The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
orKmg raper series The Implications of First-Order Risk Aversion for Asset Market Risk Premiums Geert Bekaert, Robert J. Hodrick and David A. Marshall Working Papers Series Macroeconomic Issues Research Department Federal Reserve Bank of Chicago December 1994 (W P-94-22) FEDERAL RESERVE BANK OF CHICAGO T h e Implications of First-Order Risk Aversion for Asset M a r k e t Risk P r e m i u m s Geert Bekaert Stanford University Robert J. Hodrick Northwestern University and NBER David A. Marshall Federal Reserve Bank of Chicago and Northwestern University January 14, 1994 The authors have benefitted from the comments of Stan Zin and Andy Abel as well as numerous participants at seminars and conferences. We thank the seminar participants at Duke University, the Federal Reserve Bank of Chicago, Northwestern University, the University of Chicago, the University of Pennsylvania, the University of Southern California, the American Finance Association, a CEPR conference at the University of Limburg, and a conference on Capital Market Integration at the London School of Economics. T h e Implications of First-Order Risk Aversion for Asset Market Risk P r e m i u m s Abstract Existing general equilibrium models based on traditional expected utility preferences have been unable to explain the excess return predictability observed in equity markets, bond markets, and foreign exchange markets. In this paper, we abandon the expected-utility hypothesis in favor of preferences that exhibit first-order risk aversion. We incorporate these preferences into a general equilibrium two-country monetary model, solve the model numerically, and compare the quantitative implications of the model to estimates obtained from U.S and Japanese data for equity, bond and foreign exchange markets. Although increasing the degree of first-order risk aversion substantially increases excess return predictability, the model remains incapable of generating excess return predictability sufficiently large to match the data. We conclude that the observed patterns of excess return predictability are unlikely to be explained purely by time-varying risk premiums generated by highly risk averse agents in a complete markets economy. T h e Implications of First-Order Risk Aversion for Asset M a rket Risk P r e m i u m s 1. Introduction It is generally accepted that excess returns on a variety of assets are predictable. This is true for returns in the equity markets, bond markets, and foreign exchange markets of various countries. One interpretation of this evidence is that equilibrium risk premiums on all assets are highly variable. Yet, existing rational expectations general equilibrium models based on traditional expected utility preferences have been unable to generate risk premiums that are sufficiently variable to be consistent with the observed predictability of returns. Consequently, in this paper we abandon the expected-utility hypothesis in favor of preferences that exhibit first-order risk aversion.1 With these preferences, agents are substantively averse to even small gambles. Hence, a small degree of uncertainty in the exogenous environment of economic agents can potentially induce relatively large fluctuations in agents’ intertemporal marginal rates of substitution. This, in turn, implies large fluctuations in expected rates of return on a variety of assets. Our goal is to determine whether a general equilibrium model incorporating preferences that exhibit first-order risk aversion is quantitatively consistent with the predictability of returns and with other time series properties observed in the data from the foreign exchange market, the equity markets, and the bond markets of the U.S. and Japan. Other papers that propose first-order risk aversion as an explanation for asset pricing anomalies include Epstein and Zin (1990, 1991) and Bonomo and Garcia (1993). In particular, Epstein and Zin (1991) are unable to reject the overidentifying restrictions implied by a closed economy model, analogous to the model of Hansen and Singleton (1982), when first-order risk aversion is assumed. Their approach 1 The concept o f first-order risk aversion was introduced by Segal and Spivak (1990). 1 requires the researcher to choose a proxy for the unobservable rate of return on aggregate wealth, and their inference about the validity of the model depends on this choice. In an open economy setting, the choice of a proxy for the return on aggregate wealth is problematic. Hence, we do not follow the approach of Epstein and Zin (1991). Instead of testing the first-order conditions of the model, we explicitly solve a two-country monetary model for the endogenous moments of interest. In our model, consumption of two goods equals the outputs of two countries, which are assumed to be exogenous endowment streams. The two money supplies are also exogenous. The growth rates of these exogenous processes follow a discrete Markov chain that is estimated from U.S and Japanese data using the method of Tauchen and Hussey (1991). The equilibrium processes for returns and other endogenous variables are found by numerically solving a system of Euler equations. Having solved the model, we generate a variety of statistics that provide evidence on the predictability of the model’s returns. The performance of the model is evaluated by comparing these statistics to the corresponding statistics in the data. The remainder of the paper is organized as follows. In section 2, we present evidence on the predictability of excess rates of return in the dollar-yen foreign exchange market, in the dollar and yen discount bond markets, and in the equity markets. These stylized facts provide the set of statistics that we would like the model to match. In section 3, we discuss intuitively why time-varying equilibrium risk premiums could be an explanation of the statistics described in section 2. Section 4 introduces the concept of first-order risk aversion, and section 5 incorporates these preferences into a formal dynamic model. In section 6 we derive the model’s equilibrium conditions for endogenous financial variables. Section 7 describes our procedure for calibrating the model, and section 8 presents our results. Section 9 compares our results with Epstein and Zin (1991), and section 10 provides concluding comments. 2. Some Stylized Facts on Excess Return Predictability In this section we document the predictability of excess rates of return on discount bonds, 2 equities, and foreign money markets using regression analysis. Since U.S. and Japanese data are the exogenous processes of the model, we report results only for these two countries. Nevertheless, the evidence is consistent across the markets of most developed countries as documented by the recent empirical studies of Harvey (1991), Bekaert and Hodrick (1992, 1993), and Solnik (1993), among others. We begin by developing notation for own-country money market and equity investments. Let If (If) be the dollar (yen) return at time t+1 from investing one dollar (yen) at time t in the nominally risk-free dollar (yen) bond. Let Rf+1 (Rf+1) be the analogous one-period dollar (yen) return in the equity market of the U.S. (Japan). Throughout the paper, lower case letters will represent either the natural logarithms of upper case counterparts or variables measured as continuously compounded rates of return. Hence, the continuously compounded excess dollar (yen) rate of return in the U.S. (Japanese) equity market is rf+1 - if (rf+1 - if). The conditional expectation of an excess return is often referred to in the literature as a risk premium, and we will use this terminology interchangeably with expected excess return.2 A. The Foreign Exchange Market If S, is the spot exchange rate at date t of dollars per yen, the dollar return to investing one dollar in the yen money market and bearing the foreign exchange risk is St+1If/St. Let Ft be the forward exchange rate of dollars per yen quoted at date t for date t+1 transactions. Then, the dollar return to investing one dollar in the yen money market and eliminating the foreign exchange risk with a forward contract is F,If/S,. Since this investment also provides a risk-free return of dollars, covered interest arbitrage implies interest rate parity: 2 An excess rate of return is the nominal rate of return on an asset in excess of the short-term interest rate. If inflation is stochastic, conditional expectations of excess rates of return will be non-zero even if agents are risk neutral, which makes use of the term "risk premium" for these conditional expectations somewhat imprecise. Engel (1992) provides a recent discussion of this issue for the risk premium in the foreign exchange market. 3 (1) The right-hand side of equation (1) is the continuously compounded forward premium or discount in the foreign exchange market. Notice that the continuously compounded excess dollar rate of return from investing in the Japanese money market is if + s,+1 - st - if, which from equation (1) is st+1 - ft. A common 'way of testing the predictability of the excess rate of return on an uncovered foreign money market investment is to regress the excess return on the forward premium: s«m - ft = & + 0,(f, " st) + W (2) The null hypothesis that the excess rate of return is unpredictable implies 0, = 0. If the point estimate of /S, differs significantly from zero, there is evidence that ex ante excess rates of return vary over time. In the empirical analysis we focus on a quarterly holding period since that is the frequency we use for the exogenous processes in simulating the model.3 The sample period for our exchange rate data is January 1976 to December 1989. The data are described more completely in Appendix A. The first row of Table 1, Panel A, displays the regression results for equation (2) using the threemonth forward premium as the predictor. As is typical in the literature, the slope coefficient of -4.016 is significantly negative.4 The R2 for the regression is .22, and the standard deviation of the fitted value of the excess return, reported in Table 1, Panel B, is 12.36%.5 These statistics indicate that excess 3 The availability of monthly observations on the three-month holding period allows us to use additional observations that increase the power of the tests but induce autocorrelation in the errors. As in Hansen and Hodrick (1980), the standard errors in Table 1 allow for the autocorrelation induced by the overlapping error structure and, additionally, for possible conditional heteroskedasticity as in Hansen (1982). 4 For the dollar values of other major foreign currencies, the estimated coefficients are also significantly below zero. Similar results arise in regressions using non-dollar exchange rates as demonstrated in Bekaert (1992) and Hodrick (1992). Bossaerts and Hillion (1991) use French franc exchange rates and find slope coefficients that are all less than zero, but not all are significantly negative. 5 All rates of return in this paper are expressed in percentage points per annum. 4 returns are quite predictable and risk premiums are quite variable. B. The Discount Bond Market Similar evidence of predictable excess holding period rates of return can be found in the discount bond market using a forward premium computed from bond prices as a predictor. Specifically, let Vkk denote the date t price of a nominally risk-free pure discount bond which pays one unit of currency at * date t+k. When necessary to avoid confusion, there will be a superscript $ or ¥ symbol on to denote the currency. Let i^k be the continuously compounded yield to maturity on a k-period bond expressed in percent per period. By definition, v,k = exp(-ki^k). (3) Let the one-period continuously compounded holding period rate of return on a k-period bond realized at time t+1 be h,+l k = ln(Vt+I k.,/Vuk), which from equation (3) can be written as K im = " (k - + ki..k- (4) In the empirical analysis we examine the one-period excess holding period rate of return on a twoperiod bond, h,+12 - i„ in a regression analogous to equation (2). For parallel structure, we define the forward premium in the bond market, denoted fb„ as the logarithm of the contractual price today for a one-period bond delivered one period from now minus the logarithm of the price today of a one-period bond: fbt = ln(Vt2/Vu ) -ln(Vu) = - 2 ^ + 2it. (5) The bond market analogue to equation (2) is hf u " i> = 00 + 0i(A\) + W <6) If excess holding period returns are unpredictable, /3, should be zero. In rows two and three of Table I, we report estimates of equation (6) for the U.S. dollar and Japanese yen discount bond markets. Since our timing interval is one quarter, ht+1>2 is the three-month 5 return on a six-month bond and fbt is the forward premium on a three-month bond to be purchased three months in the future. For the empirical analysis we have monthly observations on three-month and sixmonth Euro-dollar and Euro-yen interest rates from October 1975 to June 1990. For both the dollar and the yen markets, the estimate of /S, is -0.45, and both are significantly negative.6 While the estimated /5,’s are not as negative as the estimates from the foreign exchange market, there is strong evidence of predictability of the excess rates of return. The R2 for the U.S. market is .03, and the R2 for the yen market is .09. The standard deviations of the fitted values of the excess returns in the two markets are 0.318% for the U.S. and 0.370% for Japan. These statistics are much smaller than those in the foreign exchange market. C. The Equity Markets A similar set of results emerges from examining excess rates of return in equity markets. We focus on results in Bekaert and Hodrick (1992), who show that excess rates of return to U.S. and foreign equities are predicted by the forward premium in the foreign exchange market. Consistent with our twocountry framework, we construct a dollar world equity market excess rate of return as an equallyweighted average of the dollar excess rates of returns on the equity markets of the U.S. and Japan: r«!i - i* = [(r*, - it) * (r*. - i*) + (st<1- f,)](l/2). <7) We regress this excess return on the three-month forward premium in the dollar-yen foreign exchange market: r,:, - i* = P0 + 0,(f. - s.) + «, (8) The fourth row of Table 1, Panel A, reports a slope coefficient of -3.543, with a standard error of 0.816. As equation (7) indicates, there are three components to this world equity excess rate of return: the excess 6 These results are similar to those reported by Fama (1984b) and Stambaugh (1988) for monthly U.S. data. 6 dollar rate of return in the U.S. equity market, the excess yen rate of return in the Japanese equity market, and the excess rate of return in the foreign exchange market. The regression of the third component on the forward premium is discussed above. Regressions of the first two components on the forward premium are contained in rows five and six of Table 1, Panel A. Each of the components has a negative slope coefficient, and all but the Japanese equity coefficient are significantly negative. 3. Time-Varying Risk Premiums The patterns of predictability in excess returns documented above can, in principle, be explained by time variation in equilibrium risk premiums.7 Our paper considers this point of view. In this section we formally relate the regression evidence of the previous section to time-variation in risk premiums using a decomposition of the forward premium introduced by Fama (1984a). We then discuss what is required from an equilibrium model if it is to be consistent with the patterns observed in the data.8 Define the logarithmic risk premium in the foreign exchange market as rp, = Et(st+1) - ft. Following Fama (1984a), the forward premium can be decomposed into the expected rate of depreciation of the dollar relative to the yen minus this risk premium: fpt s ft - s, = Et(Ast4l) - rp„ (9) where A is the first difference operator. 7The literature modelling asset returns as the outcome of a dynamic, stochastic equilibrium is too vast to be reviewed here. Examples of recent papers that model excess returns in foreign exchange markets using approaches related to that used in this paper include Backus, Gregory, and Telmer (1992), Bansal, Gallant, Hussey and Tauchen (1991), and Bekaert (1993a,b). Kandel and Stambaugh (1991) use a similar approach to model excess returns in equity markets. 8 There are at least two other potential explanations of the data, which we take seriously but do not pursue in this paper. First, the stylized facts may not be representative of the true population distributions. Rather, they may be examples of problems in statistical inference caused by the adoption of a rational expectations perspective in non-experimental data. Such problems could be caused by infrequent regime changes, learning, peso problems, data snooping biases, and so forth. Second, these stylized facts may be evidence of market inefficiency. 7 Now, consider the regression of st+1 - ft on a constant and fp„ as in equation (2). By rational expectations, st+1 - ft = rpt 4- et+I, where the innovation, et+1, is orthogonal to the time t information set. Consequently, only the risk premium covaries with the regressor in equation (2), and the slope coefficient in that regression is = :cov(st4t - f t, fpt) = cov(rpt, Et(AstJ ) - var(rpt) 1 var(fpt) var(Et(Ast4l)) + var(rpt) - 2cov(rp„ Et(Ast4l))' (1Q) A decomposition similar to equation (9) can be performed for the bond market. Define the risk premium in the bond market to be rb, = E,(h,+l2 - ij. Then, the forward premium in the bond market, fbt, can be decomposed into the expected rate of change of the logarithm of the one-period bond price minus the risk premium: fbt = Et(Avu l) - rb,, (11) where vl+1 = ln(Vt+t). The slope coefficient in equation (6) would then be given by an expression analogous to equation (10) with rp, replaced by rbt and Ast+1 replaced by Av,+1. Clearly, if var(rp,) = 0, the slope coefficient in equation (2) is zero. Similarly, if var(rb,) = 0, the slope coefficient in equation (6) is zero. To generate the negative slope coefficients found in the data, the risk premiums rp, and rb, must vary through time. The decomposition (10) provides intuition regarding the amount of time-variation in risk premiums required to match the data. Consider first the case of foreign exchange. The explained variance in regression (2) is /S?var(fpl), so the finding of /3, < 1 implies var(rp,) > var(fpt). (12) Furthermore, if /3X< -1, equation (10) implies var(rpt) > cov(rpt, Et(As,4l)) > var(Et(Astn)). (13) Hence, regression results like those in the actual data require that the risk premium in the foreign 8 exchange market be more variable than the expected rate of depreciation and that the risk premium and the expected rate of depreciation covary positively. Similarly, since the slope coefficient for the equity market regression is comparable to that in the foreign exchange market regression, the implied variability of the equity risk premium is comparable to the implied variability of the risk premium in the foreign exchange market. * In the case of the bond market regressions (6), the estimated slope coefficients are insignificantly different from -0.5. Equation (10), applied to the dollar and yen bond markets, then implies var(Et(Avt<1)) ~ var(rbt) (14) That is, the variabilities of the risk premiums in the two bond markets are roughly equal to the variabilities of the expected rates of change of the logarithmic bond prices. The general asset pricing framework of Hansen and Richard (1987) provides insight into what is needed from an equilibrium model if it is to generate the requisite time-variation in risk premiums. Let Rj>l+, denote an arbitrary dollar return realized in period t + 1 from investing one dollar at time t. Hansen and Richard (1987) show that there exists a stochastic discount factor, Q*+1, satisfying9: Et[Qt»iRu*,] = 1- (15) In equilibrium models with effectively complete markets, Q*+1 equals each agent’s intertemporal marginal rate of substitution of wealth divided by the gross rate of change in the dollar price level. Equation (15) implies that substantial time variability in excess returns can be achieved only if there is substantial time-variation in the conditional second moments of the joint {Rft+1, Q?+1} process, 9 Hansen and Richard (1987) require that (i) the space of portfolio payoffs is a Hilbert space of square integrable random variables, and (ii) there are no arbitrage opportunities. 9 which, in turn, requires substantial volatility in the marginal rate of substitution.10 One way of generating highly volatile marginal rates of substitution is to assume that agents have a high degree of risk aversion. In effect, the extreme nonlinearity associated with high risk aversion can transform the uncertainty due to conditionally homoskedastic exogenous inputs into endogenous risky asset returns whose moments are conditionally quite variable. Alternatively, high volatility in marginal rates of substitution can be generated by directly assuming time-varying conditional heteroskedasticity in the exogenous driving processes (as in Bekaert (1993b)).11 It should be noted that increasing the variance of risk premiums is not sufficient to insure that conditions (12), (13), and (14) hold. The forward premiums and the expected rates of change of asset prices are also endogenous variables in the model. Changes in the model specification that increase the variances of risk premiums may also increase the variances of E,(Ast+1) and Et[Avt+1], leaving the relative variances unchanged. Furthermore, the covariances between the risk premiums and the expected rates of change of asset prices may change. Thus, while it is likely that extreme risk aversion will increase the variability in intertemporal marginal rates of substitution, it is unclear whether this is sufficient to induce the patterns of predictability in excess returns that are observed in the data. Hence, to explore the effects of increasing risk aversion we must solve the model explicitly. 4. First-Order Risk Aversion The preceding discussion makes clear the role that substantial risk aversion may play in generating the regression results described in section 2. Models using expected-utility preferences have 10 Hansen and Jagannathan (1991) develop restrictions on the mean and variance of the intertemporal marginal rate of substitution implied by the stochastic properties of observed asset returns. Cochrane and Hansen (1992) use this methodology to survey equilibrium approaches to asset pricing. They demonstrate how difficult it is to generate sufficient variability in Qf+, within the context of an equilibrium model. 11 In principle, these two approaches could be combined. For the model of this paper, however, incorporating time-varying conditional heteroskedasticity substantially increases the dimensionality of the state space, rendering the approach computationally intractable. 10 not fared well in this regard. Even the models of Backus, Gregory, and Telmer (1993) and Bekaert (1993b), which incorporate substantial time-nonseparabilities in the form of habit persistence, fail to imply sufficient predictability in excess rates of return. The reason for this failure is not difficult to see. Expected-utility preferences display second-order risk aversion. That is, in response to a lottery which is close to perfect certainty, an expected utility maximizer exhibits behavior close to risk neutrality. This is a problem for consumption-based asset pricing models because, at any given date, the conditional variance of next period’s aggregate consumption is small. One way of addressing this problem is to abandon preferences that display only second-order risk aversion in favor of preferences that imply first-order risk aversion. With this type of preference specification, agents are substantially averse to even small gambles. Epstein and Zin (1991) examine a variety of such preferences, including Gul’s (1991) disappointment aversion preferences, in order to increase the variability of agents intertemporal marginal rate of substitution. Disappointment aversion was introduced by Gul (1991) as a way of accommodating the Allais paradox within a parsimonious extension of expected utility. Camerer’s (1989) review of the experimental economics literature suggests that expected utility cannot explain the experimental evidence on preference orderings under uncertainty. Rather, what is required is a preference ordering in which outcomes are evaluated relative to some reference point. Disappointment aversion has this property. We follow Epstein and Zin (1991) in using the following simple model of disappointment aversion. A preference ordering over the space of probability distributions T (e.g., over alternative lotteries) can be represented by a certainty equivalent function /*: 7*-*R. For P € T, fi(P) is implicitly defined by A ^ 1, where K = A-prob(z > y) + prob(z < fi). a < 1. (16) If A = 1, the preferences described by equation (16) 11 correspond to expected utility with a coefficient of relative risk aversion equal to 1 - a . If A differs from unity, equation (16) can be interpreted as follows. Those outcomes below the certainty equivalent are disappointing, while those above the certainty equivalent are elating. If A < 1, the elation region is down-weighted relative to the disappointment region. The next section embeds these preferences in a two-country monetary model. 5. A Two-Country Monetary Model In the model asset prices and exchange rates are determined in a competitive equilibrium in which the demands for assets and goods are the optimal choices of a representative agent. As in Marshall (1992) and Bekaert (1993b), money is demanded by agents because consumption transactions are costly, and increasing real balance holdings decreases these transaction costs. Specifically, let the two countries be denoted as x and y, respectively. The representative agent’s consumption of the good produced in country x is c \ and the representative agent’s consumption of the good produced in country y is cy. Consumption of cx involves transaction costs measured by (17) denominated in units of c\ where Mf+, is the amount of currency x (which we call the dollar) acquired by the representative agent in period t; and Px is the price of cx at date t in units of M\ Consumption of cy involves a transaction cost of (18) denominated in units of good y, where Mf+, is the amount of currency y (which we call the yen) acquired by the representative agent in period t; and P( is the yen price of cy at date t. The timing in this model differs from the transaction-cost models of Feenstra (1986) and Marshall (1992) in that money provides transaction services in period t when it is acquired. However, money must 12 be held until the following period, so losses in purchasing power due to inflation accrue in period t + 1. This timing is imposed for tractability. With our timing, the only endogenous state variable affecting an individual agent’s decisions is the agent’s stock of wealth. If money only provided transaction services if acquired one period earlier, the agent’s stock of money would represent a second endogenous state variable. The optimality conditions would then involve the derivatives of the (unknown) value function with respect to the money-wealth ratio. To solve such a model numerically would be extremely burdensome computationally. In addition to monies, agents can hold n capital assets. Let Zj t+1 be the value (in units of cx) of the representative agent’s investment in asset i, chosen at t, and which pays off at t + 1. The gross real return to asset i (measured in units of good x received in t + 1 per unit of good x invested at date t) is denoted Ri>l+1. If St denotes the exchange rate (dollars/yen), the budget constraint for the representative agent in units of consumption good x is c,* + ♦ ^ i ( c ly + tf) + £ z.i,t*l P.x ' M P « ' m ,: < W (19) P,x where W, denotes the representative agent’s wealth at the beginning of period t: Mt* + S, M,y " W. - - l- - ‘ ■ * £ z, A , (20) The representative agent’s preferences over current and uncertain future consumption incorporate disappointment aversion as in equation (16), and are specified using the approach of Epstein and Zin (1989). Specifically, let Jt denote the vector of exogenous state variables which span the agent’s information set at date t. The utility value of Wt in the state Jt is denoted V(Wt, JJ, and is defined recursively by 13 i/p V(Wt, Jt) = max {c,*, Cyt M,:„ M*,, r ( k i v r r * ^ [ P v W „ u u,])' (21) 0 < 6 < 1, p < 1, subject to the budget constraint (19) and the wealth constraint (20), using the definition of p from equation (16). The expression p|PV(W s )|Jtj denotes the certainty equivalent of the conditional % distribution of the value function at date t+1, given information at date t. When agents make their consumption and portfolio choices, they care about two distinct effects: how their choices affect currentperiod utility, and what happens to the probability distribution of their future utility. In an expected utility framework, the latter effect is incorporated by taking the conditional expectation of next-period’s value function. In equation (21), effects of the probability distribution of future utility on current utility are captured by the certainty equivalent function p. In addition, the two effects are aggregated in equation (21) by the CES function of the form (a" + bp)tl/p), while, in the expected-utility framework, the two effects are simply added. The parameter p governs intertemporal substitution in the following, somewhat unconventional, sense: The elasticity of substitution between current utility (c’t)i(cy)1', and the certainty-equivalent of future utility, p{PV(W , , | JtJ, is given by 1/(1+p). Therefore, p determines the optimal tradeoff between present and future utility. When p is near unity, there is an extremely high degree of substitutability between these two sources of utility. Extremely negative values of p imply almost no substitutability. It should be noted that this elasticity of substitution does not directly correspond to the elasticity of substitution between current and future consumption (as studied, for example, by Hall (1988)). The more conventional notion of intertemporal substitution elasticity is a function of all the preference parameters of the model. 14 6. Equilibrium Determination of Exchange Rates and Asset Returns In order to derive equilibrium asset prices and exchange rates, we must solve the representative agent’s decision problem and impose market clearing. The agent’s optimal behavior is characterized by a set of Euler equations that involve the real return on optimally-invested aggregate wealth, which we denote R,. (An explicit characterization of Rt can be found in Appendix B.) These equations also involve the real returns, inclusive of marginal transaction cost savings, from holding dollars and yen. Let ' p,x p,:. denote the real return from holding dollars, where i i +& (22) denotes the period t partial derivative of \p* with respect to its i* argument. (\^, is defined similarly.) The real return from holding yen is Ry .t* l X . p .x s,p,:, i (23) i + ^ Note that both Rxt+, and Ry,t+i are measured in units of good x received in t+1 per unit of good x invested at date t. The first-order conditions for the representative agent’s optimal consumption, money holdings, and portfolio choices are the following:12 12 request. The derivation is a modification of the arguments in Epstein and Zin (1989), and is available upon 15 (2 4 ) Et{lA(ZtM)[ZrM - 1]} = 0, E{lA(Zt4l)Z “1Rt;11Rit4l] = Et[IA(Zt4l)], V i = x , y , l ........N, (25) where 0 1/p y P<l-i) X pa-i c,:i C..I ’1 + (26) ' *.♦« C,x c,y 1 + flu. and A if Z ^ 1 *a(Z) - (27) 1 if Z < 1 In developing the solution to the model, it is useful to define the endogenous processes for consumption-velocities of the two monies. Let vj and vj denote the consumption-velocities in countries x and y: v, ■ c‘P‘ cyPy M,t*l M,y t*l ' ‘ y• - v v — = vty 1 • (28) Other endogenous variables are the nominally risk free, continuously compounded interest rates on dollars and yen, denoted i* and i*. Nominally risk-free interest rates are functions of the marginal transaction costs with respect to real balances: its = In 1 ); *i.rt = In 111 1 +^2, The exchange rate S, is given by 16 1 (29) ptx 1 + tit V ' pty 1 + tit c,y 1 -6 6 (30) Given interest rates and the spot rate, the forward rate F, can then be computed using covered interest parity as in equation (1). 7. Calibration and Solution o f the Model In this section we describe our procedures for choosing the parameters of the stochastic processes for the exogenous variables and of the transaction cost functions. The outputs and money supplies of the two countries are assumed to be exogenous. There are no data series corresponding precisely to the endowment constructs of the model. The difficulty is that a two-country mode) cannot replicate the complexity of trade patterns that we observe in the real world. As a result, we do not take seriously the predictions of the model for quantity variables. However, the implications of the model for asset return predictability can be investigated with a plausible specification of the endowment driving processes. We calibrate the endowments and money supplies of the two countries to consumption data and money supply data from the U.S. and Japan. The growth rates of these four exogenous processes are assumed to follow a vector autoregression, which we will approximate as a discrete Markov chain. We find that a first-order VAR with conditionally homoskedastic errors fits the data well. In Table 2, Panels A and B, we display OLS estimates of this VAR. Table 2, Panel C reports statistics testing the appropriate lag length for the VAR. The Akaike and Schwarz criteria, as well as the sequential likelihood ratio tests, support the first-order VAR specification. Table 2, Panel D provides statistics testing for normality, autocorrelation, and conditional homoskedasticity of the VAR residuals. Only in the residual for the growth rate of Japanese consumption is there marginal evidence of serial correlation. For none of the residuals is there significant evidence against normality or conditional homoskedasticity. The growth rates of these four exogenous processes are approximated by a discrete first-order 17 Markov chain in which each variable can take four possible values, implying a state space with 256 possible values. The Markov chain is calibrated to the estimated VAR using the Gaussian quadrature method of Tauchen and Hussey (1991). In Table 3, we display the parameters of the first-order VAR implied by this Markov chain approximation. The parameters characterizing the Markov process VAR are virtually indistinguishable from those of the estimated VAR reported in Table 2. All parameters of the Markov process VAR (including the elements of the covariance matrix decomposition) are within onetenth of one standard error of the corresponding parameters in the estimated VAR. We take this as evidence that the discrete approximation is unlikely to distort the economic implications of the model. The parameters of the transaction cost functions (17) and (18) are chosen by fitting equations (29) to U.S. (for \p*) and Japanese (for V'5') data, as described in Appendix A. Specifically, we set ^ x(c,m) = 0.0008c4-35,m ''° 31; ^(c,m) = 0.0166c2“ ’’m*-2109. (31) Given this exogenous process, the three unknown endogenous processes R,, v}, and v{ are found by solving the three Euler equations (24) and (25) (for i = x and y) simultaneously. Since the state space is discrete, the Euler equations can be solved exactly for the 256 values of each endogenous variable. The only approximation is in the initial discretization of the driving processes. A detailed description of the solution procedure can be found in Appendix B. Once R,, vj, and v( have been determined, all other endogenous variables can be calculated from definitions and equilibrium conditions. 8. Implications of the Model for Excess Return Predictability In this section, we report results obtained from solving the model for a variety of parameters governing preferences. The quarterly subjective discount parameter /3 is fixed at (0.96)025. The choice of 6 (the weight on cx in the current-period utility) is irrelevant, since we examine rates of depreciation , rather than levels of exchange rates. The remaining parameters are varied over the following grid: A € {1.0, 0.85, 0.70, 0.55, 0.40, 0.25}, p € {0.50, -0.33, -4.0, -9.0}. We experimented initially with 18 values of a between 0.5 and -9 and found that the choice of a had virtually no effect on the moments of interest. Consequently, we only report results for a = -1. This corresponds to a coefficient of relative risk aversion of 2 in an economy with expected-utility preferences over timeless gambles. We first discuss the ability of the model to replicate the predictability of excess returns documented in section 2. We focus on three measures of predictability: the slope coefficient in the excess return regressions analogous to equations (2), (6), and (8); the R2, measured as the ratio of the variance of the expected excess return to the variance of the realized excess return; and the standard deviation of the expected excess return. All three statistics can be computed exactly given the discrete Markov chain driving process. Consider the model’s implications for the slope coefficients in the excess return regressions analogous to equations (2), (6), and (8). The results are displayed in Tables 4, 5, and 6 for the foreign exchange market and the dollar and yen discount bond markets, respectively. Table 7 displays the slope coefficient when the excess return to the aggregate wealth portfolio (which we interpret as an analogue to an unlevered equity portfolio) is regressed on the foreign exchange forward premium. It is clear from these tables that the model cannot match the slope coefficients estimated from observed data. For no combination of parameters do the regression coefficients implied by the model come close to the magnitudes reported in Table 1. For example, for the foreign exchange market regression, the estimated slope coefficient in Table 1 Panel A is -4.016, with an estimated standard error of 0.766. The most negative slope coefficient implied by the model is -0.191, which is approximately five standard errors away from the estimated value. Similarly, the slope coefficients implied by the model for the term structure regressions analogous to equation (6) (reported in Tables 5 and 6) and the equity return regressions (reported in Table 7) are extremely small, and they are all more than 3.4 standard errors away from the corresponding estimates reported in Table 1. The second measure of predictability is the model’s R2 as defined above. This theoretical R2 19 cannot be observed in the data, but a lower bound is provided by the estimated R2s reported in Table 1, Panel A. Whereas the R2s in the data are substantive, ranging between 1% and 22%, the corresponding R2s in the model are negligible, all being less than .2%. The third measure of the predictability of excess returns is the variability of the explained component of excess returns. As with the R2discussed above, a lower bound for this measure in the data is provided by the standard deviation of the fitted value of the excess return regressions reported in Table 1, Panel B. As with the previous two measures, the model is unable to reproduce the variability observed in the data. For example, the standard deviation of the fitted value of St+, - f, in Table 1 is 12.4%. The largest value of the standard deviation of E,(s,+l - f,) from the model, reported in Table 4, is 0.356%, which is over thirty times too small. Analogously, the standard deviation of the fitted value of the excess world equity return in Table 1 is 10.9%. The largest standard deviation of Et(rt+1 - ij from the model, reported in Table 7, is 0.175 %, which is over sixty times too small. The standard deviations of the fitted values of the excess returns in the discount bond markets are 0.318% and 0.370% for the dollar and the yen markets, respectively. The maximum value of the standard deviations of the expected excess returns, reported in Tables 5 and 6, are 0.13% and 0.06% respectively. These results are somewhat disappointing to those who favor risk-based explanations for the predictability of excess returns. To further explore the role of risk aversion in generating predictability in excess returns, we next examine how the predictions of the model change as we increase the importance of the first-order risk aversion by lowering A. In all cases, setting A = 1 results in extremely small values for the slope coefficients. However, it is not generally true that increasing the amount of risk aversion (decreasing A) implies more negative slope coefficients. Furthermore, a large degree of risk aversion is not systematically associated with a particular sign of the regression coefficient. For example, the coefficients corresponding to A = .40 and A =.25 in Tables 4 through 7 are as likely to be positive as to be negative. Thus, even if it were assumed that agents in the economy display extreme 20 risk aversion, it is not at all clear whether this would improve the performance of the model along this dimension. To see why the model fails to replicate the observed slope coefficients, it is useful to return to the discussion of section 3. In that section, we argued that substantial time-variation in risk premiums is necessary if a model is to match the patterns found in the data. Examination of Tables 4 through 7 reveals that the variances of the ex ante risk premiums are unambiguously increasing as the degree of first-order risk aversion increases. For foreign exchange, the standard deviation of the risk premium increases by a factor of 100 when A moves from 1 to .25. For discount bonds and the aggregate wealth portfolio, the standard deviation of the risk premium increases at least twenty-fold when A moves from 1 to .25. Similarly, the R2’s in all markets increase dramatically as first-order risk aversion is increased. The reason why these dramatic increases in risk-premium volatility do not imply comparable increases in the magnitude of the slope coefficients in the prediction regressions is that these coefficients are functions of moments in addition to the variances of the risk premiums. As shown in equation (10), the slope coefficients also depend on the variances of the expected asset price changes and on the covariances between the expected changes in asset prices and the risk premiums. These moments are also affected by changes in the parameter governing first-order risk aversion. In particular, Tables 5 and 6 show that the variances of the expected changes in the prices of one-period discount bonds actually decrease unambiguously as A decreases. The variance of the expected change in the spot foreign exchange rate is not monotonic in A. As shown in Table 4, decreasing A from unity initially reduces this variance, while further reductions in A increase it. The value of A at which this variance is at a minimum depends on p. The effects of increased first-order risk aversion on the covariances between the ex ante risk premiums and the expected changes in asset prices are also in Tables 4 through 6. In the foreign exchange market, decreasing A unambiguously increases this covariance. In the discount bond markets, the response of this covariance to increased risk aversion is not monotonic, and depends 21 on the value of p. Our model also has implications for the unconditional mean equity premium and the unconditional standard deviations of financial variables, which provide additional dimensions to assess the model’s performance. In Table 7, increasing the amount of first-order risk aversion dramatically increases the unconditional mean excess equity return. As A is reduced from 1 to .25, the mean equity risk premium increases by a factor of approximately 20. This increase is not sufficient to match the data as the largest mean equity premium generated by our model simulations is 3.5%. While this is substantially below the value of 8.4% estimated from our data set, the equity return data correspond to a levered portfolio, while the equity return computed in our model is unlevered. The result are comparable to those of Bonomo and Garcia (1993) for homoskedastic driving processes. These authors are able to increase the mean equity risk premium significantly by employing a richer driving process that incorporates regime switching. Table 8 displays standard deviations implied by the model. In comparing Table 8 with Table 1, Panel B, notice that the magnitudes of the standard deviations in the model are almost always smaller than the corresponding statistics in the data. In particular, the standard deviation of currency depreciation is approximately 2.5 times higher in the data than in the model, and the standard deviation of the equity risk premium is approximately three times higher in the data than in the model. When p = -9, the standard deviation of the forward premium in the model is only 50% lower than that in the data; for the other values of p, the variability of the forward premium is almost an order of magnitude too low. Although the model underpredicts the variability of both expected and realized excess returns, the parameterizations of the model that generate the largest variances of expected rates of return tend to overpredict the variances of the forward premiums in the discount bond markets. For example, with p = -9 and A = .25, the standard deviations of the forward premiums in the dollar and yen discount bond market are 3.81% and 2.56%, compared to 0.71% and 0.83% in Table 1, Panel B. 22 The source of this problem is as follows. In order to generate high volatility in excess returns, the model must generate high volatility in the conditional second moments of intertemporal marginal rates of substitution. Unfortunately, parameterizations of the model which do this also imply highly volatile spot interest rates. A similar problem has been noted in a closed-economy model by Heaton (1991). Consequently, one challenge for this class of models is to accommodate highly variable expected and realized excess returns on risky assets while keeping short-term interest rates relatively non-volatile. 9. On the Success of Epstein and Zin (1991) Epstein and Zin (1991) are unable to reject the overidentifying restrictions implied by their single country model with preferences incorporating first-order risk aversion, which suggests considerable support for this approach to asset pricing. Our approach is less successful. How can we explain the differences in findings? According to the Euler equation (25), the implications of these models for asset returns are summarized in the behavior of the asset pricing operator Et[IA(Zt-,)] -1 ♦1• This operator is a function of Rt+1, the return to the aggregate wealth portfolio. Euler equation estimation requires an observable analogue to this asset pricing operator, and Epstein and Zin use the return on a value-weighted portfolio of equities as their empirical measure of Rt+1. This procedure is clearly subject to Roll’s (1977) critique, a point acknowledged by Epstein and Zin. Furthermore, with this approach, the empirical asset pricing operator is a function of the returns on the equity assets being priced. The operator partially inherits the statistical properties of observed equity returns, so it has less difficulty replicating the behavior of observed excess equity returns. In contrast, we derive Rt+1 by explicitly solving the model’s equilibrium as a function of the growth rates of output and money in the two 23 countries. Nowhere do we use data on asset returns in deriving the asset pricing operator. To ask the pricing operator, derived in this way, to replicate the stochastic properties of equity returns is a much tougher test of the model than the Epstein-Zin procedure. It is not surprising that we find more evidence against the model. 10. Conclusions In this paper, we ask whether high levels of risk aversion can explain the observed predictability of excess returns within the context of a frictionless, representative agent model. In order to give this explanation the best chance for success, we assume that agents’ preferences display first-order risk aversion. This preference specification implies that agents respond more strongly to consumption risk than would be the case under conventional Von Neumann-Morgenstern preferences. Yet, even this more extreme form of risk aversion can explain only a small fraction of the predictability of excess returns found in the data. Furthermore, we find that the slope coefficients in equations predicting excess returns do not increase monotonically with increased risk aversion. The level of risk aversion affects not only the variability of risk premiums, but also the second moments of other endogenous variables which affect predictability. The resulting implications for the signs and magnitudes of these slope coefficients are ambiguous. Taken together, the results of this paper suggest that the predictability of excess returns cannot be fully explained simply by modifying preference assumptions. A more promising approach may be to abandon the assumption that the empirical distribution in the data set is a good proxy for agents’ subjective distribution over future variables. Rational optimizing models that do not impose this assumption include learning models, models with peso-problems, and some models with regime switching. It is hoped that these alternative approaches will have more success in explaining excessreturn predictability than approaches based solely on modelling agents’ aversion to consumption risk. 24 APPENDIX A: DESCRIPTION OF DATA The interest rate data are monthly series on three-month and six-month Euro-dollar and Euro-yen rates, obtained from the Harris Bank database at the University of Chicago. Monthly exchange rate data are from daily bid and ask rates from Citicorp Database Services and are described in detail in Bekaert and Hodrick (1993): The money supplies for the U.S. and Japan are quarterly series taken from International Financial Statistics (IFS Series 34). Growth rates are deseasonalized by regressing on four dummies. The consumption data are Nondurables and Services from the OECD Quarterly National Accounts. The Japanese data include the Semi-durables category, as this category is included in the U.S. Nondurables series. Per capita data on money supplies and consumption were derived by using linear interpolations from the annual population series 99z from IFS. The transaction cost technology parameters are considered to be part of the exogenous environment and are calibrated from the model’s implications for money demand, as summarized by equations (29). These equations imply linear relationships between the logs of current dollar and yen velocities of circulation and the logs of the respective interest rate divided by one plus the interest rate. The calibration is done by linear regression using quarterly Eurocurrency interest data and nominal velocity. The velocity series used is computed using nominal GDP, taken from OECD Quarterly National Accounts, divided by the money series described above. GDP velocity is used because it implies more reasonable parameters for the transaction cost function than consumption velocity. The use of GDP velocity can be justified by noting that money in actual economies intermediates many more transactions than just consumption transactions. See Marshall (1992) for a fuller discussion of this issue. APPENDIX B: SOLUTION PROCEDURE The key step in solving the model is to solve numerically the Euler equations (24) and (25) for 25 the endogenous variables vj, v(, (defined in equations (28)) and R,+i (the return to the aggregate wealth portfolio). We do so by using a finite-state Markov chain to approximate the exogenous driving process (see Tauchen and Hussey (1991)). We then solve the model exactly for this approximate driving process. In this appendix we describe the solution procedure in greater detail. Let ej denote the total output of good x at date t, let denote the output o f good y at date t, and let Mt+i and M(+1 denote the supplies o f dollars and yen respectively, available for use in mediating transactions at date t. (These money stocks are dated t + 1 because it is assumed that the loss in value from inflation accrues to the agent in t+ 1 .) Let g, denote the vector o f growth rates o f outputs and money supplies in the two countries: S, It is assumed that {ej, e^, MJ+I, M^+,} is an exogenous process whose law o f motion is known. First, we show how equations (24) and (25) can be written in terms o f g, and the three endogenous processes {vf, v(, Rt+1}. Using (17), (18), and the requirement that, in equilibrium, the output o f each good must either be consumed or used as transaction costs, we can write consumption growths, marginal transaction costs, and inflation rates as functions o f {gt, vf, ^x ^X ct+l _ et+l _X X Ct i + (32) = e& ' i cty ety i + tfvtr,)£-‘ rn = (34) c-i (35) tf, = & { w ) & = X(1 -i/)(vt‘)' & = p,:. _ vt:, c,x h * m K ii p,y > \ p,y *. C,y m V,y C,y, (38) ,:, ,!2 (39) M tyi The next step is to formally characterize R,+„ the return to the aggregate wealth portfolio. Since we define the return to money inclusive of marginal transaction costs, $ jit and $ !>t, we must incorporate these marginal transaction costs into the definition of the portfolio weights for the aggregate wealth portfolio. Formally, let 27 w t ■ w t- S , I l (c,’ * t f ) Pt + --- V'zt p. s t— Pt (40) Wt denotes wealth available for asset purchases at date t, adjusted for marginal transaction costs. The portfolio weights on the aggregate wealth portfolio are defined in terms o f \Vt. Let wxt+1 and wyt+1 denote the portfolio' weight on MJ+1 and respectively: M,t*l (41) (i + &)) /W. K M y (42) W y.ul S P, Let wi t+1 denote the portfolio weight on asset i: w. = Z,MX (43) i = Note that the weights sum to unity: N w.i,t*l E i-1 + W (44) x,t*l + wy,t*l = 1A* The return to the aggregate wealth portfolio is defined as follows: R .-i s ’ i-1 (45) where RM+l and Ry,t+, are defined in equations (22) and (23). Aggregate wealth evolves according to w,., - m In a single-good nonmonetary model, the market return can be expressed as a function o f the wealth/consumption ratio and the growth rate o f consumption. It is convenient to express R,+1 in a similar 28 way. To do so, define c, = W, - W„ and let the "wealth/consumption ratio" Wt/ct be denoted wc,. Equation (46) then implies R- ' WC.-1 '' * ' The transaction cost functions \p* and ^ are homogeneous o f degree one, so one can use Euler’s theorem, along with equation (30), to show that ' c ' C« C,x ■ 1 (48) 1 + 'Put By using equations (32) - (39) and equation (48) in equations (22), (23), (26), and (47), one can write the endogenous processes Rx>t+1, Ry.,+i, Zt+1, and R,+l as functions o f (gt+1, vj, v?+1, v(, v(+1, wc„ wct+,}. It follows that the three-equation system consisting of equation (24) and equation (25) with i = x and i = y, can be expressed in terms of {gl+1, vf, vf+1, v(, v(+„ wct, wct+1}. Let this three-equation system be denoted Et[f vt*l»vt*>v£i,vty,wct4l,wcj = 0 (49) where f is a known function. Our task then is to find a stochastic process {vj, v(, wc,} which satisfies equations (49) for the given g, process. Following Tauchen and Hussey (1991), we approximate g, by a finite-state Markov chain. The discretization uses Gaussian quadrature. In the results presented in section 7, we allow each o f the four elements o f g, to take on 4 values, implying 256 possible states o f the economy. The endogenous processes, vj, v(, wc,, are then represented by vectors with 256 elements each, to be determined by solving system (49). The conditional expectation can be evaluated exactly (given the discrete approximation) since the state transition probabilities are known. We reduce the computational burden of this solution algorithm by assuming that the growth rates of c} and c( are observed, rather than 29 the growth rates of output in the two countries. This enables us to solve system (49) recursively: the elements of {wc„ vx} do not depend on the third equation in (49). Therefore, the 512 elements of {wc,, vx} are found by simultaneously solving the 512 equations represented by the first two equations of (4 9 ), each evaluated at each of the 256 states. Given these values for {wq, vx}, the 256 elements of vy are found by solving the% 256 equations represented by the last equation of (49) evaluated at each of the 256 states. Having solved for (wc„ vx, vy}, the remaining endogenous variables can be computed using equation (47) and equations (32) - (39). 30 References Backus, D., A. Gregory and C. Telmer, 1993, "Accounting for Forward Rates in Markets for Foreign Currency”, Journal of Finance 48. 1887-1908. Bansal, R., A. R. Gallant, R.Hussey and G. Tauchen, 1991, "Nonparametric Structural Estimation of Models for High-Frequency Currency Market Data," manuscript, Duke University. Bekaert, G., 1992, "The Time-Variation of Expected Returns and Volatility in Foreign Exchange Markets," Working'Paper No. 1226, Stanford University, Graduate School of Business. Bekaert, G., 1993a, "Exchange Rate Volatility and Deviations from Unbiasedness in a Cash-in-Advance Model," Journal of International Economics forthcoming. Bekaert, G., 1993b, "The Time Variation of Risk and Return in Foreign Exchange Markets: A General Equilibrium Approach," Working Paper No. 1276, Stanford University, Graduate School of Business. Bekaert, G. and R. Hodrick, 1992, "Characterizing Predictable Components in Excess Returns on Equity and Foreign Exchange Markets," Journal of Finance 47, 467-509. Bekaert, G. and R. Hodrick, 1993, "On Biases in the Measurement of Foreign Exchange Risk Premiums," Journal of International Money and Finance 12, 115-138. Bera, A. K. and C. M. Jarque, 1982, "Model Specification Tests: A Simultaneous Approach," Journal of Econometrics 20, 59-82. Bonomo, M. and R. Garcia, 1993, "Disappointment Aversion as a Solution to the Equity Premium and the Risk-Free Rate Puzzles," Cahier 2793, University de Montreal. Bossaerts, P. and P. Hillion, 1991, "Market Microstructure Effects of Government Intervention in the Foreign Exchange Market," Review of Financial Studies 4, 513-541. Camerer, C.F., 1989, "Recent Tests of Generalizations of Expected Utility Theory," manuscript. Cochrane, J. and L. Hansen, 1992, "Asset Pricing Explorations for Macroeconomics," NBER Macroeconomics Annual 1992 7. 115-165. Cumby, R. E. and J. Huizinga, 1992, "Testing the Autocorrelation Structure of Disturbances in Ordinary Least Squares and Instrumental Variables Regressions, Econometrica 60, 185-196. Engel, C., 1992, "On the Foreign Exchange Risk Premium in a General Equilibrium Model," Journal of International Economics 32, 305-319. Epstein, L. and S. Zin, 1989, "Substitution, Risk Aversion and the Temporal Behavior of Consumption and Asset Returns, Econometrica 57, 937-969. 31 Epstein, L. and S. Zin, 1990, "’First-Order’ Risk Aversion and the Equity Premium Puzzle," Journal of Monetary Economics 26, 387-407. Epstein, L. and S. Zin, 1991, "The Independence Axiom and Asset Returns," NBER Technical Working Paper No. 109. Fama, E., 1984a, "Forward and Spot Exchange Rates," Journal of Monetary Economics 14, 319-338. Fama, E., 1984b, "The Information in the Term Structure," Journal of Financial Economics 13, 509-528. Feenstra, R., 1986, "Functional Equivalence Between Liquidity Costs and the Utility of Money," Journal of Monetary Economics 17, 271-291. Gul, F., 1991, "A Theory of Disappointment Aversion," Econometrica 59, 667-710. Hall, R. E., 1988, "Intertemporal Substitution in Consumption," Journal of Political Economy 96, 339357. Hansen, L. P., 1982, "Large Sample Properties of Generalized Method of Moments Estimators," Econometrica 50, 1029-1054. Hansen, L. and R. Hodrick, 1980, "Forward Rates as Predictors of Future Spot Rates," Journal of Political Economy 8 8 , 829-853. Hansen, L. P., and R. Jagannathan, 1991, "Implications of Security Market Data for Models of Dynamic Economies," Journal of Political Economy 99, 225-262. Hansen, L. P. and S. Richard, 1987, "The Role of Conditioning Information in Deducing Testable Restrictions Implied by Dynamic Asset Pricing Models," Econometrica 55, 587-613. Hansen, L. P. and K. Singleton, 1982, "Generalized Instrumental Variables Estimation of Nonlinear Rational Expectations Models," Econometrica 50, 1269-1286. Harvey, C. R., 1991, "The World Price of Covariance Risk," Journal of Finance 46, March 1991, 111155. Heaton, J., 1991, "An Empirical Investigation of Asset Pricing with Temporally Dependent Preference Specifications," Manuscript, M.I.T. Hodrick, R.,1992, "An Interpretation of Foreign Exchange Market Efficiency Tests," manuscript, Kellogg Graduate School of Management, Northwestern University. Kandel and Stambaugh, 1991, "Asset Returns and Intertemporal Preferences," Journal of Monetary Economics 27. 39-71. Marshall, D., 1992, "Inflation and Asset Returns in a Monetary Economy," Journal of Finance 47, 13151342. Newey, W. and K. West, 1987, "A Simple, Positive Semi-definite, Heteroskedasticity and 32 Autocorrelation Consistent Covariance Matrix," Econometrica 55, 703-708. Roll, R., 1977, "A Critique of the Asset Pricing Theory’s Tests, Part I" Journal of Financial Economics 4, 129-176. Segal, U. and A. Spivak, 1990, "First-Order versus Second-Order Risk Aversion," Journal of Economic Theory 51. 111-125. Sims, C., 1980, "Macroeconomics and Reality," Econometrica 48, 1-49. * Solnik, B., 1993, "The Performance of International Asset Allocation Strategies Using Conditioning Information," Journal of Empirical Finance 1, 33-55. Stambaugh, R. 1988, "The Information in Forward Rates: Implications for Models of the Term Structure, " Journal of Financial Economics 21, 41-70. Tauchen, G. and Hussey, 1991, "Quadrature-Based Methods for Obtaining Approximate Solutions to Nonlinear Asset Pricing Models," Econometrica 59, 371-396. 33 Table 1 Panel A: Regression Results Dependent Variable Coef. on constant $t+i ” % 16.271 (3.674) Coef. on fpt Coef. on ft? Coef. on ft? -4.016 (0.766) R2 .2 2 0 +1.2 * i? 0.038 (0.050) hi+,.2 - s 0.075 (0.019) rT+, - i? 21.540 (4.864) -3.543 (0.816) .139 r?+. - i? 11.413 (4.971) -2.024 (0.900) .041 rT+i - 15.397 (4.807) -1.045 (0.954) .013 i\ -0.450 (0.129) .028 -0.448 (0.028) Panel B: Means and Standard Deviations Variable Mean Standard Deviation Ast+1 5.119 25.019 fa 3.698 3.077 fitted st+, - f, 1.421 12.355 ft? 0.124 0.707 0.094 1.892 fitted h*+1.2 - i* 0.094 0.318 ft? 0.116 0.826 M ut +l,2 - iv ll 0.128 1.259 fitted h*+12 - i* 0.128 0.370 rw - is rt-i *t 8.440 29.204 fitted r", - i* 8.440 10.899 ^t +1,2 " 34 .086 Table 1 (continued) Notes: The data are monthly observations on quarterly rates. The sample period is from January 1976 to December 1989 for exchange rates and equities and from October 1975 to June 1990 for interest rates. All rates are measured as percentage points per annum. Time subscripts denote quarters. The logarithms of the dollar/yen spot and forward exchange rates are denoted st and ft. The quarterly rate of depreciation is Ast+1; the three-month forward premium on the yen in terms of the dollar is denoted the quarterly dollar excess return on the world equity market (an equally-weighted average of the dollar excess returns to U.S. and Japanese equities, defined in equation (7)) is r*+, - i*; the three-month dollar excess return to U.S. equities is rf+1 - i*; the three month yen excess return to Japanese equities is r*+3 - if; h?+1>2 it (hl+i.2 * •*) is the quarterly excess dollar (yen) return from t to t + 1 obtained by holding dollar (yen) discount bonds that mature at t+ 2 ; fbf (fbj) is the one-quarter-ahead forward premium, defined in equation (5), in the dollar (yen) discount bond market. In Panel B, the variable "fitted s, +1 - ft" is the fitted value of regression (2 ); the variable "fitted hf+1>2 - i?" ("fitted h*+12 - i*") is the fitted value of regression (6 ) using data from the dollar (yen) bond market; the variable "fitted r* +1 - i*" is the fitted value of regression (8 ). The numbers in parentheses are standard errors, which are heteroskedasticityconsistent and are corrected for the serial correlation induced by the overlap in the data using the method of Newey and West (1987). 35 Table 2 VAR Estimates and Diagnostics for the Exogenous Processes Panel A: Parameter Estimates for the VAR constant gm,., gXt-i g°t-i gy«-i R2 gmt .610 (.446) .195 (.151) .022 (.437) -.182 (.074) .368 (.184) .074 gx, .629 (.141) .033 (.038) .281 (.118) -.016 (.024) .076 (.062) .057 gi\ .439 (.835) .023 (.228) 1.310 (.763) -.493 (.129) -.259 (.370) .198 gyt 1.118 (.287) .137 (.085) -.060 (.246) -.015 (.050) -.171 (.142) .009 Panel B: Cholesky Decomposition of the Covariance Matrix of the VAR Residuals .01261 (.00152) .00114 (.00034) .00054 (.00287) -.00056 (.00073) .00382 (.00043) .00332 (.00272) -.00038 (.00087) .02326 (.00177) .00371 (.00097) .00762 (.00059) 36 Table 2 (Continued) Panel C: Selection Criteria for the VAR Order VAR order Akaike Criterion Schwarz criterion 1 -36.61 -36.05 2 -36.45 -35.33 3 -36.43 -34.74 Likelihood ratio tests 2 19.06 (.266) 2 vs. 3 23.94 (.091) 1 vs. Panel D: Residual Diagnostics 1(4) Q2(4) Ku Sk BJ Eq. 1 2.481 (.894) 5.916 (.206) .172 (.784) .005 (.987) 0.075 (.963) Eq. 2 6 .1 1 0 (.191) 7.039 (.134) 1.137 (.070) -.046 (.883) 3.308 (.191) 7.198 (.126) 4.036 (.401) -.547 (.383) .2 0 0 1.169 (.558) Eq. 3 (.523) 14.27 -.500 .232 2.123 1.182 (.426) (.007) (.713) (.459) (.554) Notes: The sample period is 1974:4 to 1990:1. gm, denotes the growth rate of the dollar money supply; gxt denotes the growth rate of U.S. consumption; gn, denotes the growth rate of the yen money supply; gyt denotes the growth rate of Japanese consumption. Estimates of the VAR coefficients and the decomposition of the error covariance matrix are obtained by OLS and reported in Panels A and B with heteroskedasticity-consistent standard errors in parentheses. The appropriate lag length for the VAR minimizes the Akaike or Schwarz criterion in Panel C. The likelihood ratio test is a sequential test of a VAR(n) versus a VAR(n+1). The test statistic has a x 2 distribution with degrees of freedom equal to the number of coefficients being restricted by the lower VAR order. The statistic incorporates the Sims (1980) correction. Marginal levels of significance are in parentheses. Panel D reports statistics and associated p-values for various residual diagnostic tests. Column 1 reports the Cumby-Huizinga (1992) 1-test for serial correlation of the residuals. Column 2 reports the Ljung-Box test statistic, applied to squared residuals, as a test for ARCH. Columns 3, 4, and 5 test for normality of the residuals. Ku is the normalized kurtosis coefficient and Sk the normalized skewness coefficient. Their asymptotic distribution is N(0,24/T), N(0,6/T) respectively, with T the sample size, under the null of normality. BJ is the Bera-Jarque (1982) test fpr normality and is x2(2). Eq. 4 37 Table 3 VAR Parameters Induced by the Discretized Markov Chain Approximation (i) Implied VAR Coefficients constant gm,., gXt-l gn,-, gy»-i gpl, .610 .194 .022 -.181 .368 gXt .629 .033 .280 -.016 .075 gn. .443 .023 1.299 -.489 -.255 gy. 1.118 .137 -.062 -.014 -.170 (ii) Cholesky Decomposition of the Implied Residual Covariance Matrix .01261 .00114 .00054 -.00056 .00382 .00331 -.00038 .02312 .00369 .00762 Notes: The discrete state space approximation to the VAR in Table 2 is computed using the Gaussian quadrature procedure of Tauchen and Hussey (1991). Each variable is allowed to take four possible values, implying a discrete state space with 256 elements. Variable definitions are given in the notes to Table 2. 38 Table 4 Implications of the Model for the Foreign Exchange Market Regression • p = .5 A = 1.0 A = .85 A = .70 A = .55 ft -.007 - .0 1 2 -.068 -.097 R2 l.OxlO"7 2 . 1xl 0 '5 8 . 8 xl 0 '5 .00016 00030 .00133 = -3 .042 .085 .116 .159 .332 ff[Et(As,+1)] .228 .229 .226 .230 .274 .370 -.0004 .0 0 1 1 .0038 .0087 .0269 .1017 13, -.007 -.023 -.057 -.107 .035 -.044 R2: l.OxlO"7 1 .8 xl 0 '5 7.7xl0'5 .00015 .00029 .00116 .003 .039 .080 .113 .155 .309 .236 .231 .229 .228 .269 .363 -.0004 .0003 .0034 .0076 .0256 .0941 Pi -.0 0 1 -.003 -.006 - .0 2 1 -.009 .057 R2 l.OxlO'7 2 . 1 xl 0"5 8 .6 xlO-J .00017 .00032 .00133 .003 .042 .085 .118 .164 .334 .485 .472 .461 .443 .438 .509 - .0 0 0 1 .0 0 1 1 .0061 .0099 .0255 .1190 .0 0 1 .003 .005 .007 .008 .009 2 . 1 xl 0'5 8 .2 xl 0 '5 .00015 .00032 .00138 .004 .044 .088 .1 2 0 .173 .356 2.284 2 .2 0 0 2 .1 0 0 1.973 1.807 1.575 .0157 .0311 .0422 .0570 .1472 ff[rpJ ff[E,(Ast+1)] ^[rpJ a[Et(Ast+I)] cov[rp„ E,(Ast+1)] Pi R2 p = -9 -.191 .003 cov[rp„ El(Ast+1)] P .038 A = .25 ff[rpj cov[rp„ El(Asl+1)] p = -.33 A = .40 ff[rpj a[Et(Asl+1)] cov[rp„ El(Ast+,)] 1.7x1O'7 .0046 Notes: The logarithms of the dollar/yen spot and forward exchange rates are denoted s, and f„ and rpt = El(st+1-Q. /?, denotes the slope coefficient in the regression st4l - ft = /S0 + j8 ,(ft - st) + et4l. EtCXj+j) denotes the expectation of x, +1 conditional on date t information, a[xj denotes the unconditional standard deviation of xt, and cov[xt, yj denotes the unconditional covariance. R2 = var(Et(sl4,-ft))/var(st4l-ft). All moments reported are the exact population moments implied by the model at the indicated parameter specifications, given the Markov transition matrix for the exogenous process gt. This transition matrix was computed using Gaussian quadrature from the estimated VAR, as described in Appendix B. 39 Table 5 Implications of the Model for the Dollar Discount Bond Market Regression p = .5 A = 1.0 A = .85 A = .70 A = .55 A = .40 A = .25 is, - .0 0 0 0 1 - .0 0 0 2 2 -.00026 .00013 -.00196 .00084 R2 1 .0 x 1 0 '* 7.7X10-6 2 .6 x 1 0 5.4xl0's .00014 .00053 .00067 .0 0 1 2 1 .00172 .00272 .00512 .2 2 0 .218 .215 .2 1 1 .206 .199 -5.0xl0'7 - 1 .0 x 1 0 s -l.OxlO-5 8 .6 x 1 0 * -7.6xl0'5 -7.2x10* Pi .00005 .00042 -.00031 .00046 .00048 -.00205 R2: 5.0x10'* 1.4xl0'5 5.1x10 s .0 0 0 1 2 .00024 .00087 ff(rpt) Et[Av*+1] .00005 .00078 .00146 .00219 .00297 .00528 .230 .226 .2 2 1 .214 .206 .192 3X10-6 -2 .2 x 1 0 s .00013 2 .6 x 1 0 2.9xl0'5 -4.8x10 s Pi -.00006 -.00031 -.00113 -.00128 -.0 0 1 2 2 -.00491 R2 4.0x10* 4.7x10^ 2 .2 xl 0 '5 4.9xl0'5 .0 0 0 1 0 .00034 .00003 . <*rb*) E,[Av2+I] covfrbf, E,(Av*+1)] p = -.33 covfrb*, Et(Av*+1)] p = -3 s a(rb*) .0 0 0 2 .0 0 2 1 .0044 .0062 .0083 .0135 E,[Av*+1] 1.109 1.074 1.030 .977 .910 .805 -.00036 -.00118 -.00119 -.00095 0.00303 -.0006 - .0 0 1 2 -.0018 -.0025 -.0047 1.9x10 s 7.4x10 s .00019 .00045 .0 0 1 1 2 cov[rb*), Et(Av*+1)] P = -9 s -.00007 Pi - .0 0 0 1 R2 1 . 1x 1 0 ^ a(rb?) .006 .024 .045 .067 .093 .123 E,[Av*+l] 5.995 5.725 5.406 5.016 4.516 3.798 -.0047 -.0176 -.0318 -.0421 -.0342 -.0528 cov[rbf), E,(Avf+,)] Notes: h*+u denotes the continuously compounded one-period holding period returnon two-period dollar discount bonds; ifdenotes the continuously compounded dollar spot interest rate; Av*+,denotes the rate of change in the logarithm of the price of one-period dollar bonds; and rb* = Et(h*+12-is). /S, denotes the slope coefficient in the regression h*,>2- i,s = var^E,(ht*l2-i*))/var(h**,2-it$). See also the note to Table 4. 40 P0 + /S,(fbts) + et4,. R2 = Table 6 Implications of the Model for the Yen Discount Bond Market Regression P = -5 A = 1.0 A = .85 A = .70 A = .55 A = .40 A = .25 fit -.00013 .0 0 0 0 2 -.00044 -.00052 .00214 .00169 R2 2 .0 xl 0 '7 5.0xl0'5 .0 0 0 1 2 .00024 .00054 .00195 ?(rb») .00007 .00091 .00176 .00247 .00352 .00658 .249 .247 .243 .238 .230 .223 -8 x 10 ‘6 2x1O'4 -2.3xl0'5 -2.3x10 s .000126 .000127 fit - .0 0 0 2 1 -.00024 -.00095 -.00118 .00059 -.00036 R2: 3.5xl0'7 3.1xl0‘5 .00013 .00027 .00059 .00193 <r(rb?) .0 0 0 1 2 .00113 .00226 .00313 .00442 .00757 .347 .340 .333 .322 .306 .288 -2 xl 0 6 -3x1O'6 - .0 0 0 1 0 - .0 0 0 1 1 .00007 .00003 fit -.00027 -.00081 -.00176 -.00256 -.00276 -.00532 R2 3 .2 x 10 ‘7 1.2x1 O’5 4.8x10‘5 .0 0 0 1 0 .0 0 0 2 0 .00067 o i jb \) .0004 .0023 .0045 .0062 .0080 .0131 .950 .919 .883 .836 .774 .690 -.00024 -.00068 -.00136 -.00176 -.00160 -.00239 fix -.0003 -.0009 -.0017 -.0028 -.0039 -.0065 R2 7.2xl0 7 1.4x10 s 5.3xl0'5 .00013 .00029 .00073 a [ E ,( Av*+I)] cov[rbI, E,(AvI+1)] p = -.33 a[Et(AvI+I)l cov[rb][) E,(Av*+1)] p = -3 a[E,(Av*+I)] cov[rb^, Et(Avf+I)] p = -9 o is b \) .036 .013 .024 .034 .046 0628 a[E,(Av{+1)] 4.062 3.872 3.648 3.374 3.025 2.541 0.00422 -.01396 -.02262 -.03069 -.03398 -.03876 cov[rb*, E,(Av*+,)] Notes: h*+1>2 denotes the continuously compounded one-period holding period return on two-period yen discount bonds; i* denotes the continuously compounded yen spot interest rate; Avf+, denotes the rate of change in the logarithm of the price of one-period yen bonds; and rbf = E ^+ ^-if). /S, denotes the slope coefficient in the regression ht4i,2 - i* = /S0 + /3,(fbtr) + et<l. R2 = v a ^ E ^ ^ -i^ /v a r^ h ^ -i* ). See also the note to Table 4. 41 Table 7 Implications of the Model for the Excess Dollar Return on Aggregate Wealth Panel A: Predictability of Excess Dollar Return on Aggregate Wealth A = 1.0 , R2 0 P = -5 o[E,(rl+1-i«)] p = -.33 /S, R2 ^[Et(rt+j-iJ] , R2 0 p = -3 <r[Ei(rt+l-i,)] , R2 0 p = -9 ^[E,(rt+rO] A = .85 .003 - .0 0 1 l.OxlO"8 A = .70 A = .55 -.015 -.039 4.2x10-6 2 . 1x 1 0 .0 0 1 .013 - .0 0 1 1 .0 x 1 0 '* .0 0 1 .0 0 0 1 .0 x 10 '8 s A = .40 -.034 A = .25 -.085 4.6x10 s .0 0 0 1 2 .00042 .028 .042 .068 .128 -.0 0 1 -.004 -.049 -.040 .006 3 .6 x 1 0 "* 2 .0 x 1 0 s 4.6x10 s .0 0 0 1 1 .00039 .0 1 2 .028 .042 .067 .123 -.008 - .0 2 0 .0 0 2 .0 0 2 .025 3.1x10-* 1.4x10 s 3.4x10 s .0 0 0 1 0 .00033 .0 0 1 .0 1 2 .025 .039 .068 .1 2 2 .0 0 1 .003 .006 .0 1 0 8 . 1 x 10 -* 2 .8 xl 0"5 .028 .051 3.7xl0'7 .006 .009 .009 7.5xl0"s .00016 .00040 .082 .115 .175 Panel B: Mean of (rt+, - i,) A = 1.0 A = .85 A = .70 A = .55 A = .40 A = .25 p = .5 0.060 0.227 0.430 0 .6 8 8 0.999 1.510 p = -.33 0.062 0.238 0.447 0.718 1.047 1.563 p = -3 0.077 0.315 0.597 0.940 1.356 1.991 p = -9 0.168 0.655 1.205 1.843 2.591 3.566 Notes: r, denotes the continuously compounded dollar return to the aggregate wealth portfolio; i, denotes the continuously compounded dollar spot interest rate. /S, denotes the slope coefficient in the regression r«.i " », = /30 + - st) + etM. R2 = var(Et(rul -it)) / var(rl4l -it). See also the note to Table 4. 42 Table 8 Implications of the Model for Unconditional Standard Deviations Panel A: (st+1 - s,) A = 1.0 A = .85 A = .70 A = .55 A = .40 A = .25 p = .5 9.118 9.121 9.122 9.125 9.110 9.093 p = -.33 9.166 9.121 9.120 9.125 9.111 9.081 p = -3 9.184 ' 9.186 9.188 9.193 9.175 9.167 p = -9 10.066 10.026 9.921 9.978 9.833 9.702 Panel B: (ft - st) A = 1.0 A = .85 A = .70 A = .55 A = .40 A = .25 p = .5 .230 .227 .225 .221 .215 .209 p = -.33 .237 .233 .228 .223 .212 .198 p = -3 .486 A ll .456 .436 .410 .364 9 = “9 2.282 2.193 2.087 1.955 1.784 1.521 Panel C: (r, +1 - i,) A = 1.0 A = .85 A = .70 A = .55 A = .40 A = .25 p = .5 6.152 6.157 6.165 6.170 6.178 6.188 p = -.33 6.230 6.235 6.240 6.249 6.255 6.254 p = -3 6.802 6.791 6.782 6.765 6.741 6.714 P = -9 10.002 9.850 9.672 9.458 9.163 8.746 Panel D: (hf+12 - i*) A = 1.0 A = .85 A = .70 A = .55 A = .40 A = .25 p = .5 0.244 0.242 0.239 0.235 0.230 0.222 p = -.33 0.214 0.210 0.205 0.200 0.193 0.179 p = -3 0.998 0.966 0.928 .0.882 0.822 0.728 P = -9 5.759 5.503 5.199 4.827 4.352 3.699 43 Table 8 (Continued) Panel E: (hf+u - if) A = 1.0 A = .85 A = .70 A = .55 A = .40 A = .25 p = .5 0.166 0.164 0.162 0.159 0.154 0.149 p = -.33 0.206 0 .2 0 2 0.197 0.191 0.182 0.172 p = -3 0.697 0.674 0.647 0.612 0.567 0.505 P = -9 3.685 •' 3.510 3.304 3.054 2.735 2.292 Notes: See Tables 4-7. 44 Working Paper Series A series of research studies on regional economic issues relating to the Seventh Federal Reserve District, and on financial and economic topics. REGIONAL ECONOMIC ISSUES Estimating Monthly Regional Value Added by Combining Regional Input With National Production Data WP-92-8 Philip'R. Israilevich and Kenneth N. Kuttner Local Impact of Foreign Trade Zone WP-92-9 David D. Weiss Trends and Prospects for Rural Manufacturing WP-92-12 William A. Testa State and Local Government Spending—The Balance Between Investment and Consumption WP-92-14 Richard H. Mattoon Forecasting with Regional Input-Output Tables WP-92-20 P.R. Israilevich, R. Mahidhara, and G.J.D. Hewings A Primer on Global Auto Markets WP-93-1 Paul D. Ballew and Robert H. Schnorbus Industry Approaches to Environmental Policy in the Great Lakes Region WP-93-8 David R. Allardice, Richard H. Mattoon and William A. Testa The Midwest Stock Price Index—Leading Indicator of Regional Economic Activity WP-93-9 William A. Strauss Lean Manufacturing and the Decision to Vertically Integrate Some Empirical Evidence From the U.S. Automobile Industry WP-94-1 Thomas H. Klier Domestic Consumption Patterns and the Midwest Economy WP-94-4 Robert Schnorbus and Paul Ballew 1 Working paper series continued To TradeorNot toTrade: Who ParticipatesinRECLAIM? WP-94-11 Restructuring& WorkerDisplacementintheMidwest WP-94-18 ThomasH.KlierandRichardMattoon PaulD.BallewandRobertH.Schnorbus ISSUES IN FINANCIAL REGULATION IncentiveConflictinDeposit-InstitutionRegulation: Evidencefrom Australia EdwardJ.KaneandGeorgeG.Kaufman WP-92-5 CapitalAdequacy and theGrowth ofU.S.Banks WP-92-11 Bank Contagion: Theory and Evidence WP-92-13 TradingActivity,ProgarmTradingandtheVolatilityofStockReturns WP-92-16 PreferredSourcesofMarket Discipline: Depositorsvs. SubordinatedDebt Holders WP-92-21 HerbertBaerandJohnMcElravey GeorgeG.Kaufman JamesT.Moser DouglasD.Evanoff An InvestigationofReturnsConditional onTradingPerformance JamesT.MoserandJackyC.So The EffectofCapitalon PortfolioRiskatLifeInsuranceCompanies ElijahBrewerHI,ThomasH.Mondschean,andPhilipE.Strahan W P-92-24 WP-92-29 A Framework forEstimatingtheValueand InterestRateRiskofRetailBank Deposits DavidE.Hutchison,GeorgeG.Pennacchi WP-92-30 CapitalShocks andBank Growth-1973 to 1991 WP-92-31 The ImpactofS&L FailuresandRegulatoryChanges on theCD Market 1987-1991 WP-92-33 HerbertL.BaerandJohnN.McElravey ElijahBrewerandThomasH.Mondschean 2 Working paper series continued Junk Bond Holdings, Premium Tax Offsets, and Risk Exposure at Life Insurance Companies WP-93-3 Elijah Brewer III and Thomas H. Mondschean Stock Margins and the Conditional Probability of Price Reversals WP-93-5 Paul Kofman and James T. Moser Is There Lif(f)e After DTB? Competitive Aspects of Cross Listed Futures Contracts on Synchronous Markets WP-93-11 Paul Kofman, Tony B o u w m a n and James T. Moser Opportunity Cost and Prudentiality: A RepresentativeAgent Model of Futures Clearinghouse Behavior WP-93-18 Herbert L. Baer, Virginia G. France and James T. Moser The Ownership Structure of Japanese Financial Institutions WP-93-19 Hesna Genay Origins of the Modern Exchange Clearinghouse: A History of Early Clearing and Settlement Methods at Futures Exchanges WP-94-3 James T. Moser The Effect of Bank-Held Derivatives on Credit Accessibility WP-94-5 Elijah Brewer III, Bernadette A. Minton and James T. Moser Small Business Investment Companies: Financial Characteristics and Investments WP-94-10 Elijah Brewer III and Hesna Genay MACROECONOMIC ISSUES An Examination of Change in Energy Dependence and Efficiency in the Six Largest Energy Using C o u n tries-1970-1988 WP-92-2 Jack L. Hervey Does the Federal Reserve Affect Asset Prices? WP-92-3 Vefa Tarhan Investment and Market Imperfections in the U.S. Manufacturing Sector WP-92-4 Paula R. Worthington 3 Working paper series continued Business Cycle Durations and Postwar Stabilization of the U.S. Economy WP-92-6 Mark W. Watson A Procedure for Predicting Recessions with Leading Indicators: Econometric Issues and Recent Performance W P -92-7 James H. Stock and Mark W. Watson Production and Inventory Control at the General Motors Corporation During the 1920s and 1930s W P -92-10 Anil K. Kashyap and David W. Wilcox Liquidity Effects, Monetary Policy and the Business Cycle W P -92-15 Lawrence J. Christiano and Martin Eichenbaum Monetary Policy and External Finance: Interpreting the Behavior of Financial Flows and Interest Rate Spreads W P -92-17 Kenneth N. Kuttner Testing Long Run Neutrality W P -92-18 Robert G. King and Mark W. Watson A Policymaker's Guide to Indicators of Economic Activity Charles Evans, Steven Strongin, and Francesca Eugeni W P -92-19 Barriers to Trade and Union Wage Dynamics W P -92-22 Ellen R. Rissman Wage Growth and Sectoral Shifts: Phillips Curve Redux W P -92-23 Ellen R. Rissman Excess Volatility and The Smoothing of Interest Rates: An Application Using Money Announcements W P -92-25 Steven Strongin Market Structure, Technology and the Cyclicality of Output W P -92-26 Bruce Petersen and Steven Strongin The Identification of Monetary Policy Disturbances: Explaining the Liquidity Puzzle W P -92-27 Steven Strongin 4 Working paper series continued Earnings Losses and Displaced Workers WP-92-28 Louis S. Jacobson,Robert J. LaLonde, and Daniel G. Sullivan Some Empirical Evidence of the Effects on Monetary Policy Shocks on Exchange Rates WP-92-32 Martin Eichenbaum and Charles Evans An Unobserved-Components Model of Constant-Inflation Potential Output WP-93-2 Kenneth N. Kuttner Investment, Cash Flow, and Sunk Costs WP-93-4 Paula R. Worthington Lessons from the Japanese Main Bank System for Financial System Reform in Poland Takeo Hoshi, Anil Kashyap, and Gary Loveman WP-93-6 Credit Conditions and the Cyclical Behavior of Inventories WP-93-7 Anil K. Kashyap, O w e n A. Lamont and Jeremy C. Stein Labor Productivity During the Great Depression WP-93-10 Michael D. Bordo and Charles L. Evans Monetary Policy Shocks and Productivity Measures in the G-7 Countries WP-93-12 Charles L. Evans and Fernando Santos Consumer Confidence and Economic Fluctuations WP-93-13 John G. Matsusaka and Argia M. Sbordone Vector Autoregressions and Cointegration WP-93-14 M a r k W. Watson Testing for Cointegration When Some of the Cointegrating Vectors Are Known WP-93-15 Michael T. K. Horvath and M a r k W. Watson Technical Change, Diffusion, and Productivity WP-93-16 Jeffrey R. Campbell 5 Working paper series continued Economic Activity and the Short-Term Credit Markets: An Analysis o f Prices and Quantities WP-93-17 Benjamin M. Friedman and Kenneth N. Kuttner Cyclical Productivity in a Model of Labor Hoarding WP-93-20 Argia M. Sbordone The Effects of Monetary Policy Shocks: Evidence from the Flow of Funds WP-94-2 Lawrence J. Christiano, Martin Eichenbaum and Charles Evans Algorithms for Solving Dynamic Models with Occasionally Binding Constraints WP-94-6 Lawrence J. Christiano and Jonas D.M. Fisher Identification and the Effects of Monetary Policy Shocks WP-94-7 Lawrence J. Christiano, Martin Eichenbaum and Charles L. Evans Small Sample Bias in GMM Estimation of Covariance Structures WP-94-8 Joseph G. Altonji and Lewis M. Segal Interpreting the Procyclical Productivity of Manufacturing Sectors: External Effects of Labor Hoarding? WP-94-9 Argia M. Sbordone Evidence on Structural Instability in Macroeconomic Time Series Relations . Stock and M a r k W. Watson WP-94-13 The Post-War U.S. Phillips Curve: A Revisionist Econometric History WP-94-14 James H Robert G. King and M a r k W. Watson The Post-War U.S. Phillips Curve: A Comment WP-94-15 Charles L. Evans Identification of Inflation-Unemployment WP-94-16 Bennett T. McCalhun The Post-War U.S. Phillips Curve: A Revisionist Econometric History Response to Evans and McCallum WP-94-17 Robert G. King and M a r k W. Watson 6 Working paper series continued Estimating Deterministic Trends in the Presence of Serially Correlated Errors WP-94-19 Eugene Canjels and M a r k W. Watson Solving Nonlinear Rational Expectations Models by Parameterized Expectations: Convergence to Stationary Solutions WP-94-20 Albert'Marcet and David A. Marshall The Effect of Costly Consumption Adjustment on Asset Price Volatility WP-94-21 David A. Marshall and Nayan G. Parekh The Implications of First-Order Risk Aversion for Asset Market Risk Premiums WP-94-22 Geert Bekaert, Robert J. Hodrick and David A. Marshall 7