do something" during recessions supports the case for a long-run anchor to the price level. Sluggish Deposit Rates: Endogenous Institutions and Aggregate Fluctuations 23 by Joseph G. Haubrich This paper provides an equilibrium analysis of how endogenously aris ing financial institutions alter the impact of macroeconomic shocks. It explains the low volatility (sluggishness) of bank interest rates relative to other short-term rates and illustrates a powerful principle: When aggre gate disturbances also have distributional consequences, the shock can change the pattern of prices specified by efficient contracts. Interest-rate sluggishness arises because banks provide insurance against individual uncertainty, which itself is affected by economic conditions. Economic Review is published quarterly by the Research Depart ment of the Federal Reserve Bank of Cleveland. Copies of the Review are available through our Public Affairs and Bank Relations Depart ment, 1-800-543-3489. Coordinating Economist: James B. Thomson Advisory Board: David Altig Erica L. Groshen William P. Osterberg Editors: Tess Ferg Robin Ratliff Design: Michael Galka Typography: Liz Hanna Opinions stated in Economic Review are those of the authors and not necessarily those of the Federal Reserve Bank of Cleveland or of the Board of Governors of the Federal Reserve System. Material may be reprinted provided that the source is credited. Please send copies of reprinted material to the editors. ISSN 0013-0281 Intervention and the Bid-Ask Spread in G-3 Foreign Exchange Rates by W illiam P. Osterberg W illiam P. Osterberg is an economist at the Federal Reserve Bank ot Cleveland. The author is grateful to David Altig, Jeffrey Hallman, Owen Humpage, Jacky So, and James Thomson for use ful comments and suggestions, and to Rebecca Wetmore Humes for research assistance. Introduction Tests of the efficiency of foreign exchange markets continue to proliferate. Because these markets have become worldwide in scope and nonstop in operation, economists have been able to test many hypotheses about how infor mation becomes incorporated into prices and transferred between markets in different loca tions. However, the finding that forward rates for foreign exchange are not unbiased predic tors of future spot rates remains without a coherent explanation. It seems reasonable to speculate on the role that central-bank intervention plays in such find ings. After all, central banks may possess infor mation not available to other traders. However, since central banks usually have not made avail able to the public accurate information about their daily foreign exchange activities, it has been difficult to determine if intervention influ ences foreign exchange market efficiency. Greater interest in central-bank intervention has also been stimulated by an increase in the frequency of intervention. During the period of ostensibly floating rates, central-bank interven tion policy has been, at various times, designed http://fraser.stlouisfed.org/ either to influence the level of the exchange Federal Reserve Bank of St. Louis rate or to reduce its volatility. Specifically, as dis cussed by Funabashi (1989) and Dominguez (1990), soon after the Plaza accord in September 1985, the finance ministers of the G-5 (France, Germany, Japan, the United Kingdom, and the United States) agreed to reduce the dollar’s ex change value. Then, at the Louvre meeting in 1987, they decided to shift to a regime of stabili zation. Thus, there is a clear interest in analyz ing the impact of intervention on both the level and volatility of exchange rates. This paper examines the relationship between G-3 (Germany, Japan, and the United States) central-bank intervention and bid-ask spreads in the German mark/U.S. dollar (DM/$) and yen/U.S. dollar (yen/$) spot and forward foreign exchange markets. Bid-ask spreads may be related to vola tility and risk. The examination is stimulated by the speculation of Bossaerts and Hillion (1991), henceforth referred to as B/H, that an intraweekly pattern in intervention explains the intraweekly pattern in bid-ask spreads that they observed for the currencies in the European Monetary System. They determine that bid-ask spreads are higher on Fridays and that taking account of such asym metry alters conclusions regarding the efficiency of forward markets. B/H surmise that the higher Friday bid-ask spreads are related to market participants’ anticipation that decisions about intervention will be undertaken on weekends. Here, I utilize official intervention data to see if G-3 intervention influences G-3 spreads. To the best of my knowledge, this is the first time that such an investigation has been undertaken. The organization of this paper is as follows. Section I reviews selected literature on forward market efficiency and the impact of central-bank intervention. Section II discusses the data on inter vention and the bid-ask spreads. In the third sec tion, I examine 1) intraweekly patterns in bid-ask spreads, 2) holiday effects, 3) intraweekly patterns in intervention, 4) bid-ask spreads over periods of nonintervention versus intervention, and 5) Granger-causality tests for the intervention-spread relations. Section IV concludes and summarizes. I find that 1) bid-ask spreads are higher on Fridays for both spot and forward G-3 exchange rates, 2) intervention is no more likely to occur on Mondays than on other days, 3) for both cur rencies, periods of purported intervention are associated with lower, rather than higher, bidask spreads, 4) conditional on whether or not intervention occurred, expectations of interven tion seem to be associated with lower spreads, and 5) intervention generally does not Grangercause spreads. Overall, there appears to be little evidence to support the view that spreads widen in anticipation of intervention. A more plausible view is that the expectation of intervention has a negative impact on spreads. A structural model of the relations among intervention, spreads, and volatility would be necessary to address these issues in more detail. and asks from their books of orders. Flood sug gests that adverse selection costs and inventory holding costs are likely influences on marketmaker spreads. Adverse selection influences spreads if market-makers confront traders who have inside information and who are thus able to speculate against the market-maker. Inven tory holding costs are influenced by the possi bility of unfavorable price changes during the time that currencies are held. Flood submits that models of brokers’ spreads are less applicable to the foreign exchange mar ket, where (unlike in securities markets) brokerage and market-making are separated. One distinction between the two activities is that brokerage main tains the anonymity of the transacting parties. It is worth noting that U.S. intervention operations util ize both market-makers, who generally are com mercial and investment banks, and brokers. “Secret” intervention occurs via a broker. Interven tion via a market-maker may increase spreads, since a market-maker could view the intervening central bank as having inside information. This is the mechanism to which B/H refer. Early empirical work by Fieleke (1975) and Overturf (1982) shows that spreads are positively related to foreign-domestic interest differentials and exchange volatility. Allen (1977) provides a theoreti cal rationale for the volatility-spread relation. Black (1989), Boothe (1988), Glassman (1987), and Wei (1991) find that spreads are positively related to transactions volume. Although a lower rate of trans actions could influence the risk component of the spread by increasing the length of time an open position would be held, it is also possible that vola tility and volume are determined simultaneously. I. Related Literature Foreign Exchange Bid-Ask Spreads A small body of literature focuses on the determi nation of foreign exchange bid-ask spreads. Flood (1991) provides a summary of the theory and points out the difficulties in applying to the foreign exchange market the framework used to analyze securities market spreads.1 A unique aspect of spread determination in exchange markets is that two trading structures coexist. There are marketmakers, who provide both bids and asks upon demand, and brokers, who quote the best bids 1 George, Kaul, and Nimalendran (1991) provide a recent summary of the findings regarding spreads in equity markets. They conclude that only 8 to 13 percent of the spreads can be explained by adverse selection and http://fraser.stlouisfed.org/ claim that the predominant influence on equity spreads is processing costs. Intervention and Risk Premia Though B/H contend that intervention influences the spread, most investigations have viewed inter vention as influencing a risk premium defined in other terms, as discussed below. However, it is not at all clear that a significant risk premium exists (see Hakkio and Sibert [1991]). The existence of a time-varying risk premium is only one of the pos sible explanations of the finding that the forward rate is not an unbiased and efficient predictor of future spot rates. The unbiasedness and efficiency of forward rates has been widely tested by analyzing varia tions on equation ( l) .2 ■ Federal Reserve Bank of St. Louis ■ 2 Baillie and McMahon (1989) and Hodrick (1987) provide com prehensive reviews of this literature. (U Et (SH i ) = FIH II, where the left side is the expectation at time t of the spot exchange rate k periods in the future and the right side is the forward rate at time t for a transaction at time t + k. In practice, Et (St+k) is replaced by S[+k, S and F are often replaced by their logarithms or by (St+ k—St)/St and (.Ft—St)/St; and an equation such as (2) is analyzed.3 G) (5 ,+ * - 5 , ) / J , —oc + (3 (Ft t+k —St ) / St + ul+ k . As summarized by Baillie (1989), a consensus against unbiasedness has emerged— the hypoth esis that a = 0 and that (3 = 1 is usually rejected. One possible explanation is the presence of a risk premium. Equation (1) could be expected to hold purely as the outcome of arbitrage among riskneutral speculators who can take an open position in the forward market based on their expectation of the future spot rate at which positions would have to be covered. On the other hand, the portfolio-balance approach to exchange-rate de termination considers risk-averse investors who choose holdings of assets denominated in differ ent currencies. If such assets are imperfect substi tutes, then factors such as relative asset supplies will influence exchange rates and imply rejection of the unbiasedness hypothesis. B/H suggest that the frequent use of the aver age of the bids and asks in equations (1) and (2) is inappropriate and claim that intervention is responsible for an asymmetry of the true price around the average of bids and asks. Other possible theoretical explanations include the inappropriateness of the rational expectations assumption (Frankel and Froot [1987D, the pos sibility that policy changes would lead to ex post biasedness even if unbiasedness held ex ante (Lewis [1988]), anticipation of real exchangerate changes (Levine [19891), and the existence of liquidity premia (Engel [1990]).4 A variety of approaches, summarized by Hodrick (1987), imply a time-varying risk pre mium. Lucas (1978) relates the risk premium to ■ 3 These transformations ameliorate problems introduced by the nonstationarity of exchange rates (see Baillie and McMahon [1989] or Hodrick [1987] for details) and Siegel’s paradox. Siegel's paradox states that if equa tion (1) holds when S and F are expressed as units of currency A per unit of currency B, then it cannot also hold for the inverse rates, because E(MK) and 1/E(X) are not equal. ■ 4 Other possibilities include Siegel's paradox and transactions costs. Research has generally concluded, however, that these are not im http://fraser.stlouisfed.org/ portant empirically. See Baillie and McMahon (1989) for a summary. Federal Reserve Bank of St. Louis the conditional covariance between a long posi tion in the forward market and the marginal rate of substitution between future and current con sumption. Hodrick (1989) shows how the risk premium in the forward market can be more directly related to the conditional variance of market fundamentals, such as money supply and government spending. Osterberg (1989) modifies Hodrick’s paper to show how interven tion can influence the risk premium in the for ward market. In general, evidence in favor of the existence of a risk premium in the forward market is weak (see Engel and Rodrigues [1989], Kaminsky and Peruga [1990], and Mark [1988]). This may result in part from using data of no higher than monthly frequency in analyzing the relationship between the forward-rate forecast error and either consumption or money.5 Vola tility measures such as conditional variance ex hibit less time variation when constructed from data of lower frequency. Measurement and testing issues are also in volved with the controversy over the existence of risk premia in forward rates 6 B/H determine that the use of the average of bids and asks in tests of forward market efficiency ignores the information contained in the bid-ask spread, biases the test results, and distorts the magnitude of the implied risk premium. In this paper, I focus on the authors’ contention that the bidask spread widens when the market anticipates intervention, because the possibility of interven tion induces an adverse selection problem for market-makers or brokers. B/H conclude that the spreads are wider on Fridays. Other investigators have found evidence of day-of-the-week effects in foreign exchange markets. Glassman (1987) finds that bid-ask spreads are higher on Fridays and on days before market holidays. So (1987) confirms pre vious findings that exchange rates on Monday ■ 5 There are indirect approaches to testing for a risk premium using daily data. One approach is that taken by Levine (1989), who tests the im plication of many asset-pricing models that the risk premium embedded in the forward rate is exactly equal to the risk premium in the differential in real interest rates. Giovannini and Jorion (1987) test for the influence of various proxies for a risk premium, such as lagged forward rates and squared interest rates. ■ 6 Bekaert and Hodrick (1991) discuss the impact on the measure ment of risk premia of 1) matching forward and spot quotes so as to be consistent with settlement conventions in the foreign exchange markets and 2) the use of averages of bids and asks. ■ 7 Thaler (1987) summarizes the evidence regarding day-of-theweek effects in equity markets, but finds no consistent explanation of the results. Negative returns from Friday to Monday are due to the change from the Friday close to the Monday open. Highest returns are on Wed nesday and Friday. Returns also tend to be lower on days before holidays. and Wednesday tend to be higher than on Thurs day and Friday. McFarland, Petit, and Sung (1982) contend that these findings may be related to set tlement conventions and to the fact that money supply announcements are made on Thursday. Baillie and Osterberg (1991) examine the forwardrate error with daily data and find that conditional variances are higher on Fridays and before holi days. Baillie and Bollerslev (1989) estimate a generalized autoregressive conditional heteroscedasticity (GARCH) model for daily exchange rates and determine that conditional variances are higher on Mondays and lower on Thursdays. Humpage and Osterberg (1992) estimate a GARCH model for the risk premium implied by the deviation from uncovered interest parity for the G-3 currencies. They find that the risk pre mium for the DM/$ is lower on Thursdays and that the conditional variance of the deviation for the yen/$ is higher on Fridays and around holi days. Hsieh (1988) concludes that daily exchangerate distributions are not independently and identically distributed across days and that there are no day-of-the-week effects in the mean of the exchange-rate change. However, he does find that variances are larger when the trading period spans a weekend or holiday. Channels of Influence for Intervention The linkage between intervention and bid-ask spreads has not previously been examined. In stead, studies of intervention view it as influencing risk premia or conditional variances. Most analyses have concentrated on sterilized intervention, partly because there is interest in whether it can be viewed as a policy lever in addition to monetary and fiscal policies.8 Unsterilized intervention is equivalent to monetary policy. The two major channels through which steril ized intervention can influence exchange rates are the portfolio-balance channel and the signal ing channel.9 Sterilized intervention alters the ■ 8 A country sterilizes its intervention when it negates the initial impact ol the intervention on its money supply through an offsetting open-market transaction. For example, when U.S. authorities purchase marks with dollars, the supply of dollars is increased. Selling U.S. government securities in the same amount as the intervention removes dollars and sterilizes the intervention. ■ 9 Some authors have suggested other channels. Humpage (1988) finds that intervention sometimes provides “news” other than about future monetary policy. Dominguez (1988) discusses how intervention can have an influence by misleading exchange market participants. The vast major ity of theoretical and empirical research focuses on the portfolio-balance http://fraser.stlouisfed.org/ and signaling channels. Federal Reserve Bank of St. Louis relative supplies of domestic and foreign bonds and, if investors are risk averse and if domestic and foreign bonds are imperfect substitutes, leads to a readjustment of rates of return via the exchange rate; this is the portfolio-balance mechanism. The impact of intervention operat ing through the portfolio-balance channel can be mitigated by three conditions: 1) perfect sub stitutability, 2) Ricardian equivalence, under which consumers perfectly anticipate future taxes associated with the change in government debt, and 3) the slight effect of intervention on asset supplies. The signaling channel is usually analyzed with in the asset-market approach to exchange-rate determination. Exchange rates equal the present discounted value of future economic fundamen tals. If monetary authorities have inside informa tion, intervention may signal future monetary policies. For example, a sterilized purchase of marks by the United States may lead to an appre ciation in marks (a decrease in the DM/$ rate) if the purchase is believed to signal inside informa tion (more expansionary U.S. monetary policy) that increases the expected future exchange rate. The question arises as to why intervention is the type of signal chosen. One answer may be that it gives authorities an incentive to follow through with the expected policy. For example, if authorities have just purchased foreign currency, they may wish to see an appreciation in its value. O n the other hand, since intervention does not re quire an immediate change in the monetary base, market participants may be misled. However, if the subsequent monetary policy is not consistent with that implied by the initial action, the effective ness of future intervention may be reduced. This has led some to suggest that intervention is an effective signal only if followed by consistent mon etary policy. If this is true, however, it is not clear that intervention is independent of monetary pol icy. Humpage (1991) discusses concerns asso ciated with this point. Empirical evidence suggests that the signal ing channel is probably of more significance than the portfolio-balance channel. Early stud ies of the latter, summarized by Obstfeld (1988), generally find that intervention has little impact or that coefficients’ signs are inconsistent with theory. One reason for the small estimated im pact is that intervention is minute relative to the outstanding stocks of assets. Another reason may be that calculation of asset supplies pre cludes the use of high-frequency data. Studies that utilize relatively high-frequency data have found signaling effects. Dominguez (1988) examines weekly data on money surprises, exchange rates, and intervention and concludes that the effectiveness of intervention as a signal depends on the credibility of the implied monetary policy. In a later paper, Dominguez (1990) finds the distinction between coordinated and unilateral intervention to be important. If the mechanism was portfolio balance, only the change in relative asset supplies would matter.10 Few studies use both daily exchange-rate data and official intervention data, as does this paper. Dominguez (1990), Loopesko (1984), and Humpage and Osterberg (1992) use official data to examine the impact of intervention on the risk premium implied by deviations from uncovered interest parity. All three studies find significant effects of intervention. Baillie and Humpage (1992) estimate a simultaneous sys tem in which intervention either “leans against the w ind” or seeks to stabilize volatile markets. They determine that intervention influences the conditional variance of the exchange rate. Bail lie and Osterberg (1991) examine intervention’s impact on the conditional mean and variance of the daily forward-rate forecast error, finding that U.S. purchases of foreign currency influence the conditional mean. If efficiency is assumed, the mean is interpreted as a risk premium. B/H and Hung (1991) both view intervention as operating via the market microstructure of heterogeneous traders. In B/H, traders face the possibility that the central bank may decide to push the rate down or up. As a result, traders may find that they have offered to buy too high or to sell too low. In either case, the dealer sets a wider bid-ask spread. Hung (1991) considers a signaling role for intervention that differs from that discussed by Dominguez. If doubts about credibility make intervention an ineffective signal of monetary policy, and if the market is without a strong di rection, public intervention can influence the trading strategies of chartists or other nonfunda mental traders. A strong implication of this is that the central bank must know the current market trading strategies. In addition, the ability of intervention to increase or decrease volatility depends on market conditions. For example, if the dollar is acknowledged to be overvalued but is still moving upward, the Fed would prefer to wait until a short-term downward movement ■ 10 Dominguez and Frankel (1991) and Ghosh (1989) attempt to distinguish between portfolio-balance and signaling channels. Using monthly data, Ghosh finds that portfolio-balance variables add a small but significant effect to exchange rates. With weekly data, Dominguez and Frankel determine that the signaling mechanism enhances the portfolio http://fraser.stlouisfed.org/ effect. Federal Reserve Bank ofbalance St. Louis began, which it could encourage through secret intervention. Selling dollars with this downward trend would increase volatility. However, if the dollar is on a strong downward trend, the Fed could help it move down and decrease volatility by countering short-term upward movements. II. Data The exchange-rate data were provided by the Federal Reserve Bank of New York. At 10:00 a.m. of each day on which the New York market is open, the Bank obtains both bid and ask quotes for the spot and forward rates for the DM/$ and the yen/$. The intervention data were provided by the Board of Governors of the Federal Reserve System.11 I analyze four series: U.S. purchases of dollars vis-à-vis the mark, Ger man purchases of dollars (sales of marks), U.S. purchases of dollars vis-à-vis the yen, and Jap anese purchases of dollars (sales of yen). The sample period is from August 6, 1985 to September 6, 1991. However, because not all Japanese and German holidays coincide, the num ber of observations differs for the two exchange rates under examination. The intervention data are close-of-business (COB) net daily purchases, measured in $1 million units. The following analy sis attempts to account for the fact that the foreign exchange quotes are not contemporaneous with the intervention numbers. Unfortunately, the avail able data do not permit discrimination between interventions that occur via a broker and those that occur via market-makers. III. Results Table 1 presents the bid-ask spreads for both the spot and forward rates for the DM/$ and yen/$ for each day of the week. Beneath the spreads are the t-statistics for the hypothesis that each day’s spread is equal to the Friday spread. Except for the Tuesday numbers for both the spot and forward spreads for the yen/$, the two-tailed test indicates rejection of the null at the 5 percent level. In all cases, the null is rejected at 10 percent. The Friday versus nonFriday tests are consistent with these results. Table 2 looks at holiday effects in the spreads. This focus is motivated by three facts: First, mar kets by definition are closed on holidays as well as on weekends (although markets may be open elsewhere in the world on U.S., Gennan, or ■ 11 The data on U.S. intervention are now publicly available from Publications Services, Board of Governors of the Federal Reserve System. D TABLE 1 Daily Patterns in Bid-Ask Spreads DM/$ Spot T-stat. Forward T-stat. N Yen/$ Spot T-stat. Forward T-stat. N Monday Tuesday Wednesday Thursday Friday Non-Friday 6.360E-4 5.l60a 7.6l6E^i 4.984a 6.465E-4 4.818a 7.687E-4 4.779a 304 6.796E-4 3.523a 8.034E-4 3.497a 302 7.774E-4 6.544E-4 5.985a 7.778E-4 266 6.534E-4 4.405a 7.754E^i 4.4lOa 311 6.123E-2 2.712a 7.316E-2 2.531a 263 6.384E-2 1.760b 7.52E-2 1.735b 304 6.222E-2 2.468a 7.370E-2 2.465a 303 6.186E-2 2.530a 7.378E-2 2.359a 298 9-065E-4 9 .6 6 0 a 304 1,183 6.872 6.233 3.324a 7.407E-2 3-181a 1,168 8.062E-2 298 a. Significant at the 5 percent level for a two-tailed test. b. Significant at the 10 percent level for a two-tailed test. NOTE: Entries for “spot” and “forward” are the average bid-ask spreads. The t-tests are for the differences from the Friday spreads. “N ” indi cates the num ber o f observations. SOURCE: Author’s calculations. Japanese bank holidays). If spreads are higher on Fridays because markets are going to be closed and prices therefore cannot “reveal” in formation, spreads may also be higher on days before holidays. Second, an examination of the intervention data shows that intervention does not occur on weekends, although it does some times occur on U.S., German, or Japanese holidays in markets that are still open. If market participants are aware of these facts, and if an ticipated intervention widens spreads, then spreads will indeed be wider on days before holidays. Third, since more holidays are on Mondays than on any other day, the “Friday effect” could be a “holiday effect.” In order to focus on the possible influence of intervention on spreads, I isolate a pure holiday effect by controlling for whether or not the day before a holiday falls on a Friday. I also present the com parisons necessary to detect a pure Friday effect. The results show that spreads are higher on days before holidays, but there is mixed evi dence of a pure holiday effect. First, although spreads are higher on Fridays before holidays than on other Fridays, the difference is not sig nificant for any of the four spreads. Second, for other days before holidays, both spot and for ward spreads are wider for the DM/$ rates, but not for the yen/$ rates. There is also mixed evi dence for a pure Friday effect. In terms of both http://fraser.stlouisfed.org/ currencies and spreads, Fridays not before Federal Reserve Bank of St. Louis holidays are higher than non-Fridays not before holidays. However, there are no significant dif ferences between Fridays not before holidays and non-Fridays not before holidays. These comparisons provide no compelling reason to think that higher spreads on Fridays are due to the fact that many Fridays fall before holidays on which intervention may occur. The last column of table 2 compares spreads on days before single holidays with spreads on days before consecutive holidays. The spreads on days before multiple holidays are lower than, but not significantly different from, days before single holidays. The remaining tables present information about the relationship between the daily and holiday patterns in spreads and intervention.12 Ideally, data on expected intervention would be used to test the hypotheses presented by B/H. Newspapers regularly report intervention. Such reports, however, often either mention inter vention that did not occur or fail to note actual intervention (see Klein [19921). Another consid eration is that while the foreign exchange quotes are as of 10:00 a.m., the intervention data are as of COB. ■ 12 Intervention rarely occurred on holidays. The United Slates and Germany intervened five and nine times, respectively, in the DM/$ mar ket. The United States and Japan intervened eight and 13 times, respec tively, in the yen/$ market. 8 TABLE 2 Friday and Day-BeforeHoliday Effects in Bid-Ask Spreads DM/$ Spot T-stat. (H) T-stat. (F) Forward T-stat. (H) T-stat. (F) N Yen/$ Spot T-stat. (H) T-stat. (F) Forward T-stat. (H) T-stat. (F) N A B C D E F G H Before ~A Fri., A Fri., -A -Fri., A ~Fri., ~A Multiple Single 8.015E-4 6.718E-4 3.694a 8.430E-4 7.664E-4 1.290 7.600E-4 6.502E-4 2.355a 6.250E-4 8.097E-4 -0.868 9.292E-4 7.961E-4 3.595a 7.733E-4 2.406a 7.500E-4 9.376E-4 -0.844 90 1,397 1,138 4 86 6.943E-2 6.320E-3 2.038a 6.210E-2 1.341 6.423E-2 7.020E-2 -0.672 8.223E-2 7.489E-2 2.253a 7.380E-2 1.488 7.577E-2 8.323E-2 -0.779 101 1,365 1,118 13 88 0.955 9.669E-4 0.820 45 7.147E-4 0.692 8.443E-2 0.678 51 8.960 1.130 259 6.815E-4 0.570 7.983E-2 0.744 247 5.408a 8.916E-4 5.390a 45 6.735E-2 2.905a 8.007E-2 2.708a 50 a. Significant at the 5 percent level for a two-tailed test. NOTE: Entries for “spot” and “forward” are the average bid-ask spreads. “N ” indicates the number of observations. Explanation o f columns: A: Days before market holidays B: C: D: E: F: (~A) Days not before market holidays (Fri., A) Fridays before market holidays (Fri., ~A) Fridays not before market holidays (~Fri., A) Non-Fridays before market holidays (~Fri., ~A) Non-Fridays not before market holidays G: Days before multiple, consecutive market holidays H: Days before single market holidays Explanation o f t-statistics: (H), (F) distinguish tests designed to isolate pure day-before-holiday and Friday effects, respectively. B: Days before holidays compared to days not before holidays C: Fridays before holidays compared to non-Fridays before holidays D: Fridays before holidays compared to Fridays not before holidays E: Fridays not before holidays compared to non-Fridays not before holidays F: Non—Fridays before holidays compared to non-Fridays not before holidays H: Days before multiple holidays compared to days before single holidays SOURCE: Author’s calculations. Table 3 presents the daily variation in fre quency of intervention. B/H suggest that deci sions about intervention took place over the weekend for the currencies in the European Monetary System. If this were true for the G-3, we may expect to see more intervention occur ring on Mondays. However, there is no signifi cant evidence that this is the case. Rather than define periods of intervention as days on which intervention officially occurred ex post, in table 4, I use two measures of ex http://fraser.stlouisfed.org/ pected intervention. Panel A compares the bidFederal Reserve Bank of St. Louis ask spreads over periods usually thought of as times of intervention as opposed to “noninter vention” periods. Ignored for the moment is the issue of whether intervention actually occurred at these times. The intervention periods are defined as 9/1/85 to 12/31/85, 9/1/86 to 1/1/87, 2/1/87 to 6/1/87, and 10/1/87 to 12/31/87. The most noteworthy dates are 9/22/85 (Plaza ac cord), 2/23/87 (Louvre accord), and 10/19/87 (the U.S. stock market crash). Dominguez (1990) presents reasons to focus on the wider time frames utilized here. The nonintervention 9 T ABL E 3 Day-of-the-Week Effects in Intervention DM/$ U.S. T-stat. Germany T-stat. N Monday Tuesday Wednesday 0.1312 0.1158 0.5759 0.1897 0.0614 311 0.1118 0.7198 0.1447 1.5022 304 0.1325 -0.0306 0.1523 1.245 302 0.1513 -0.6724 0.1809 0.3303 304 0.1278 0.1684 0.1671 0.9663 1,221 0.1151 0.7794 0.1875 1.4067 304 0.1089 1.0135 0.1848 1.4877 303 0.1342 0.0915 0.2114 0.6902 298 0.1174 0.6899 0.1946 1.1843 298 0.1189 0.8079 0.1945 1.5089 1,203 0.1917 266 Yen/$ U.S. T-stat. Japan T-stat. N 0.1367 0.2358 263 Thursday Friday Non-Mond NOTE: Entries for each country are the proportion o f days on which intervention occurred. T-statistics are for the difference between the M on day numbers and other days. “N” indicates the num ber o f observations. SOURCE: Author’s calculations. TABLE 4 Bid-Ask Spreads: Intervention Periods vs. Nonintervention Periods Panel A: Purported Intervention? Panel B: Two Consecutive Days Panel C: Expected vs. Unexpected, Realized vs. Unrealized 1) Yes 2) No 1) Int. 2) Non. A :l, B:1 A: 1, B:2 A:2, B:1 A:2, B:2 DM/$ Spot T-stat. Forward T-stat. N 8.342E-4 -1.308 9930E-4 -5.465a 339 8.744E-^t 6.670E-4 -0.472 7.987E^ -0.211 111 6.82 IE-4 7.707E-4 8.050E-4 9361E-4 1,145 41 8.323E-4 -1.011 9-883E-^ -0.832 229 9.455E-4 -1.576 1.084E-3 -1.300 11 8.856E-4 -1.973a 1.042E-3 -1.779b 222 Yen/$ Spot T-stat. Forward T-stat. N 7.799E-2 -2.099a 9.290E-2 -1.975a 339 8.41 IE-2 7.091E-2 3.855a 8.464E-2 4.524a 147 6.176E-2 7.393E-2 7.309E-2 8.907E-2 1,098 61 7.671E-2 -0.667 9.113E-2 -0.465 234 8.182E-2 -1.354 9.655E-2 -1.262 44 8.317E-2 -1.880b 9.820E-2 -1.818b 161 1.030E-3 246 9-890E-2 246 a. Significant at the 5 percent level for a two-tailed test. b. Significant at the 10 percent level for a two-tailed test. NOTE: T-statistics for panels A and B are for the intervention-nonintervention difference. T-statistics for panel C are for the differences from the A :l, B:1 spreads. “N ” indicates the num ber o f observations. Explanation o f panel C: A: 1, A: 1, A:2, A:2, B: 1: B:2: B: 1: B:2: Days Days Days Days o n w hich intervention was expected and realized o n which intervention was expected but not realized o f “surprise” intervention o n which intervention was neither expected nor realized SOURCE: Author’s calculations. 10 ■ 13 I am grateful to Jacky So for suggesting this further refinement. Because B/H claim that anticipation of intervention widens spreads, theirs is a claim about weak-form market efficiency. Use of actual, confidential http://fraser.stlouisfed.org/ intervention data is relevant for tests of strong-form efficiency. Federal Reserve Bank of St. Louis findings in panel B had for concluding that inter vention lowers spreads. More important, how ever, panel C is contrary to the B/H hypothesis that expectations of intervention increase bidask spreads. Causality should not be inferred from correla tions such as those presented here. While B/H contend that spreads widen in anticipation of in tervention, at times intervention has been in tended to counter volatility. Bid-ask spreads may in part reflect volatility, and thus interven tion and bid-ask spreads may be correlated be cause of attempts to counter volatility reflected in spreads. In the absence of a fully specified model of the determinants of the spreads and of the re sponse of intervention to market movements, I utilize the concept of Granger-causality to learn more about the temporal relations between spreads and intervention. Granger-causality util izes equations of the form <7 M II p (3) bSSi S t - i= 1 /+ Xf ii yfe= 1 bS i j I t- j + us n 7=1 r (4) M period is defined as all other days. For purposes of comparability, the panel A calculations leave out the post-1987 subsample. Both DM/$ and yen/$ spreads are significantly lower during periods of purported intervention. Panel B of table 4 compares spreads from days within actual intervention periods with days from periods when intervention did not occur. Specifi cally, if either the United States or Germany was intervening on day t- 1 and on day t, the 10:00 a.m. day t quote on the DM/$ is said to be from a period of intervention. If both countries were not intervening on either day, the quote is from a non intervention period. In effect, this indicates that if there was intervention on day t-\ (ex post) and intervention as of COB on day t, it is likely that, at 10:00 a.m. on day t, traders perceived that they were in the midst of a period of intervention. Table 4 shows that the yen/$ spreads were significantly higher during these periods, while the DM/$ rates were lower, though not significantly so. Panel C further refines these measures of ex pected intervention.13 The periods of purported intervention analyzed in panel A might be better thought of as periods when intervention was likely to have been anticipated. The “two con secutive days” criterion utilized in panel B may better identify periods of actual intervention. Thus, one possible explanation of the higher spreads for the yen/S in panel B may be that not all intervention that occurred during two consecutive days was anticipated. Days that fell into the first columns of both panels A and B may more closely identify intervention that was both expected and realized. Days that fell into both of the second columns tell us when inter vention neither occurred nor was expected. The in-between cases are when days met only one of the criteria. Panel C provides the results for all four cases. All four of the t-statistics imply significant dif ferences at the 10 percent level, and the relative magnitudes of the spreads are consistent with my interpretation of panel A- Spreads are lower when actual intervention was expected than when intervention was neither expected nor realized. Spreads when intervention was expected but not realized lie between the “expected intervention” and “neither” cases. In addition, conditional on whether intervention occurred as defined by the panel B criterion, spreads are lower when inter vention was anticipated, as defined in panel A. This weakens the qualification that the yen/$ S ^isk ^ n il t - 1 St- k + + uIt 1= l Here, / and S are each regressed on past val ues of themselves and on lagged values of the other variable. / Granger-causes S if past values of / improve upon the ability of past values of S to predict S. Since the focus is on whether inter vention Granger-causes spreads, I test for the significance of the bSI's.14 However, before esti mating these equations, I test for the presence of unit roots in the spreads. The presence of such effects would imply a type of nonstationarity that would invalidate the results. I consistently reject the null that such an effect existed.15 In addition, the length of the autoregressions, p, q, r, and s, must be chosen. I arrive at a lag length of 20 by considering successively longer lag lengths (10, 15, 20, and 25) and by testing whether the addi tional terms are significant. ■ 14 Alternative concepts of, and tests for, causality are presented by Granger and Newbold (1986). ■ 15 These tests were performed with both the Dickey-Fuller and Phillips—Perron procedures, both with and without deterministic trends. The number of lagged first differences on the right side was the minimum number to produce residuals that were free of serial correlation as meas ured by Box—Ljung Q statistics. Baillie and McMahon (1989, pp. 105— 107) discuss these test procedures. The results of the unit root tests are available from the author. J T A B L E 5 | intervention. In this paper, I use official data on Granger-Causality Tests: Intervention to Spreads, Significance Levels Full Sample 9/9/8512/31/86 1/1/8712/31/89 U.S.-Germany Int.—»Spreads 0.4978 0.4260 0.3657 U.S.-Japan Int.—»Spreads 0.9680 0.0001 0.6717 NOTE: Significance levels are for the likelihood ratio tests o f whether the vec tor of intervention terms Granger-causes the vector o f spreads. SOURCE: Author’s calculations. Table 5 presents the results of the tests for Granger-causality from intervention to spreads.16This is done for each currency, so that when DM/$ (yen/$) spreads are on the left side, then lagged DM/$ (yen/$) spreads, lagged Ger man (Japanese), and lagged U.S. intervention are on the right side. For the full sample, there is no evidence of Granger-causality from inter vention at conventional levels of significance. Table 5 also presents the results of the same causality tests when the sample was split at the end of 1985 and the second subperiod ends at the close of 1986. Hung (1991) suggests that the impact of U.S. intervention on unexpected vol atility changed over these periods in response to different market conditions, as discussed above. U.S. and Japanese intervention Grangercauses yen/$ spreads for the first subperiod. No such effect is found for the three other tests. It is well known, however, that such tests should not be interpreted in terms of structural models. IV. Summary In a recent article, Bossaerts and Hillion (1991) present evidence that tests of forward market efficiency that ignore variation in the bid-ask spread are biased, at least for currencies in the European Monetary System. They observe that spreads are wider on Fridays and speculate that this may be due to anticipation of central-bank ■ 16 Tests of whether spreads Granger-cause intervention would need to be strongly qualified due to the nature of the distribution of the in tervention variables (many observations are clustered at zero). This prob lem, however, does not invalidate the tests for Granger-causality from http://fraser.stlouisfed.org/ intervention. Federal Reserve Bank of St. Louis intervention to see if it can explain intraweekly patterns in G-3 spreads. The tests confirm the tendency for Friday spreads to be higher than for other days of the week and also find some evidence of holiday effects. However, there is no evidence that intra weekly patterns in intervention are related to the patterns in spreads. In addition, I find no evi dence to support the conclusion that anticipa tion of intervention widened spreads. Last, Granger-causality tests suggest that intervention generally does not lead spreads. Although I cannot interpret such results in terms of a structural model, previous research has documented that intervention influences risk premia and that conditional variances ex hibit intraweekly variation. Intervention policies at times have been explicitly designed to respond to volatility. Further investigation into the relations among intervention, spreads, and volatility would be greatly facilitated by a struc tural model. References Allen, William A. “A Note on Uncertainty, Trans actions Costs, and Interest Parity,”Jo u rn a l o f Monetary Economics, vol. 3, no. 3 (July 1977), pp. 367-73. Baillie, Richard T. “Econometric Tests of Ration ality and Market Efficiency,” Econometric Reviews, vol. 8, no. 2 (1989), pp. 151-86. ______ , and Tim Bollerslev. “The Message in Daily Exchange Rates: A Conditional-Variance Tal e," Jo u rn a l o f Business a n d Economic Sta tistics, vol. 7, no. 3 (July 1989), pp. 297-305. Baillie, Richard T., and Owen F. Humpage. “PostLouvre Intervention: Did Target Zones Stabilize the Dollar?” Federal Reserve Bank of Cleveland, Working Paper 9203, Febmary 1992. Baillie, Richard T., and Patrick C. McMahon. The Foreign Exchange Market: Theory a n d Econ ometric Evidence. New York: Cambridge University Press, 1989. Baillie, Richard T., and William P. Osterberg. “The Risk Premium in Forward Foreign Exchange Markets and G-3 Central Bank Intervention: Evidence of Daily Effects, 1985-1990,” Federal Reserve Bank of Cleveland, Working Paper 9109Ju ly 1991. Bekaert, Geert, and Robert J. Hodrick. “O n Biases in the Measurement of Foreign Ex change Risk Premiums,” National Bureau of Economic Research, Working Paper No. 3861, October 1991. Black, Stanley W. “Transactions Costs and Vehicle Currencies,” International Monetary Fund, IMF Working Paper WP/89/96, Novem ber 1989. Boothe, Paul. “Exchange Rate Risk and the BidAsk Spread: A Seven Country Comparison,” Economic Inquiry, vol. 26, no. 3 (July 1988), pp. 485-92. Bossaerts, Peter, and Pierre Hillion. “Market Microstructure Effects of Government Inter vention in the Foreign Exchange Market,” Review o f F inancial Studies, vol. 4, no. 3 (1991), pp. 513-41. Dominguez, Kathryn M- "The Informational Role of Official Foreign Exchange Interven tion Operations: An Empirical Investigation,” Harvard University, working paper, Novem ber 1988. ______ . “Market Responses to Coordinated Cen tral Bank Intervention,” Carnegie-Rochester Conference Series on Public Policy, vol. 32 (1990), pp. 121-64. ______ , and Jeffrey A. Frankel. “Does Foreign Exchange Intervention Matter? Disentangling the Portfolio and Expectations Effects,” Har vard University and National Bureau of Eco nomic Research, manuscript, June 1991. Engel, Charles. “The Risk Premium and the Li quidity Premium in Foreign Exchange Markets,” Federal Reserve Bank of Kansas City, Research Working Paper No. 90-07, December 1990. ______ , and Anthony P. Rodrigues. “Tests of International CAPM with Time-Varying Co variances, "Jo u rn a l o f Applied Econometrics, vol. 4, no. 2 (April-June 1989), pp. 119-38. Fieleke, Norman S. “Exchange-Rate Flexibility and the Efficiency of the Foreign-Exchange Markets,”Jo u rn a l o f F in an cial a n d Q u a n titative Analysis, vol. 10, no. 3 (September 1975), pp. 409-28. Flood, Mark D. “Microstructure Theory and the Foreign Exchange Market,” Federal Reserve Bank of St. Louis, Review, vol. 73, no. 6 (November/December 1991), pp. 52-70. Frankel, Jeffrey A., and Kenneth A. Froot. “Using Survey Data to Test Standard Proposi tions Regarding Exchange Rate Expecta tions,” Am erican Economic Review, vol. 77, no. 1 (March 1987), pp. 133-53. Funabashi, Yoichi. M anaging the Dollar: From the P laza to the Louvre. Washington, D.C.: Institute for International Economics, 1989. George, Thomas J., Guatam Kaul, and M. Nimalendran. “Estimation of the Bid-Ask Spread and Its Components: A New Ap proach,” Review o f F in an cial Studies, vol. 4, no. 4 (1991), pp. 623-56. Ghosh, Atish R. “Is It Signalling? Exchange Inter vention and the Dollar-Deutschemark Rate,” Princeton University, working paper, Septem ber 1989Giovannini, Alberto, and Philippe Jorion. “Inter est Rates and Risk Premia in the Stock Market and in the Foreign Exchange Market,” Jou r n al o f International Money a n d Finance, vol. 6, no. 1 (March 1987), pp. 107-23. Glassman, Debra. “Exchange Rate Risk and Transactions Costs: Evidence from Bid-Ask Spreads "Jo u rn a l o f International Money a n d Finance, vol. 6, no. 4 (December 1987), pp. 479-90. Granger, C.W.J., and Paul Newbold. Forecasting Economic Time Series. New York: Academic Press, Inc., 1986. Hakkio, Craig S., and Anne Sibert. “The Foreign Exchange Risk Premium: Is It Real?” Federal Reserve Bank of Kansas City, Research Work ing Paper 90-02, February 1991. Hodrick, Robert J. The Em pirical Evidence on the Efficiency o f Forward a n d Futures For eign Exchange Markets. Chur, Switzerland: Harwood Academic Publishers, 1987. 13 ______ . “Risk, Uncertainty, and Exchange Rates,”Jo u rn a l of Monetary Economics, vol. 23, no. 3 (May 1989), pp. 433-59. Lucas, Robert E., Jr. “Asset Prices in an Exchange Economy,” Econometrica, vol. 46, no. 6 (November 1978), pp. 1429-45. Hsieh, David A. “The Statistical Properties of Daily Foreign Exchange Rates: 1974-1983,” Jo u rn a l o f International Economics, vol. 24, no. 1/2 (February 1988), pp. 129-45. Mark, Nelson C. “Time-Varying Betas and Risk Premia in the Pricing of Forward Foreign Ex change Contracts,” Jo u rn a l o f F in an cial Economics, vol. 22, no. 2 (December 1988), pp. 335-54. Humpage, Owen F. “Intervention and the Dollar’s Decline,” Federal Reserve Bank of Cleveland, Economic Review, vol. 24, no. 2 (Quarter 2 1988), pp. 2-16. ______ . “Central-Bank Intervention: Recent Litera ture, Continuing Controversy,” Federal Reserve Bank of Cleveland, Economic Revieu>, vol. 27, no. 2 (Quarter 2 1991), pp. 12-26. ______ , and William P. Osterberg. “Intervention and the Foreign Exchange Risk Premium: An Empirical Investigation of Daily Effects,” Global Finance Journal, forthcoming, 1992. Hung, Juann H. “The Effectiveness of Sterilized U.S. Foreign Exchange Intervention: An Em pirical Study Based on the Noise Trading Ap proach,” Federal Reserve Bank of New York, Research Paper No. 9117, May 1991. Kaminsky, Graciela, and Rodrigo Peruga. “Can a Time-Varying Risk Premium Explain Excess Returns in the Forward Market for Foreign Exchange?”Jo u rn a l o f International Economics, vol. 28, no. 1/2 (February 1990), pp. 47-70. Klein, Michael W. “The Accuracy of Reports of Foreign Exchange Intervention,” Tufts University, working paper, March 1992. Levine, Ross. “The Pricing of Forward Exchange Rates,”Journal o f International Money an d Finance, vol. 8, no. 2 (June 1989), pp. 163-79. Lewis, Karen K. “The Persistence of the ‘Peso Problem’ when Policy Is Noisy "Jo u rn a l o f International Money a n d Finance, vol. 7, no. 1 (March 1988), pp. 5-21. Loopesko, Bonnie E. “Relationships among Ex change Rates, Intervention, and Interest Rates: An Empirical Investigation,” Jo u rn a l o f International Money a n d Finance, vol. 3, no. 3 (December 1984), pp. 257-77. McFarland, James W., R. Richardson Petit, and Sam K. Sung. “The Distribution of Foreign Ex change Price Changes: Trading Day Effects and Risk Measurement,” Jo u rn a l o f Finance, vol. 37, no. 3 (June 1982), pp. 693-715. Obstfeld, Maurice. “The Effectiveness of ForeignExchange Intervention: Recent Experience,” National Bureau of Economic Research, Work ing Paper No. 2796, December 1988. Osterberg, William P. “Intervention and the Risk Premium in Foreign Exchange Rates,” Federal Reserve Bank of Cleveland, Working Paper 8908, August 1989. Overturf, Stephen Frank. “Risk, Transactions Charges, and the Market for Foreign Ex change Services,” Economic Inquiry, vol. 20, no. 2 (April 1982), pp. 291-302. So, Jacky C. “The Distribution of Foreign Ex change Price Changes: Trading Day Effects and Risk Measurement— A Comment, "Jo u r n al o f Finance, vol. 42, no. 1 (March 1987), pp. 181-88. Thaler, Richard. “Anomalies— Seasonal Move ments in Security Prices II: Weekend, Holi day, Turn of the Month, and Intraday Effects,” Jo u rn a l o f Economic Perspectives, vol. 1, no. 1 (Fall 1987), pp. 169-77. Wei, Shang-Jin. “Anticipation of Foreign Exchange Volatility and Bid-Ask Spreads,” Board of Governors of the Federal Reserve System, Inter national Finance Discussion Paper No. 409, August 1991. Bl An Ebbing Tide Lowers All Boats: Monetary Policy, Inflation, and Social Justice by David Altig David Altig is an economist at the Federal Reserve Bank of Cleve land. The author thanks Stephen Cecchetti for useful comments. It is essential that the direction o f public policy be well targeted to the nature o f the problem it is seeking to ameliorate.... But only in the con text o f prudent, noninflationary expansion o f money a n d credit are such improvements likely to be lasting. — Alan Greenspan, December 18, 1991 Introduction During periods of slow growth and rising unem ployment, the dynamics of the economic policy debate inevitably reveal an almost irresistible sen timent for stimulative monetary policies. To cite a current example, the steady march of the unem ployment rate from 5.3 percent in mid-1990 to 7.3 percent as of April 1,1992 has been matched on the monetary policy front by persistent calls for the Federal Reserve to take action that would ensure an economic recovery regardless of any longer-term price-level consequences. The dual circumstances of lower-than-expected inflation and slow growth of the M2 monetary aggregate have reinforced this pressure. At the same time, the reluctance of private-market participants to fully incorporate recent inflation outcomes in their http://fraser.stlouisfed.org/ inflation expectations, coupled with the persistent Federal Reserve Bank of St. Louis steepness of the yield curve, suggests that infla tion fears are very real to the decision-makers whose behavior ultimately determines the course of the economy.1 Still, at times like this, there are always many who feel that the inflationary risk inherent in an aggressive monetary policy is worth taking if such a policy can effectively stimulate economic activity, especially since the costs of recessions and slow-growth periods are unequally distrib uted throughout the population. This sentiment is forcefully expressed in the book H ard Minds, Soft Hearts: Tough-Minded Economics fo r a Just Society, written by economist Alan Blinder of Princeton University (see Blinder [1987]). As the evidence presented in the next section makes ■ 1 The spread between three-month T-bill and 30-year Treasury bond yields reached a record high of 436 basis points in the week ended April 24,1992. With respect to inflation expectations, the following quote is from the April 1992 issue of the Federal Reserve Bank of Cleveland's Economic Trends: ‘‘The P-Star model, which links the trend in M2 growth to future inflation, projects continued downward pressure on the inflation rate through 1993.... Apparently, private forecasters are not as optimistic about the near-term inflationary trends. The Blue Chip consensus forecast shows the GDP im plicit price deflator edging up to slightly more than 3 percent next year.” The first-quarter 1992 number for the deflator indi cates that these forecasts are well founded. 15 clear, unemployment disproportionately burdens lower- and middle-class workers relative to more affluent Americans, while inflation, to the extent that it affects income distribution at all, appears to do just the opposite. In Blinder’s words: Sometimes inflation is piously attacked as the “crudest tax,” meaning that it weighs most heavily on the poor.... On close examination, the “crudest tax” battle cry is seen for what it is: a subterfuge for protecting inflation’s real victims, the rich.... [Elvery bit of evidence I know of points in die same direc tion: inflation does no special harm to die poor.... The meager costs that inflation poses on the poor are dwarfed by the heavy price the poor are forced to pay whenever the nadon embarks on an anti inflation campaign.... (p. 54) Two important features of the evidence to which Blinder refers deserve further comment. First, most of the evidence points to the distribu tion of income rather than to the level of income. The former is a somewhat strange measure of welfare: I would gladly see you gain a zillion dollars of real output if doing so would obtain a billion for me, even if the distribution of our in comes becomes more unequal in the process. Second, and more critically, the evidence cited by Blinder focuses on cyclical fluctuations in eco nomic activity. Few economists believe that lower unemployment can be “traded” for higher inflation in the long run. Consequently, a more accurate statement would be that the meager costs inflation poses on the poor are dwarfed in the short run by the heavy price this segment of the population is forced to pay when the nation embarks on an anti-inflation campaign. Some empirical and theoretical arguments for factoring the long-run costs of inflation into calculations of the “fairness” of anti-inflation policies are presented in section II. These argu ments refer primarily to the resource cost to the average individual and thus do not directly address the fairness issue. However, the argu ments do relate inflation to reductions in the overall level of GDP and hence indirectly bear on welfare considerations, to the extent that the burden of falling income is in the long run shared by the less-than-wealthy. A more direct argument is presented in sec tions III and IV, by way of a simple model that illustrates how the long-run costs of inflation arise due to distortions created by a tax system based on nominal income. Although the world I consider is highly stylized, it captures some key elements of the real world: The tax system is im perfectly indexed for inflation. There are “rich” people and “poor” people. Rich people own capital; poor people do not. The share of the economic pie earned by rich people is larger than the percentage of the total population they represent. Also, inflation raises the tax burden of the rich relatively more than that of the poor and, consistent with empirical evidence, does lit tle to change the distribution of income. Within this model, inflation-induced tax in creases on capital definitely hurt the poor. Because inflation effectively raises the tax on capital, a sustained increase in price-level growth ultimately results in a lower capital stock, reduced output, and lower productivity for all workers. Declining output and productiv ity can be expected to fall especially hard on the poor because they start from a lower stan dard of living to begin with. The example given by this simple model is not provided as an argument for eschewing dis cretionary, short-run stabilization policies as rationalized by variants of the Phillips curve model that serve as the foundation of Keynesian economics — even in its more recent incarna tions.2 Although I am skeptical of the Keynesian framework, neo or otherwise, as a useful guide for policymaking, the purpose of this paper is not to engage in a theoretical or philosophical quarrel with the proponents of activist monetary policy.3 Instead, I attempt to show that the “fair ness” objectives that motivate people to urge the Federal Reserve to “do something” when eco nomic activity drags also dictate that the Fed achieve a long-run goal of maintaining price stability. In broad terms, I am arguing that, if we adopt Blinder’s arguments as a guide to shortrun monetary policy, we should symmetrically adopt procedures that provide a long-run anchor to the price level in order to ensure against the possibility of making the cure worse than the illness. I. Inflation, Unemployment, and the Size Distribution of Income The perception that inflation does no special harm to the poor arises from studies that ■ 2 The volumes edited by Mankiw and Romer (1991) are an excel lent introduction to some of the important works in the “New Keynesian” literature. ■ 3 Those readers who are interested in such a quarrel are referred to Barro (1989). 16 T A B L E 1 The Effect of U n e m p lo ym e n t and Inflation on Incom e Shares Quintile Real Per Capita GNP 1 0.111 2 Post-1983 Trend Lagged Dependent Variable Inflation Unemployment 0.016 (1.1) 0.012 (1.0) -0.076 (3.3) -0.082 (4.7) -0.043 (1.0) 0.694 (6.7) 0.8034 (1.3) -0.122 (1.8) -0.052 (1.5) 0.610 (6.8) 0.9426 3 -0.088 (1.2) 0.014 (1.0) -0.038 (2.0) -0.018 (0.5) 0.669 (6.4) 0.8140 4 0.254 (2.8) -0.022 (1.7) -0.018 (1.0) -0.070 (2.2) 0.396 0.8143 0.003 (0.01) -0.041 (1.2) 0.175 (3.5) 0.123 (1.2) 0.700 (7.5) 5 Adjusted R 2 (2.9) 0.8501 NOTE: Standard errors are in parentheses. SOURCES: U.S. Department of Commerce, Statistical Abstract o f the United States, 1990, and Economic Report o f the President, 1991. examine the effects of macroeconomic variables on the share of income received by distinct population quintiles. These share data, col lected and reported by the U.S. Department of Commerce, are obtained by ranking the income of all households from lowest to highest and calculating the percentage of total income that accrues to the first (lowest-income) one-fifth of households, the second one-fifth of households, and so on, up to the last one-fifth, who have the highest incomes in the population. The effect of macroeconomic activity on these income shares can be seen by examining the re sults of the regressions reported in table 1. The regressions measure the effect of unemployment and inflation on the income share of each popula tion quintile after controlling for the level of per capita income, lagged share values (essentially a catchall for the effects of omitted variables), and a shift in the income distribution that appears to have occurred subsequent to 1983.4 The results in table 1 indicate that the burden of unemployment clearly falls on the lowerincome quintiles. The jobless rate is negatively related to the share of income received by the three lowest-income quintiles and is positive for the upper two.5 Inflation, on the other hand, has no statistically significant effect on the dis tribution of income. As indicated by the Blinder quotation in the introduction, these results are consistent with http://fraser.stlouisfed.org/ the bulk of the evidence on income inequality Federal Reserve Bank of St. Louis in the United States.6 However, the information provided by studies of this sort is of a very par ticular type. Specifically, the regression results indicate only that, on a year-to-year basis, infla tion does not reduce the relative share of in come received by the lower-income quintiles. They do not tell us anything about the long-run effects of sustained inflation on the level of in come for any particular income class. In fact, if inflation has adverse effects on the long-run level of income, the poor may indeed be hurt — and perhaps hurt disproportionately in utility terms — even though their relative ■ 4 This regression model follows that reported in a recent paper by Cutler and Katz (1991). Although I use a different sample period than they do, the results in table 1 are qualitatively sim ilar to their findings. The post-1983 shift toward greater inequality in income distribution is an interesting phenomenon that appears to have resulted from a significant structural shift in the employment patterns of skilled versus unskilled labor. I recommend the Cutler—Katz paper to those readers interested in a thorough discussion of this change. ■ 5 Note that, by construction, the income shares over all five quin tiles must sum to one. Thus, a significant negative effect of some variable on the income share of one group must be offset by positive effects on one or more other quintiles. ■ 6 Buse (1982) finds a sim ilar resultfor Canada. Specifically, he discovers that inflation does not significantly affect the share of income received by different income quintiles. Interestingly, neither does he find a significant effect arising from unemployment rates. However, other labor market variables, specifically the employment and labor participa tion rates, are found to influence income distribution, with greater em ployment and participation related to less income inequality. income shares are not reduced. I turn now to a brief overview of the empirical evidence on the relationship between inflation and the long-run level of output. II. Is Inflation Harmful to the Economy in the Long Run? A recent study by Charles T. Carlstrom and William T. Gavin of the Federal Reserve Bank of Cleveland attempts a direct comparison of the welfare impli cations of the effects of disinflationary policies in both the short and long run (see Carlstrom and Gavin [1991])- The authors argue that, in terms of forgone output for the average individual, the long-run “shoe-leather” costs of a steady 4 percent inflation rate are similar in magnitude to the shortrun costs that would typically be attributed to a tight-money policy that reduced the rate of infla tion from 4 percent to zero.7 More generally, simple correlations do suggest that economic growth is negatively related to infla tion. Using data from the International Financial Statistics, Gomme (1991) reports that “...62 of 82 countries exhibit a negative correlation between inflation and per capita real output growth.” More complicated statistical examinations — essentially regressions of cross-country growth rates on a variety of political and economic variables— yield mixed conclusions. But, as convincingly argued by Levin and Renalt (1991), nonrobustness appears to be a generic weakness of the methodology employed in such studies. Two features of these cross-country studies may help to explain this nonrobustness. First, there is a subtle point to be made here about the correlations between growth and inflation. In standard neoclassical growth models, the growth rate of the economy is exogenous and constant. In particular, the growth rate of in come is not affected by inflation even though the level of income is.8 Thus, the absence of a significant correlation between inflation and the long-run growth rate of the economy does not necessarily imply that a particular level of infla tion will fail to reduce per capita income below the level attainable at lower inflation rates. Second, the relationship between inflation and long-run economic performance may operate through indirect and complicated channels. One such possibility is the interaction between infla tion and the tax system. Although indexing has been partially implemented in many countries, in cluding the United States, extant indexing schemes are generally insufficient to remove the distortions created by inflation/tax interactions.9 Although it is true that such interactions provide revenue that might be channeled to productive uses by funding desirable government expenditures or by reducing the level of government debt, research in progress by Charles Carlstrom and me suggests that allow ing inflation to interact with the structural tax sys tem is not an efficient way to raise revenue.10 In the next section, I examine a simple model economy in which inflation distortions arise through exactly this channel. Specifically, inflation is allowed to interact with a tax system based on nominal wage and capital income. The model is chosen to illustrate a rather straightforward point — that inflation can have deleterious long-run effects on the economic well-being of both the rich and poor, without affecting either the growth rate of the economy or the distribution of income. ■ ■ 7 Shoe-leather costs are defined as the value of real money balances that would be held by individuals if the inflation rate were zero instead of 4 percent. An even more dramatic comparison of the welfare costs of short-run versus long-run changes in economic resources, although one not directly related to inflation, was given by Robert E. Lucas, Jr. in his 1985 Vrjo Jahnsson Lectures (see Lucas [1987], section III). He posed the following question: What is the maximum percentage of per-period consumption a rep resentative individual would willingly give up in exchange for 1) a complete smoothing of short-run (or cyclical) fluctuations in consumption or 2) an in crease in the long-run (or trend) growth rate of consumption from 2 to 3 per cent? Using plausible values for individual risk preferences, volatility in consumption, and so on, Lucas argues that the amount of consumption that would be forgone in exchange for higher long-run consumption growth is several hundred times the amount that would be given up to eliminate shorthttp://fraser.stlouisfed.org/ run fluctuations. Federal Reserve Bank of St. Louis 8 The assumption of exogenous, or policy-invariant, growth rates typical of the neoclassical growth framework presented here has recently been challenged by proponents of so-called endogenous growth models. Good overviews of the neoclassical and endogenous growth frameworks can be found in two papers by Sala-i-Martin (1990a, 1990b). A short and informal presentation of the issue is provided in an article entitled “Eco nomic Growth: Explaining the Mystery," published in the January 4,1992 edition of The Economist. See also Mankiw, Romer, and Weil (1990) for a skeptical empirical assessment of the endogenous growth framework. ■ 9 ■ 10 See Altig and Carlstrom (1991b). This message is implicit in Altig and Carlstrom (1991a). Bear in mind that we are not referring to issues related to seigniorage, or the “ inflation tax,“ per se. See Cooley and Hansen (1989,1991) and Gomme (1991) for recent analyses of the welfare implications of revenue collec tion through seigniorage. U III. A Simple Model11 To illustrate the argument, I present a simple general-equilibrium framework that admits two types of individuals: those who earn income solely through wages and those w ho earn both labor and capital income. Each of the groups arises endogenously as a result of its preferences. Members of the first group, who earn only labor income in equilibrium, allocate their earnings according to their own life-cycle consumption needs. Those in the second group care not only about their own life-cycle consumption, but also about their children’s consumption. These altruistic impulses effectively make the planning horizon of this group infinite. They therefore have a much stronger motive for saving than the first group and, in equilibrium, end up owning the entirety of the economy’s capital stock. For simplicity, and with obvious motiva tion, the first group will be referred to as “poor” and the second will be referred to as “rich.”12 Each generation in this model lives, with abso lute certainty, for two periods, which I refer to as the young and old phases of life. Labor is inelastically supplied in each period, and the productivity of labor, identical for rich and poor, is the same when young and old. I assume that a fraction £ of each generation is rich and 1-e is poor.13 The population growth rate is assumed to be zero, and the aggregate capital stock, wages, and the interest rate are determined by 1) the aggregate production ■ 11 The model developed in this section is similar to that presented in section V(b) of Altig and Davis (1992). ■ 12 Some readers may be uncomfortable with the model's implica tion that rich people “care” about their children but poor people do not. Such an implication, however, is more apparent than real. First, the group I have designated as poor (because they have no capital income) is pre sented as nonaltruistic for convenience only. As long as the degree of al truism is lower for one group than the other, the equilibrium outcome will be such that the group with the higher degree of altruism will own the en tire capital stock, even if it is more altruistic by an infinitesimally small amount. Second, a more general model than the one I use here could allow the effective degree of altruism to be related to an individual’s level of wealth. Thus, a framework in which bequest levels depend on the serendipitous mortality history of a given family line could result in the same type of sorting I exploit here, even though the utility functions of all individuals are identical. 13 Mankiwand Zeldes (1991) reportthat, in 1984, some portion of wealth was held as stock for approximately 25 percent of the families sur veyed in the University of Michigan’s Panel Study of Income Dynamics. (This figure does not include equity implicitly held through pension plans.) These families accounted for approximately 40 percent of total disposable income. As described below, our model will be parameterized such that 25 percent of the population holds capital, with the shares of income accruing to the rich http://fraser.stlouisfed.org/ and poor according fairly closely with this evidence. technology, 2) the government’s tax and expen diture policy, and 3) the saving and consump tion decisions of the two groups. The government raises revenue by applying a uniform flat tax rate, p , to nom inal labor and capital income. In other words, the tax code is not indexed for inflation. Although the actual U.S. personal tax code is partially indexed, ad justments for inflation are far from perfect. In particular, the indexing provisions in the current tax code would not vitiate the overstatement of capital income that is critical for the results reported here.14 Denoting variables associated with the rich by superscript R and those for the poor by super script P, the government’s budget constraint is (1) c,+ r f + r f = p [ (r i +7ii M ,+ ( l + 7Tr) U’tLt\, where Gt represents government purchases of output, Tf and T f are transfer payments to the rich and poor, respectively, rt is the real return to capital (the interest rate), wt is the real wage, n, is the exogenously determined infla tion rate, At is aggregate capital holdings, and Lt is aggregate labor supply.1’ Government spending is not productive, nor does it substitute for private consumption. In what follows, I examine the steady-state, or long-run, effects of a change in the inflation rate on the level of income and lifetime consumption of the rich and poor. Subscripts indicating time periods will therefore be dropped. To further streamline the presentation, superscripts denoting rich and poor will be suppressed except when necessary. Readers who have no special interest in the details of the model can, without loss of continuity, skip to the next section, which presents the numerical results. The utility function of each individual who, in equilibrium, is rich is given by (2) U (c v c2, Uk ) = In (q ) + p/rc (c2) + y U\, where cx and c2 denote own consumption in the first and second period of life, (3 is a subjec tive time-discount factor, U*k is the maximum attainable utility of the individual’s child, and y is the rate at wrhich a parent discounts his or her ■ Federal Reserve Bank of St. Louis ■ 14 See Altig and Carlstrom (1991b) for a more detailed discussion of inflation indexing in the U.S. personal tax code. The corporate tax code contains no indexing provisions. ■ 15 There is no “money” in the model. Inflation is introduced as the exogenous rate of depreciation of an arbitrary unit of account. child’s utility. If y = 1 , parents weight their child’s utility equally to their own. Using analogous notation, the utility function of each individual who, in equilibrium, is poor is For both groups, the intertemporal first-order condition for utility maximization is given by (3) For the group with preferences given by equa tion (2), the first-order condition governing intergenerational transfers, g, is U (c v c2) = /«(Cj) + $ ln (c 2) . Equations (2) and (3) are maximized subject to the budget constraints (11) (12) (4) q + g + Tx= w [1 —p (1 + 7t) ] + a and (5) c2 + T2 + b = w\l - p (1 + 7i) ] + { 1 + r[l - p (1 + 7t) ]} « , where g represents transfers received by children, b represents transfers given by parents, and a represents asset holdings. Note that b = g = 0 for individuals with preferences given by equation (3). Also, recall that a p = 0 in equilibrium. Production is undertaken by profit-maximizing, competitive firms that apply competitively ob tained capital and labor inputs to a CobbDouglas technology, given by (6) y - Ke, where y and k are, respectively, per capita out put and the per capita capital stock, and 0 is a parameter that measures capital’s share of total output. The profit-maximizing conditions of firms imply that the aggregate wage and interest rate are given by (7 ) r - 6k0-1 and (8 ) w = ( 1 - Q ) k *. Along with the government’s budget constraint given in equation (1), the specification of the model is completed by the goods-market and capital-market clearing conditions. Because capi tal does not depreciate and the population is sta tionary, government purchases and aggregate consumption, C, exhaust total output. The capital stock is simply the sum of asset holdings by the rich and poor, with the latter, once again, being zero in equilibrium. The two market-clearing con ditions are thus given by (9) y = G+ C and http://fraser.stlouisfed.org/ (1 0 ) k = eaR+ (1 - e) ap . Federal Reserve Bank of St. Louis c2 = (3 (1 + r) cx . c2 = y c l k , where c]k is the children’s first-period consump tion. Because every generation’s consumption is the same in a steady-state equilibrium, clk = cv Equations (12) and (13) thus imply that (3 (1 + r) = y when the transfer motive is operative for the group with preferences indicated by (2). Com bined with equation (7), this condition implies that the per capita capital stock is given by (13) K = r i= J L r ± llf iJ E le ^ L e p y ( l- p ) -I IV. How Inflation Hurts the Poor The model is constructed so that the effects of inflation work through interactions with the tax system. Thus, as is clear from equation (1), an increase in the inflation rate (it) raises the amount of revenue collected by the government even when real income (r A + w) is unchanged. Because the model does not incorporate government debt, satisfaction of the government budget constraint requires either an increase in government expen ditures, an increase in transfer payments, or some combination of the two. Aggregate and individual consumption levels thus depend on the nature of the fiscal policy regime. The results of three distinct fiscal policy ex periments are presented in this section. In the “benchmark” model, tax revenues and govern ment purchases of real output are endogenously determined. A second case, which I refer to as the “progressive-transfer” model, maintains a constant, exogenous level of government pur chases, transferring all surplus revenues to the poor. Results in the third case, which I call the “revenue-neutral” model, are obtained by assuming a constant level of government pur chases, with all surplus revenues used to in crease transfer payments such that the net tax payments of each cohort remain constant. Equa tions (4) to (13) can be combined to obtain con sumption levels under each of the fiscal regimes. The solutions are given in the appendix. T A B L E 2 Simulated Steady-State Effects of Inflation/Tax Interactions Income Share Consumption Loss from Inflation (percent) Model Rich Poor Rich Poor Benchmark model, zero inflation 0.44 0.56 __ __ Benchmark model, 4 percent inflation 0.44 0.56 3.1 2.2 Progressive transfer, 4 percent inflation 0.43 0.57 3.1 0.0 Revenue neutral, 4 percent inflation 0.44 0.56 1.5 1.3 SOURCE: Author’s calculations. invariant to the rate of inflation. Despite this, the lifetime consumption opportunities of the poor fall by as much as 2.6 percent. Only when all surplus revenues from inflation are trans ferred to the poor is this group unharmed by in flation. And even in this case, their lot is not improved. It is clear, then, that evidence regard ing income distribution is of limited value as a measure of the welfare consequences of infla tion on the poor.1 More directly, the poor are decidedly hurt by inflation, even though these adverse conse quences do not manifest themselves in lost in come shares. It is possible that in a more fully articulated model, the poor might actually gain in the short run. However, if the effects of infla tion emphasized here capture some important part of economic reality, such a gain would be transitory. If inflation is harmful in the long run, the less affluent will not be exempt. V. The Moral of the Story The results of the three distinct fiscal policy experiments, presented in table 2, are obtained as suming that capital’s share of output is 25 percent (0 = 0.25), the productivity factor in each period of life ( a ^ , a / , a 2p, and a / ) equals 0.25, the subjective discount factor ((3) equals 0.778, the in come tax rate is 20 percent (p = 0.20), 75 percent of the population is poor and 25 percent is rich (e = 0.25), and the rich weight the utility of their children equally to their own (y = l ).16 For each of the experiments, I calculate the relative share of income received by the rich and poor populations, as well as the change in lifetime consumption for each group in a steady state as the rate of inflation is increased to 4 per cent from the benchmark case with zero infla tion. The results in the second row of table 2 are obtained from the benchmark model (with 4 percent inflation), the results in the third row correspond to the progressive-transfer model, and the results in the fourth row are obtained from the revenue-neutral model. Table 2 conveys the central message of this paper: The distribution of income, as measured by relative shares of personal income (total out put less government purchases), is virtually This paper is a cautionary tale for the “soft hearted”: Attempts to alleviate the burden of unemployment on the less well-to-do through expansionary monetary policy may hurt the clien tele it is supposed to serve if, ultimately, the policy leads to higher long-run rates of inflation. This study is not, however, a criticism of fine-tuning attempts per se. Current Fed policy may or may not fall vic tim to the “too much, too late” syndrome (that is, too rapid an expansion of the money supply at too late a stage in the slowdown to prevent upward pressure on the price level once the recovery begins in earnest). But if policy mistakes do occur, shortrnn monetary medicine could further harm those who are most affected by recession, slow growth, and diminished income levels. Fortunately, the presumed trade-off between a monetary policy that responds to short-run eco nomic circumstances and one that maintains price stability in the long run is a false exchange. By set ting long-run price-level targets collateralized with credible and clearly articulated enforcement mech anisms, the Fed would be free to pursue stabiliza tion efforts aggressively without destabilizing inflation expectations or ultimately risking higher- ■ ■ 16 Although the results are sensitive to the choice of e, this value accords fairly well with evidence concerning the actual distribu tion of income. The poor segment of the model population receives a higher share of total personal income than the rich, but the poor repre sent three-quarters of the population. The rich, who make up only one quarter of the population, receive almost 44 percent of personal income. http://fraser.stlouisfed.org/ See footnote 13. Federal Reserve Bank of St. Louis 17 A 2.6 percent reduction may not seem like much, especially when stacked against the potential costs of unemployment. But 2.6 per cent of lifetime consumption may be larger than you think. With a sus tainable real consumption level of $20,000 per year, a 55-year planning horizon, and a 5 percent real rate of return, a loss of this magnitude would be equivalent to a current lump-sum tax on the order of $10,000, or half a year’s consumption. 21 than-desired inflation paths that are difficult to reverse after the fact. Creating such a policy environment is, of course, easier said than done, but certainly no more difficult than determining an effective way to exploit notoriously slippery Phillips curve trade-offs. Furthermore, institutional rules that advance price stability while maintaining flexi bility over monetary policy choices in the short run do exist. William Gavin, of the Federal Reserve Bank of Cleveland, and Alan Stockman, of the University of Rochester, have recently presented such a proposal (see Gavin and Stockman [1992]). This, and related work, de serves the attention of anyone interested in the long-run welfare of rich and poor alike. where xNtT is all tax revenues net of capital in come taxes paid by the rich, C p is the total con sumption by the poor, and Z is the (exogenous) aggregate labor supply. Note that C R is obtained by first solving for consumption by the poor. ProgressiveTransfer Model For the poor: f e t + p Z k (r+ „ ,l n) 2 ( 1 —e) Appendix ~ T| + P <p r 2 Consumption Solutions for the Alternative Fiscal Regimes „ p (r+7i) r\ (oCj + oc2) [1 - ( 1 + an exogenous lump-sum tax payment of the poor when young, and x2 is an exogenous lump-sum tax payment of the poor when old. Consumption by the rich is k0 - (1 - e) C p - G] e (l+ y ) RevenueNeutral Model n ) p] w For the poor: <ptt + P) 1+ where (p = l + r ( l - p ) - p 7 t . The consumption solution for the rich is f ( 1 - (p) [ Z k 8 - (1 - e ) Cp] ^ l e ( 1 + y ) [ 1 - c p - p ( r + ju) ] e p (p ( a ] + q 2) ( p + 7i) p (r+7i) = ( l + r) ---- --- , r 2= l + ------ , X j is c, = In the benchmark model, government expendi tures are endogenous. The poor’s first-period consumption is , 2(1 -e) r , + p <pr 2 where y [Z Benchmark Model cn =■ r £T + p Z K ( r + 7 t ) ^ This appendix presents steady-state consump tion solutions for the rich and poor when young (that is, for q ^ a n d q * ). Solutions for old-age consumption are given by these expressions and equation (11). Asset levels are then given by equations (4) and (5). Superscripts indicating rich and poor are suppressed except where absolutely necessary. q w (fj a , + a 7 T ,) c, = - xNHT £ ( l + y ) [ l — cp — p ( r + 7 t ) ] . 1 +P a. (« ii + T— \+ -r ) w ~ (h1 + 71 T + "r ) Given the consumption solutions for the poor, the consumption solutions for the rich have the same form as in the progressive-transfer model. References Altig, David, and Charles T. Carlstrom. “Inflation, Personal Taxes, and Real Output: A Dynamic A n a ly s is Jo u rn a l o f Money, Credit, a n d Banking, vol. 23, no. 3, part 2 (August 1991a), pp. 547-71. ______ , a n d _______ . “Bracket Creep in the Age of Indexing: Have We Solved the Problem?” Federal Reserve Bank of Cleveland, unpub lished manuscript, 1991b. Altig, David, and Steve J. Davis. “The Timing of Intergenerational Transfers, Tax Policy, and Aggregate Savings,” Am erican Economic Review, forthcoming, 1992. Barro, Robert J. “New Classicals and Keynesians, or the Good Guys and the Bad Guys,” Na tional Bureau of Economic Research, Working Paper No. 2982, May 1989Blinder, Alan S. H ard Heads, Soft Hearts: ToughM inded Economicsfo r a Just Society. Reading, Mass.: Addison-Wesley Publishing Co., 1987. Buse, Adolf. “The Cyclical Behaviour of the Size Distribution of Income in Canada: 1947-78,” C anadian Jo u rn a l o f Economics, vol. 15, no. 2 (May 1982), pp. 189-204. Carlstrom, Charles T., and William T. Gavin. “Zero Inflation: Transition Costs and Shoe-Leather Benefits,” Federal Reserve Bank of Cleveland, Working Paper 9113, October 1991. Cooley, Thomas F., and Gary D. Hansen. “The Inflation Tax in a Real Business Cycle Model,” Am erican Economic Review, vol. 79, no. 4 (September 1989), pp. 733-48. ______ , a n d _______ . “The Welfare Costs of Moderate Inflations,” Jo u rn a l o f Money, Credit, a n d Banking, vol. 23, no. 3, part 2 (August 1991), pp. 483-503. Cutler, David M., and Lawrence F. Katz. “Macroeconomic Performance and the Disadvan taged,” Harvard University, unpublished manuscript, September 1991. Gavin, William T., and Alan C. Stockman. “A Price Objective for Monetary Policy,” Federal Reserve Bank of Cleveland, Economic Com mentary, April 1, 1992. Gomme, Paul. “Money and Growth Revisited,” University of Western Ontario, unpublished manuscript, November 1991. Levin, Ross, and David Renalt. “A Sensitivity Analysis of Cross-Country Growth Regres sions,” World Bank, unpublished manuscript, March 1991. Lucas, Robert E., Jr. Models o f Business Cycles. New York: Basil Blackwell, 1987. Mankiw, N. Gregory, and David Romer, eds. New Keynesian Economics, vols. 1 and 2. Cambridge, Mass.: MIT Press, 1991. ______ , _______ , and David N. Weil. “A Contri bution to the Empires of Economic Growth,” Harvard University, unpublished manuscript, September 1990. Mankiw, N. Gregory, and Stephen P. Zeldes. “The Consumption of Stockholders and Non stockholders, vJo u rn a l o f F in an cial Econom ics, vol. 29, no. 1 (March 1991), pp. 97-112. Sala-i-Martin, Xavier. “Lecture Notes on Econom ic Growth (I): Introduction to the Literature and Neoclassical Models,” National Bureau of Economic Research, Working Paper No. 3563, December 1990a. ______ . “Lecture Notes on Economic Growth (II): Five Prototype Models of Endogenous Growth,” National Bureau of Economic Re search, Working Paper No. 3564, December 1990b. 23 Sluggish Deposit Rates: Endogenous Institutions and Aggregate Fluctuations by Joseph G. Haubrich Introduction The interest rates that banks pay on deposits move more slowly than money-market interest rates, a phenomenon documented in several recent studies (Flannery [1982], Hannan and Berger [1991], and Neumark and Sharpe [1992]). Understanding deposit-rate sluggishness has im portant direct consequences for comprehending money demand and bank profitability, as well as indirect consequences for understanding almost all industrial pricing. However, even when this recent work takes an explicitly microeconomic approach, it does not consider market conditions that lead to the exis tence of banks. It may therefore distort the lessons of sluggishness both for macroeconomics and for industrial structure. This paper approaches the Issue in terms of the microfoundations of banking. Although this theory may not be all-inclusive and may work in combination with other effects, ignoring it may mean that previous explanations of interest-rate sluggishness are misleading and that attempts to draw parallels with other indus tries regarding price rigidities could be biased. The sluggish adjustment of bank interest rates relative to prevailing market rates, as shown in http://fraser.stlouisfed.org/ figures 1 and 2, has puzzled economists since at Federal Reserve Bank of St. Louis Joseph G. Haubrich is an economic advisor at the Federal Reserve Bank of Cleveland. The author thanks Peter Garber, Robert King, Jeremy Siegel, and James Thomson for helpful criticism. least the mid-nineteenth century. Figure 1 com pares the savings bond deposit rate with the commercial paper rate from 1840 to 1899. Figure 2 compares the same rate paid on savings bank deposits with the interest rate charged on call money from 1857 to 1899- In both cases, the bank rate shows substantially less movement than the market rate.1 In fact, bank interest rates appear to be even more rigid than predicted by this paper. The stability of nominal rates, even in the face of the inflation of the 1850s and the deflation preceding resumption of the gold standard in 1879, suggests that for some reason, interest rates did not index to the inflation rate or to the money supply. Many of the price and nonprice constraints producing macroeconomic behavior originate not from an auction market, but from an organi zation. Banks, labor contracts, and corporations set interest rates, wages, and prices. I contend that such institutions arise to solve problems of risk and private information— precisely those problems associated with a recession, which ■ 1 For evidence on twentieth-century inflexibility, as well as explana tions based on exogenously motivated banks, see Flannery (1982), Klein (1972), Weber (1966), and the references cited therein. 24 F I G U R E 1 Regular Deposit Rate and Commercial Paper Rate, Yearly Averages Percent io 14 12 - I 10 - Commercial A paper rate I 8 -A 6- f \ 1 / -- \ \ / \ J 420 1840 / Savings bank deposit rate 1845 1850 1855 1860 1865 1870 1875 1880 1885 1890 1895 1900 1855 I860 1865 1870 1875 1880 1885 1890 1895 1900 SOURCE: Homer (1977). F I G U R E 2 Regular Deposit Rate and Call Money Rate, Yearly Averages Percent 16 14 12 10 8 6 4 2 0 1840 1845 1850 a. Data were unavailable prior to 1857. SOURCE: Homer (1977). changes the uncertainty that is the very basis of the institution. Thus, the equilibrium prices faced by agents adjust in a way that no market could mimic. Individual agents respond to a macroeconomic shock only after it has been filtered through an organization. Derivative markets then react and alter individuals’ response to disturbances. This paper builds on the recent informationbased banking models of Diamond and Dybvig (1983), Smith (1984), and Haubrich and King (1990). As in those papers, banks in this model arise endogenously in response to a demand for insurance against private risk. Banks are the optimal contract arising from uncertainty. The macroeconomic approach leads to some modi fications, however. These changes should pro vide a picture of banks that can be more easily and realistically integrated with aggregate fluc tuations. Diamond and Dybvig introduce a basic insurance-theoretic banking model in which the bank insures individuals facing a privately observable preference risk: Some individuals die 25 early and therefore need to consume early. Be cause it is costly to remove goods from storage early, such individuals face a liquidity problem. A deposit bank, by setting proper interest rates, can pool the risk between those who die and those who survive. The present paper makes several changes in that basic structure. First, the uncertainty generat ing the bank is somewhat different. The privately observed shock alters endowments, not prefer ences, which seems to capture more realistically what actually constrains agents' liquidity. It also seems more plausible that these endowment shocks are correlated with aggregate disturbances. Also, the shock is a continuous random variable. The continuum, in combination with the endow ment risk, allows use of the optimal taxation litera ture deriving from Mirrlees (1971) to provide a clearer picture of the insurance role of banks. This in turn sets the stage for the second and main innovation of the paper: the interaction between the aggregate shock and individual uncertainty. This interaction takes a particular fonn. In creases in the underlying productivity of the economy, leading to higher market interest rates, induce greater individual uncertainty. This assumption has previously been presented in various forms, but it is by no means obviously true. Analysis along these lines produced the neo-Keynesian concept of autonomous invest ment, which is investment driven not by de mand or savings, but by technological advances and the introduction of new products. It plays a prominent role in the business cycle theories of such diverse authors as Robertson (1915) and Hicks (1950), and also shares the property that low values imply a small, uniform advance while high levels mean a divergence of growth across industries and firms.3 The assumption also suggests the effects of aggregate disturbances, such as business cycles, on the distribution of in come. For example, Dooley and Gottschalk (1984) ■ 2 In Diamond and Dybvig (1983), insurance against private preference shocks is complete due to restrictions on preferences. Haubrich and King (1990) analyze a richer environment, in which insurance against privately observable income shocks is desirable. But in the Haubrich—King setup, in surance is incomplete because there is a trade-off between insurance and intertemporal efficiency. Both papers concentrate on the form of the banking contract, not on its interaction with macroeconomic shocks. ■ 3 For applied work justifying the stylized fact of a positive relation between the level of autonomous investment and its dispersion, see the historical section of Schumpeter (1939) or Safarian (1959, chapter 6). For a different view, see Sheffrin (1984). http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis find the variance of weekly earnings to be nega tively correlated with the unemployment rate.4 Some macroeconomic work based on con tract theory' makes similar assumptions. Gross man, Hart, and Maskin (1983) consider shocks that increase the dispersion of the value of mar ginal product. Haubrich and King (1991) posit a link between the size and dispersion of mone tary shocks as an incentive for sticky nominal price contracts. This paper differs in the sense that it intro duces endogenously arising financial institutions as a response to the uncertainty and traces the consequences of those institutions. In section I, the economic environment is specified and the standard representative-agent solution is dis cussed. The forces motivating the endogenous formation of banks are then presented in sec tion II, under the assumption that there are no aggregate shocks. With that analysis in hand, the mutual interaction of banks, private risks, and aggregate shocks is explored in section III. A final section summarizes and concludes. I. The Economic Environment This investigation begins by specifying a hypo thetical stochastic economy with three basic ele ments central to the problem at hand. First, agents face an intertemporal decision problem concerning the correct amounts of storage and consumption. Second, the aggregate opportuni ties vary in a stochastic fashion; that is, there exist shocks common across all individuals. Third, agents face idiosyncratic, privately observable risks concerning their income (endowment). This paper examines the simplest hypothetical econ omy that incorporates these features. The econ omy lasts for three periods, T= 0,1,2. The two consumption periods allow intertemporal choice, and the stochastic intertemporal terms of trade pro vide the aggregate disturbance. There is also uncer tainty due to environmental randomness in T= 1, which is private information. ■ 4 The data show a positive correlation between unemployment and variance of annual earnings, however. More generally, income disper sion across agents appears to be positively associated with growth (see Danziger and Gottschalk [1986]). Robinson (1972) also emphasizes the macroeconomic consequences of the increased dispersion of incomes resulting from growth and technological progress. 26 T A B L E 1 ! Intertemporal Technology The Storage Technology for Three Periods T= 0 Return -1 T= 1 T= 2 1 0 0 R SOURCE: Author. Tastes Agents are identical, with the following constant elasticity of substitution (CES) utility function: (1) U — G (u ) , where u (cp c2) = (c1 1~'/o + (3c2 1~>/a) 07(G_ and G (u ) = 1/(1 - y) w1_Y. Three important parameters specify prefer ences: (3, the discount factor; o, the intertemporal elasticity of substitution; and y , the rate of relative risk aversion toward variation in lifetime wealth. In the economies studied below, agents face un certainty about lifetime wealth, so that we can meaningfully separate attitudes about risk aversion from those concerning the time pattern of con sumption. Once individuals enter period 1, they face neither uncertain income nor risky assets. Thus, agents formulate consumption plans contin gent on the level of lifetime wealth. Lifetime utility, but not the consumption strategy, depends on the risk-aversion parameter y . Endowments Each individual has an endowment of a single good in each period. At periods 0 and 2, all agents have identical endowments and y2 . At period 1, each individual receives a privately ob servable income level ^(0 ) = + 0, where y 1 is the level of per capita income. Consumers know yx at T= 0, and they leam 0 at T = 1. The idiosyncratic component of income, 0, is con tinuously distributed on (0, 0) with density function / ( 0 , x) having E (0) = 0 and E (0 |x) = 0 (x is an aggregate shock discussed later). I assume a continuum of traders indexed at period 1 by the realized value of 0. Thus, the analysis proceeds as if each value of the distri bution is realized (see Judd [1985]). Along with preferences and endowments, the actors in the model have a storage technology, that is, an intertemporal production function that rewards long-term storage. Goods stored in T- 0 pay no net interest if removed in period 1, but pay a gross return R > 1 if left until T= 2, as shown in table 1. This provides a tractable case in which the time paths of investment projects are somewhat irre versible. An alternative motivation Ls that individ uals (banks) cannot costlessly liquidate assets before their maturity. Economywide movements are captured by introducing randomness into the intertemporal technology. R, the technological rate of return, varies positively with the aggregate shock x. Individ uals observe x costlessly and perfectly at T= 0, so that they know R(x) from the beginning. Fur thermore, the distribution of 0 depends on the aggregate shock. A higher value of x induces a mean-preserving spread on the distribution of 0 ,/ (0 ), subjecting agents to more risk. This assumption is designed to capture the view that progress benefits some individuals more than others. Schumpeter (1939) assigns this view a major role: Industrial change is never harmonious advance with all elements o f the system actually moving, or tending to move, in step. At any given time, some industries move on, others stay behind; and the dis crepancies arising from this are an essential element in the situations that develop, (pp. 101-102) Thus, I separate the effects of an aggregate shock into two components. One is an increase in the productivity of long-term storage, whereby a positive x increases R. The other is an increase in the dispersion of the random variable 0. Follow ing Rothschild and Stiglitz (1970), I let the shift put more weight in the tails of the distribution. ’ These effects cause /(0 , x) to become riskier (in the sense of a mean-preserving spread) with increases in x and cause R (x) to increase in x. That is, the shock raises market (or technological) interest rates. Conversely, a negative shock decreases R and reduces the dispersion of 0. This connection between a macroeconomic variable (R ) and a microeconomic variable (the ■ 5 As the authors point out, this sort of mean-preserving spread corresponds to natural economic measures of increasing dispersion. Any risk-averse individual w ill prefer the old distribution, and the new dis tribution w ill equal the old distribution plus a noise term. 27 T A B L E 2 Observation of Shocks for Three Periods T = T = O x realized 1 0 realized T = 2 R Cr) paid off / ( 0, x) known agents' diversifiable risk by exploiting the pro duction structure of the economy. This section abstracts from aggregate shocks in order to examine the nature of the emergent institutions more clearly. Demand for Insurance SOURCE: Author. individual’s endowment risk) is critical in study ing the behavior of optimal bank contracts in this economy. Because individuals can observe jc at 7’= 0, knowledge of R (x) a n d / ( 0 , x) sim plifies the analysis by reducing the problem to comparative statics on the distribution of 0. Addi tionally, this specification abstracts from the uncer tainty about aggregate shocks and instead empha sizes their distributional consequences. I thus con centrate on the direct effects of the aggregate shocks, not on uncertainty about them. To recapit ulate, then, agents observe x, and thus /(0 , x) in period 0, and 0 obtains in period 1 (see table 2). As a benchmark for comparison with later results, consider the macroeconomic effects of an aggregate shock in this economy without contracts. The individual uncertainty about the distribution of income has no effect on aggre gate variables, so it makes sense to examine only the average individual. The increased dis persion caused by the impulse has no effect on aggregate variables: The per capita change in con sumption and savings is the same as if the distribu tion of income had been entirely ignored. The simplicity of this macro model underscores a point generic to models of this class; namely, this simple economy can be understood in an aggre gate sense by ignoring individual differences and by focusing on the average agent. II. Economic Institutions and the Exchange of Risk When facing diversifiable risk, however, agents in this economy will not accept the market struc ture imposed above. The ability to write con tracts at T= 0 means that they can improve upon their initial position by creating a richer institu tional structure. In the simple world considered here, banks arise endogenously to meet that demand for insurance. The bank is able to pool Whether the market system produces a bank, an insurance company, or a security market depends on the information structure of the economy. If 0 were public information, a regular insurance con tract with premiums and payoffs could protect people against the diversifiable income risk. The private character of 0 gives rise to adverse selec tion, however, and mles out such insurance. Still, since I assume that individuals may write con tracts on any observable quantity, there may be some other way to trade risk. In one case, individuals might exchange claims on long-term storage maturing in T= 2 after re ceiving their random income. Unfortunately, this ex post security market provides no improvement over autarky. In equilibrium, arbitrage opportuni ties between production and securities imply that the price of such securities must be one. If a claim on one unit in storage (R tomorrow) sold for more than one, no one would buy it, preferring instead to place one unit in productive storage. If the price were below one, no one would sell (see Diamond and Dybvig [1983] for a more detailed discussion of this point). Selling these bonds is thus equivalent to taking goods out of production. As we have seen, the ability to draw down storage stocks does not eliminate the possibility of low first-period income.6 There is still room for an institution that can provide insurance and pool risk even if private income shocks are unobservable. The Organization of Banking I define a bank as a coalition of individuals, per haps brought together by an entrepreneur, that receives a deposit <5 in T= 0 and pays interest rates r0 from T= 0 to T= 1, and r, from T= 1 to T= 2. Agents can withdraw any fraction of the account in any period. A bank is linear if the ■ 6 I assume that <t> is sufficiently large relative to y 1 and y 2 so that market equilibrium takes place “off the corner" at the aggregate level. That is, individuals w ill want to store some of 0>. Also, O is not so large relative to lifetime wealth that agents wish to deposit in T=1. 28 interest rate paid is independent of the amount in the account. A bank provides agents with a higher level of expected utility than a situation of autarky because the bank partially insures agents against income risk. The provision of in surance is typically incomplete, because the bank faces a trade-off between risk-pooling and the incentives for saving. Relative to the technological return (or, equiv alently, to ex post security markets), banks offer higher short-term yields (rQ> 1) and lower long term yields (r{ < R). This is how banks provide insurance. To determine the interest rates that actually occur, take the analysis one step further and consider the optimal linear bank.7 This bank sets rQand rxto maximize the expected util ity of agents given the total resources of the bank and the decision rules of the individuals. The analysis closely follows the optimal income taxa tion investigations of Mirrlees (1971). An individual must choose consumption and savings withdrawal given the bank’s interest rates r0 (from T= 0 to T= 1) and rx (from T = 1 to T= 2). If r0 > 1, the problem for a rational in dividual begins in period 1: (2) dw * dw * dw * --— > 0, -r— < 0, and - _ < 0 . Recall the drQ dr, d0 assumption (footnote 6) that the initial endow ment is large enough so that the withdrawal will be positive for all 0. The bank, as a coalition of individuals, wishes to maximize the depositors’ expected utility EG [u (0, r0 , rj)] subject to a resource con straint. This constraint, written as equation (3), states that the period 0 present value of assets, , must equal the present value of the liabilities both in period 1, Ew* (0, r0 , r2), and in period 2, rx[r0 O - Ew* (0, r0 , ra)l . (3) O = E w *(Q , rQ, rj) + R~' { r} [r0O - Ew * (0, r0, rx) ] 1. In other words, the bank must be able to cover all withdrawals. Notice that the bank views total withdrawals as certain. Thus, Ew* involves simply “summing” across all depositors. In addi tion to the resource constraint (3), the bank is constrained by the individuals’ decision rules, such as the withdrawal function, which is a function of bank actions r0 and rxas well as 0. max w(cj,c2) Banking and Insurance subject to (i) j / O ) + w = (ii) cv y 2 + r ^ r ^ - iv ) = c 2. The solution to this problem provides four func tions of the income shock and interest rates: an indirect utility function, v (0, rQ, ra); two con sumption functions, c* (0, r0, rx) and c* (0, r0, rx); and an optimal withdrawal function w* (0, rQ, rx). With a CES utility function, indirect utility is linear in wealth, v= CL(r^) a (r0, r-,0). Since w* = What are the characteristics of an optimal bank ing structure? First, consider a small increase in r0 from its initial position of one and a small decrease in rx. The bank must respect its budget constraint, that is, (4) 0 = drQ[ O - (1 / rx - 1 / R ) E(dc*2/d r Q) ] - drx { ( y2-E c2) + (1 / rx - 1 / R ) E [dc2 / a (1 / rx) 11/r\ . c*\~Ti(0), one can straightforwardly show that When evaluated at r, = R, expression (4) be comes simply drQO = drl(y 2 - E c * ) / r * . Since ■ 7 Haubrich and King (1990) examine such a bank, but with an o nreversible storage technology. Consideration of linear institutions un doubtedly simplifies the analysis, but more important, it prevents the formation of depositor coalitions that could arbitrage across nonlinearities in the rate structure. In other words, an interest-rate structure that is non linear in the size of withdrawals would be subject to raiding by coalitions of depositors at 7"= 1. For example, small depositors might combine funds and act as a syndicate to obtain the better rates received by large depositors. This would change the distribution (especially the expected value) of withdrawals and ruin the bank. A budget just balanced, with some individuals obtaining low interest rates, has no room for everyone to receive high rates. A competitive bank simply could not give everyone a http://fraser.stlouisfed.org/ higher interest rate. Federal Reserve Bank of St. Louis 2> Vi ’ a smaH increase in r0 requires a decrease in rv The effects on expected utility can similarly be calculated by differentiation. (5) d U = E ( G 'd v / d r 0 ) drQ + E ( G ' d v/d rx ) drx = E ( G ' a ) O dr0 - E { G ' a [ y 2- c^(0)]} drx/ r x . 29 Expression (5) indicates that increases in r0 have an identical wealth effect on all consumers, a is the marginal utility of a unit of period 1 wealth. As discussed above, a is invariant to 0 under CES utility. By contrast, the wealth effect of an increase in rx is greatest for the largest lenders in period 1, for whom y2 < c* ( 0 ) . Requiring feasibility of d r0 and d rxand rearranging the resulting expression, (6) dU = a E[G'(c*2-Ec*2 )] drx / r\. With risk aversion, G " > 0 , so that the covari ance term is unambiguously negative and a small decline in rx raises welfare. Intuitively, by raising r0 and lowering rx, the bank has shifted wealth from those with high 0 ’s to the average individual. The lucky people with high 0 ’s will attempt to smooth consumption and save the windfall, withdrawing relatively little. The lower rxpenalizes them. The unlucky people with a low 0 withdraw a lot, benefiting from the high r0. This redistribution provides insurance in T= 0, when 0 is unknown. In effect, in period 0, the bank offers an individual a security that 1) has a certain period 1 expected return (Oi/r0), 2) pays negative returns when high 0’s occur, and 3) reduces individual risks. The Optimal Linear Bank The economic intuition behind these results (small changes in r0 and rxfrom the initial posi tion r0 = 1 and rx = R) extends to interpretation of the optimal banking structure. Again, follow ing Mirrlees (1971) and Atkinson and Stiglitz (1980), I derive the result that for the CES case, the optimal level of rxsatisfies the following condition: 3c(7) dc* r1 = S ( E 2 + 82 ~ ) / ( e 2 + 82- j j + /?82 ) 3a dc*, - R - z ( e 2 , 52 , da ), where e2 is the compensated semi-elasticity of second-period consumption with respect to its price, p 2 = — . e2 is a constant because utility is C E S , e2 = ( 1 / c*), and > 0. ap2 dp2 is the effect of a wealth increment on second-period consump tion, and 82 is the risk premium of a private agent for a consumption bet of the form c2 / Ec\ . Such a bet has expected utility of one but covaries negatively with lifetime marginal utility: §2 = - { cov [<?', c2 (0) 1 /E G ' Ec2 }. Notice that risk aversion implies r, < R and thus r0 > 1, both of which preserve the flavor of the local results above. Banks and Other Structures It is worth comparing this bank with the other institutions already discussed. In autarky, each individual agent is subject to income risk. Be cause the technology is reversible, no one bene fits from being able to sell shares in an ex post security market, that is, by transferring goods from T= 2 to r = 1. A simple ex post equity market, then, does not improve upon autarky, because it cannot remove any of the income risk faced by agents. However, the optimal linear banking structure provides agents with a higher level of expected utility than an ex post market does, because it par tially insures agents against income risks. The provision of such insurance is incomplete because the bank pays for insurance by distorting the inter temporal trade-ofif facing consumers. Relative to ex post security markets, banks offer higher short term yields (r0 > 1) and lower long-term yields (rx< R). Without income uncertainty, or with full insurance from another source, the optimal bank would set r0 = 1 and rx= R, and would serve no economic purpose. Notice this classic relation between the bank and asset markets: The bank creates long-term assets from short-term liabilities. Though agents may withdraw money from their account at any time, the bank balances these withdrawals and invests partly in long-term production. A nonclassical restriction is the requirement of a choice of institution. As in other models of this sort (Diamond and Dybvig [19831, Haubrich and King [1990], andjacklin and Bhattacharya [1988]), a bank and an equity market cannot coexist. A more detailed analysis of these questions would proceed by initially characterizing Paretooptimal allocations— subject to resource and incentive constraints— and then asking whether particular market arrangements can effectively decentralize these allocations or yield Paretooptimal quantities as the outcomes of individual choices in a specified market. Because this paper concentrates on the effects of aggregate shocks, and not on the banking contract per se, it will not formalize the mechanism-theoretic approach to this problem. Additionally, a digression here 30 could not do justice to the many interesting issues that arise, and would be redundant in light of the fuller treatment of the banking con tract found in Haubrich (1988) and Haubrich and King (1990). Still, an informal discussion summarizing results from the other papers can clarify several related issues. A key question is which institutions can sup port the optimal allocations arising from the planning problem. A bank contract supports such allocations, as do some other institutions. The main difference concerns the possibility of bank runs. Adding a sequential service con straint, as in Diamond and Dybvig (1983), will create panics. However, banks without this fea ture (and indeed mutual funds issuing derivative securities) can support the optimal allocations and remain immune to panics. I consider only such stable institutions. An equity market does not support the opti mal allocation. Once a bank exists, there are individual incentives to create a stock market. This would ruin the bank, however, so the plan ner does not allow that market to open. This exclusivity seems to be a generic defect of this type of banking model. Haubrich (1988) exam ines the informational assumptions allowing such exclusion. Jacklin and Bhattacharya (1988) interpret banking regulation as a means of pre venting the arbitrage that would destroy banks. Gorton and Haubrich (1987) explore coexis tence using a somewhat different model. Finally, support for the full optimum men tioned above requires a nonlinear bank— one that pays contingent on withdrawal size. The general form of the contract remains the same, and the same techniques can be used to charac terize the interest-rate schedule, but comparative statics become intractable. The linear bank results from the arbitrage conditions discussed above, which in the planning problem take the form of "multilateral incentive compatibility con straints” (see Haubrich [1988]). The nonlineari ties that exist in the real world may result from the inability to arbitrage the bank— perhaps due to transactions costs or to the inability of group members to monitor one another. Still, the linear bank seems a useful approximation. III. Banking with Aggregate Shocks This section reintroduces fluctuations into the economy by integrating the banking sector into the basic macro model. It explores how the aggregate random variable x influences bank interest rates and in turn affects savings and con sumption. This section illustrates the importance of contracts in economies with connections between a macroeconomic variable, R, and a microeconomic variable, individuals’ endow ment risk. Recall that a positive x increases R and induces a mean-preserving spread in /(0 ), while a negative draw lowers R and reduces the dispersion of 0. In the presence of banks, this interaction has important consequences. Individuals can observe x in T= 0, so that knowledge of R(x) a n d /(0 , x) allows calcula tion of the interest rates r0 and rx. This reduces the problem to comparative statics on the dis tribution of 0 and suggests that it is not uncer tainty about aggregate shocks that drives banks’ effects on interest rates, but rather the distribu tional consequences of such shocks. It will be easier to examine these effects in three steps. First, I examine how rx changes with R if the distribution of 0 remains fixed. Next, I keep R fixed and note how rx changes with the dispersion of 0. Finally, I put the two together. Pure Aggregate Shocks The case of an aggregate shock— with no effect on the uncertainty of income— serves as a benchmark for comparison with more compli cated scenarios. With a “pure” aggregate shock, if the underlying technological rate of return R increases, the economy is richer and should be able to support a higher interest rate on bank deposits. This is indeed what happens, since drx/ dR = z ( 5 2 , dc*2 /da, e2) - r, 8 2(e2 + 8 2d c/d a + R 8 ) > 0 . Thus, the direct or “pure” effect of an aggregate shock moves both bank and market interest rates in the same direction. The second term in the equation is model specific: Because the utility function exhibits constant relative risk aversion, the increased income leads consumers to demand less insurance for a given absolute risk. This term would be absent with constant absolute risk aversion. A short calculation re veals that r0 rises with R\economically, because of a higher payoff to storage, the bank can afford to distribute more goods, and both bank and market interest rates increase. ta Pure Distribution Effects The next determination is how banks’ interest rates move when individuals are subject to greater uncertainty. I wish to sign d z / d x ; that is, to hold R fixed, but to allow x to change/(0). Equation (7) tells us rx= z ( 8 2, dc2/ d a ,£ ,)/ ? . Notice that the CES specification makes £2 constant, and the homotheticity of indifference curves implies that dc2/d a is independent of the distribution of 0. This means that the only term changed by a mean-preserving shift in /(0 ) is 82. Not surprisingly, the movement in the interest rate depends on the movement of the risk premium on period 2 consumption. Recall that a greater risk premium indicates a greater demand for insurance, which is provided by a lower interest rate. Notice that 3 r, /0 5 , = —£2R / (£ , + 8 2 + dc*2/d a ) 2 < 0. Thus, a meanpreserving spread will decrease rx if it increases 8^ Since 8, measures the risk premium on c2/E c2 , we expect it to rise with a risker c2 , which in turn is a linear function of 0. Intuitively, a positive shock, say a good harvest, will increase the uncertainty of individual incomes. This drives up 8 ,, the risk premium on the lifetime consump tion gamble, and sends rx down. The bank pools some of the increased risk by pushing rxand r() closer together, hence further redistributing in come from the lucky to the unlucky. The clear intuition on the effects of a meanpreserving spread belies the complexity of the actual calculation. The multiperiod, multiplechoice problem does not fit the one-variable techniques of Rothschild and Stiglitz (1970, 1971). In a closely related problem, calculating the change in the optimal linear income tax with a change in the ability distribution, Stern (1976) resorts to numerical examples even after specifying both utility and distribution func tions. With problems in such a simple case, it is not surprising that more general specifications prove intractable. Calculating the change in 82 is straightforward when G takes the form of log utility.8 This is the only case for which an intertemporal inves tor facing a changing investment opportunity set will act as if he were a one-period maximizer (Merton 11982]). With log utility, changes in the interest rate alone do not alter consumption or savings decisions, and the result is a one-period problem on which standard comparative static ■ 8 The dynamic asset pricing literature often exploits this tractability, http://fraser.stlouisfed.org/ which stems from the offsetting income and substitution effects. Federal Reserve Bank of St. Louis techniques can be used. In this paper, because interest rates differ across periods, individuals face a changing investment opportunity set. With that problem simplified, comparative statics on the bank problem become feasible. The ap pendix carries out the calculation for log utility and examines the robustness of the result. A meanpreserving spread also increases the risk premium in another tractable case, quadratic utility. Another way to obtain results is to restrict the distribution function. The appendix shows that for arbitrary utility functions, a two-point dis tribution yields the required result, as do certain changes related to the martingale measure of risk. Thus, although the general case seems in tractable, a number of specific results support the intuitive conclusion. Micro and Macro Shocks Together The pure aggregate shock moves the underly ing interest rate. The pure distribution effect, on the other hand, increases individual uncertainty and induces people to pool more risk by accept ing a lower interest rate. The combination of both effects means that a macroeconomic distur bance will increase bank interest rates, but by less than the underlying rate. In other words, the aggregate shock x moves R directly, increasing both rx and rQ. In fact, without changes in in dividual uncertainty, an efficient bank would raise rx proportionately with R. The distribution effect by itself lowers rx when x rises. Both effects together imply that rx moves by less than R. Further, we expect that the direct effect dom inates the distributional (indirect) effect, and both rxand R increase (that is, bank rates move less than one-to-one with the underlying inter est rates). Similarly, a negative x decreases R, and the distribution effect raises rx. Again, slug gishness results. Since the two effects of x — an increase in R and a greater dispersion of 0 — are mathematically distinct, we must simply as sume the dominance of the direct effect. This assumption accords with the macroeconomic evidence and theories mentioned in section I. This distribution effect also influences r0. The bank’s budget constraint, (3), implies that a decrease in rx requires an increase in r0. When the dispersion of 0 rises, the bank provides more insurance by increasing r0 and decreasing rx. This affects consumption and savings in two ways: The higher r0 augments the wealth of all agents as of T= 1, and the lower rx makes current con sumption more attractive. These distributional consequences counteract the intertemporal effects of the pure gain in R, which induces people to consume more later. The effect on interest rates is an immediate illustration of how contracts change the qualita tive macroeconomic behavior of this economy. As the intertemporal price, the interest rate has additional effects. In general, comparing the path of aggregate disturbances will be compli cated, but in the case of log utility, simple results emerge. The sluggish adjustment of interest rates dampens the effect of aggregate shocks on consumption and savings. Some lengthy but straightforward calculations show that (8) (9) dc* dc* 0>— zr ~ (bank) > -r-^- (no b a n k ), and ox dx dc* ox (no bank) > dc* ox (bank) > 0. Thus, though idiosyncratic risk “washes out” across all agents, it affects the economy because agents form institutions and write contracts to protect against that risk. Even if interest rates adjust one-to-one, the deviation of the bank rate from the technological rate alters behavior. More significant, however, is that the bank filters the effect of the shock by changing the underlying risk. Hence, ignoring or simply exogenously im posing institutions on a macro model seriously distorts conclusions. Figures 1 and 2 give a flavor of possible applications of this model and show that there are useful and tractable exten sions of the representative-agent framework. IV. Conclusion This paper illustrates how institutions play a cen tral role in aggregate phenomena. In this section, I argue that the results hold in a very general con text and that the general study of institutions aris ing from competition is essential for adequate macroeconomics. The analysis presented above extends beyond bank rates. Other financial institutions play a part in macroeconomic disturbances, and although this paper argues in terms of risk-pooling, the underlying ideas pertain to risk-shifting as well. The institution studied here is termed a bank, but as a pure financial intermediary, its functions may be duplicated by an appropriate derivative secu rity market. For example, consider dividend payments. When individuals face private risks, dividend payments may set the return on equity to pro vide insurance. An interaction between macroand microeconomic shocks leads to dividends that adjust slowly (Copeland and Weston [1979D. In fact, the analysis is not limited to financial institutions: Some recent work on labor con tracts also discusses the role of aggregate shocks as signals about unobservable individual distur bances. Haubrich and King (1991) examine a case in which the money supply signals individ ual dispersion, leading to the non-neutrality of perceived money. Grossman, Hart, and Maskin (1983) focus on economies where asymmetric information between firms and workers pro duces cyclical unemployment. These new markets and institutions attempt to avoid the problems of adverse selection aris ing from private information. In this sense, de rivative security markets or institutions occupy niches similar to other schemes discussed in the literature. In order for the institution to survive, the incentive structures must force agents to reveal themselves at least partially. Markets can not always completely exploit this information, because to do so would distort the incentives that allowed revelation in the first place. This paper provides an equilibrium analysis of how endogenously arising financial institu tions alter the impact of macroeconomic shocks. It explains the modifications in consumption and investment decisions as reactions to prices that react sluggishly to the underlying economic disturbances. This suggests that income distribu tion plays a major role in aggregate disturbances, such as business cycles. It also suggests that a relevant business cycle theory eventually must explicitly model why banks exist and why they take their present form. This explanation of bank rate sluggishness illustrates a powerful principle: When aggregate disturbances also have distributional consequences, the pattern of efficient contract-specified prices can change. Appendix In this appendix, I calculate the change in the risk premium 5 , caused by an increase in indi vidual uncertainty. First, recall that indirect util ity and optimal second-period consumption are (A l) v= a ( r ) 1^(0) ] and (A2) c* = r[ 1 - h (p 2) ] [w(Q) ] = q (r) [*¿>(0) ] . 8? can be written as (A3) 8 2 = - [E(v~v c2) - Ec2 Ev~y}/ E c2Ev~y that in some cases (A4) is positive. Additionally, (A4) is always positive with a discrete, symmet ric, two-point distribution. To see this, write the numerator of (A5) as Ew~ yEw~ y0 + y {Ew '~yEw~y- 'Q - E w ~ yEq~y§ ) . The first term is always negative. I can use the linearity of wealth to express w as (a ± k), where the distribution is the two-point discrete distribution with probability 1/2 on k. and -k. The sign of (A5) is then the opposite of = 1 - E{irlc2)/E c2Ev~'i. (a - k )x~y {a + k )1~y(-4 a) , which is always negative. Thus, the risk premium moves positively Using (A l) and (A2), I rearrange (A3) to obtain with x. When G is quadratic, G (x ) = x - Vi bx2, the (A4) 1 - 8 2 = E [w ($)l -y]/ E[w{$)} E[w{$)~y] . result also holds. Substitute into (A4) to obtain (A 6) To discuss how 5, changes with increases in the dispersion of 0 ,1employ the techniques of Sandmo (1970) and Rothschild and Stiglitz (1970, 1971) and stretch the distribution by replacing 0 with x 0 in order to sign d 8 , / d x . First, take the derivative: d 8 ?/d x = - [Ew(xQ) Ew (xd)~ y(d/dx) Ew (x0 1~y) - E w (x6) 1_Y Ew(xO) ■(d/dx) Ew (x6~ y)] / (EwEu~y) 2. Without loss of generality, I evaluate this expres sion at x = 1. (A5) -{Ew (d)Ew (Q )-yE[( 1 -y) w(Q)~yQ] - Ew (0) - -iEw(6) E[- y w (0)" y~ 10 ]} / (EwEw-y)2. Notice that the first and second terms of this ex pression are positive, as are all the terms after the minus sign (fourth, fifth, and sixth terms). The third term is negative when y < 1, making the entire derivative unambiguously positive. Thus, an increase in x increases 5, and de creases rv When y < 1, the sign of expression (A4) becomes ambiguous. Without explicitly determining its sign, though, we can gain some idea of its properties. Simple numerical exam ples involving uniform distributions indicate 1 - 8 2= e \i - b{ a[ a (a + 0)]} [q(a + 0) ]] E[1 - b (a a + a 0)1 E[q(a + 0)] With a mean-preserving spread on 0, only the numerator of (A6) changes, becoming E\q(1 + 0) ] - b a q E (a 2 + 2aQ) - baqE(Q 2) . The MPS on 0 increases the variance, proving the result. For general utility functions, 1 - 81can be ex pressed as a “martingale measure of risk” as in Nachman (1979, section 4.1). Then, i f / is the dis tribution for c2, G’ G' ----- A c). f* (c ) - — 7 ’ / = EG J G A c) dc Defining E t (c) - \cf* (c) d c , Nachman extends Rothschild and Stiglitz’s arguments to show Ej (c) < E(c). The assumption on the movement from f to g implies E*^(c) < E (c ). Similarly, if g is riskier th a n /* , it is also risker than/ The new expression for 1 - 8 , is £* (c) < E^ (c) < E*/ (c) < Ef(c) . Again, the desired result follows. Here, the function G is general, but a large shift in dispersion is required. E a References Atkinson, Anthony B., and Joseph E. Stiglitz. Lectures on Public Economics. New York: McGraw-Hill, 1980. Copeland, Thomas E., and J. Fred Weston. F inan cial Theory a n d Corporate Policy. Reading, Mass.: Addison-Wesley Publishing Co., 1979Danziger, Sheldon, and Peter Gottschalk. “Do Rising Tides Lift All Boats? The Impact of Secular and Cyclical Changes on Poverty,” Am erican Economic Review: Papers a n d Proceedings, vol. 76, no. 2 (May 1986), pp. 405-10. Diamond, Douglas W., and Philip H. Dybvig. “Bank Runs, Deposit Insurance, and Liquid ity,”Jo u rn a l o f Political Economy, vol. 91, no. 3 (June 1983), pp. 401-19. Dooley, Martin D., and Peter Gottschalk. “Earn ings Inequality among Males in the United States: Trends and the Effect of Labor Force Growth,”Jo u rn a l o f Political Economy, vol. 92, no. 1 (February 1984), pp. 59-89Flannery, MarkJ. “Retail Bank Deposits as Quasi-Fixed Factors of Production,” Am eri can Economic Review, vol. 72, no. 3 (June 1982), pp. 527-36. Gorton, Gary B., and Joseph G. Haubrich. “Bank Deregulation, Credit Markets, and the Control of Capital,” in Karl Brunner and Allan Meltzer, eds., Bubbles a n d Other Essays, CamegieRochester Conference Series on Public Policy, vol. 26 (Spring 1987), pp. 289-333. Grossman, Sanford J., Oliver D. Hart, and Eric S. Maskin. “Unemployment with Observable Aggregate Shocks,” Jo u rn a l o f Political Econ omy, vol. 91, no. 6 (December 1983), pp. 907-28. Hannan, Timothy H., and Allen N. Berger. “The Rigidity of Prices: Evidence from the Banking Industry,” Am erican Economic Review, vol. 81, no. 4 (September 1991), pp. 938-45. Haubrich, Joseph G. “Optimal Financial Stmcture in Exchange Economies,” International Economic Revieu\ vol. 29, no. 2 (May 1988), pp. 217-35. ______ , and Robert G. King. “Banking and In surance,”Jo u rn a l o f Monetary Economics, vol. 26, no. 3 (December 1990), pp. 361-86. ______ , a n d _______ . “Sticky Prices, Money, and Business Fluctuations,”Jo u rn a l o f Money, Credit, a n d Banking, vol. 23, no. 2 (May 1991), pp. 243-59. Hicks, John R. A Contribution to the Theory o f the Trade Cycle. Oxford: Clarendon Press, 1950. Homer, Sidney. A History o f Interest Rates, 2nd edition. New Brunswick, N.J.: Rutgers Univer sity Press, 1977. Jacklin, Charles J., and Sudipto Bhattacharya. “Distinguishing Panics and InformationBased Bank Runs: Welfare and Policy Im pli cations, "Jo u rn a l o f Political Economy, vol. 96, no. 3 (June 1988), pp. 568-92. Judd, Kenneth L. “The Law of Large Numbers with a Continuum of IID Random Variables,” Jo u rn a l o f Economic Theory, vol. 35, no. 1 (February 1985), pp. 19-25. Klein, Michael A. “O n the Causes and Conse quences of Savings and Loan Deposit Rate In flexibility,”Jo u rn a l o f Finance, vol. 27, no. 1 (March 1972), pp. 79-87. Merton, Robert C. “O n the Microeconomic Theory of Investment under Uncertainty,” in Kenneth J. Arrow and Michael D. Intriligator, eds., Handbook o f M athem atical Economics, vol. 2. New York: North-Holland, 1982. Mirrlees, James A. “An Exploration in the Theory of Optim um Income Taxation,” Review o f Economic Studies, vol. 38, no. 2 (April 1971), pp. 175-208. Nachman, David C. “O n the Theory of Risk Aversion and the Theory of Risk/'Jo u rn a l o f Economic Theory, vol. 21, no. 2 (October 1979), pp. 317-35. Neumark, David, and Steven Sharpe. “Market Structure and the Nature of Price Rigidity: Evidence from the Market for Consumer Deposits,” Quarterly Jo u rn a l o f Economics, forthcoming, 1992. Robertson, Dennis H. A Study o f Industrial Fluc tuation. London: P.S. King& Son, 1915. Robinson, Joan. “The Second Crisis of Economic Theory,” American Economic Review. Papers a n d Proceedings, vol. 62, no. 2 (May 1972), pp. 1-10. Rothschild, Michael, and Joseph E. Stiglitz. “In creasing Risk: I. A Definition,”Jo u rn a l o f Economic Theory, vol. 2, no. 3 (September 1970), pp. 225-43. ______ , a n d _______ . “Increasing Risk II: Its Eco nomic Consequences, "Jo u rn a l o f Economic Theory, vol. 3, no. 1 (March 1971), pp. 66-84. Safarian, A.E. The C anadian Economy in the Great Depression. Toronto: University of Toronto Press, 1959Sandmo, Agnar. “The Effect of Uncertainty on Saving Decisions,” Review o f Economic Studies, vol. 37 (July 1970), pp. 353-60. Schumpeter, Joseph A. Business Cycles: A Theo retical, Historical, a n d Statistical Analysis o f the Capitalist Process, vol. 1. New York: McGraw-Hill, 1939. Sheffrin, Steven M. “The Dispersion Hypothesis in Macroeconomics,” Review o f Economics a n d Statistics, vol. 66, no. 3 (August 1984), pp. 482-85. Smith, Bruce D. “Private Information, Deposit Interest Rates, and the ‘Stability’ of the Bank ing System," Jo u rn a l o f Monetary’ Economics, vol. 14, no. 3 (November 1984), pp. 293-317. Stem, Nicholas H. “On the Specification of Models of Optimum Income Taxation,”Jo u r n al o f Public Economics, vol. 6, no. 1/2 (July-August 1976), pp. 123-62. Weber, Gerald I. “Interest Rates on Mortgages and Dividend Rates on Savings and Loan Shares," Jo u rn a l o f Finance, vol. 21, no. 3 (September 1966), pp. 515-21. it i Second Quarter Working Papers Current Working Papers of the Cleveland Federal Reserve Bank are listed in each quarterly issue of the Economic Review. Copies of specific papers may be re quested by completing and mail ing the attached form below. Single copies of individual papers will be sent free of charge to those who request them. A mailing list service for personal subscribers, however, is not available. Institutional subscribers, such as libraries and other organiza tions, will be placed on a mail ing list upon request and will automatically receive Working Papers as they are published. ■ 9206 ■ 9207 Social Security and New Results on the Medicare Policy from the Impact of Central-Bank Intervention on Deviations Perspective of Genera tional Accounting from Uncovered Interest by Alan J. Auerbach, Parity Jagadeesh Gokhale, and Laurence J. Kotlikoff by Owen F. Humpage and William P. Osterberg Please complete and detach the form below and mail to: Research Department Federal Reserve Bank of Cleveland P.O. Box 6387 Cleveland, Ohio 44101 Check item(s) requested Please send the following Working Paper(s): □ 9206 □ 9207 Send to: Please print Name Address City State Zip