The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
w o r k i n g p a p e r 9 8 2 3 Central Bank Intervention and Overnight Uncovered Interest Rate Parity by Richard T. Baillie and William P. Osterberg FEDERAL RESERVE BANK OF CLEVELAND Working Paper 9823 Central Bank Intervention and Overnight Uncovered Interest Rate Parity by Richard T. Baillie and William P. Osterberg Richard T.Baillie is Professor of Economics at Michigan State University, East Lansing, Michigan, and a visiting consultant with the Federal Reserve Bank of Cleveland. William P. Osterberg is an Economist at the Federal Reserve Bank of Cleveland. The authors thank Jennifer DeRudder for research assistance. Working papers of the Federal Reserve Bank of Cleveland are preliminary materials circulated to stimulate discussion and critical comment on research in progress. They may not have been subject to the formal editorial review accorded official Federal Reserve Bank of Cleveland publications. The views stated herein are those of the authors and are not necessarily those of the Federal Reserve Bank of Cleveland or of the Board of Governors of the Federal Reserve System Working papers are now available electronically through the Cleveland Fed's home page on the World Wide Web: http://www.clev.frb.org. January 1999 Central Bank Intervention and Overnight Uncovered Interest Rate Parity By Richard T. Baillie and William P. Osterberg This paper considers the impact of U.S. and German central bank intervention on the risk premium in forward foreign exchange markets. The model estimation is facilitated with the use of daily data on overnight Eurocurrency deposit rates, so that the interest rate maturity time of one day matches the sampling interval of the data. We also use the official net daily purchases and sales of dollars vis-à-vis the German Mark by the Federal Reserve System and the Bundesbank. The model involves FIGARCH innovations to model the degree of long term dependence in the volatility process. Some support is found for the intervention variables affecting the risk premium as predicted by theory. The impact of intervention in the two years immediately following the meltdown of the equity markets in October 1987 is particularly strong. Keywords: Exchange Rates, Central Bank Intervention, Risk Premium, FIGARCH. JEL Classification numbers: C22, E41, E31 CENTRAL BANK INTERVENTION AND OVERNIGHT UNCOVERED INTEREST RATE PARITY by Richard T Baillie (Michigan State University) and William P Osterberg* (Federal Reserve Bank of Cleveland) This version: December, 1998 Address for Correspondence: Richard T Baillie, Department of Economics, Michigan State University, East Lansing, MI 48824. Phone: (517) 355-1864; email: baillie@pilot.msu.edu * The views in this paper do not necessarily reflect the views of the Federal Reserve Bank of Cleveland, or of the Board of Governors of the Federal Reserve System. The authors thank Jennifer DeRudder for research assistance. 1 Abstract This paper considers the impact of U.S. and German central bank intervention on the risk premium in forward foreign exchange markets. The model estimation is facilitated with the use of daily data on overnight Eurocurrency deposit rates, so that the interest rate maturity time of one day matches the sampling interval of the data. We also use the official net daily purchases and sales of dollars vis-%-vis the German Mark by the Federal Reserve System and the Bundesbank. The model involves FIGARCH innovations to model the degree of long term dependence in the volatility process. Some support is found for the intervention variables affecting the risk premium as predicted by theory. The impact of intervention in the two years immediately following the meltdown of the equity markets in October 1987 is particularly strong. Keywords: Exchange Rates, Central Bank Intervention, Risk Premium, FIGARCH. JEL Classification numbers: C22, E41, E31. 2 1. Introduction While central bank intervention has at times been quite substantial in the post Bretton Woods era, there continues to be controversy over its effectiveness in achieving the policy goals of either changing the level of nominal exchange rates or of reducing volatility. A large literature has examined the usefulness of intervention; see Edison (1993) and Almekinders (1995). The general conclusion seems to be that the policy either has consequences that vary with the sample period, effects that are inconsistent with the theory, or ultimately has little impact on nominal exchange rates. On the one hand, Dominguez and Frankel (1993) and Ghosh (1992) find support for the effectiveness of intervention. However, Baillie and Osterberg (1997a, 1997b) find evidence that G3 intervention either has no statistical effect, or that it has outcomes which are the opposite to those intended. Econometric analysis of the impact of central bank intervention is generally constrained by the availability of official data. Clearly the nature of the volatility of asset price markets and the likely short-lived nature of risk premia make it desirable to use daily data or very high frequency data. Ideally, we would utilize intraday data, as in the study by Goodhart and Hesse (1993). In this study we use the officially recorded net intervention by the Federal Reserve Bank over the previous 24 hours, and similar but confidential data kindly supplied by the Bundesbank. The appropriate sampling frequency of the data is related to the hypothesized transmission mechanism of intervention. Studies such as Ghosh (1992), Obstfeld (1989) and Humpage (1988) which relate intervention to portfolio balances are limited to monthly data and have been generally inconclusive as to the validity of the portfolio balance effect. The possibility that intervention signals a change in monetary policy is examined by Dominguez (1990) who uses weekly data on monetary surprises, exchange rates 3 and intervention credibility of and finds monetary its effectiveness policy. Conversely, varies if with an the !small intervention can provide a signal of monetary policy (Klein (1992)), then the impact of the signal might be largely unrelated to fundamentals such as money supply. Generally, studies that have examined the effects of daily intervention on daily spot exchange rates have either found no effects (e.g. Baillie and Osterberg (1997a)), or effects that are extremely weak, e.g. Goodhart and Hesse (1993). A small number of studies have assessed intervention s impact on the mean and/or conditional variance of deviations from uncovered interest rate parity (UIRP). Loopesko s (1984) approach utilized cumulative intervention flows from the beginning of her sample period while Dominguez (1992) analyzed the impact of daily flows. Humpage and Osterberg (1992) tried both approaches in their analysis of the conditional mean and conditional variance of deviations from UIRP. One theoretical motivation for an impact of intervention on UIRP is provided by Osterberg (1989) and Baillie and Osterberg (1997b) who formulated a two country inter-temporal asset pricing model which implied that central bank foreign exchange intervention affects the forward exchange risk premium. Baillie and Osterberg (1997b) found empirical support for intervention influencing the risk premium in the forward DM-$ and Yen-$ markets. Purchases of dollars by the Federal Reserve System were found to be associated with excess $ denominated returns, and furthermore, there was evidence that intervention increased rather than reduced exchange rate volatility. The Role of Intervention The Federal Reserve System and the Bundesbank appear to routinely sterilize their interventions so that the purchase (sale) of foreign currency is offset by a corresponding sale (purchase) of domestic government debt to eliminate the effects on 4 domestic money supply. Clearly, unsterilized intervention is equivalent to monetary policy and is more likely to directly effect exchange rates. However, even sterilized intervention might be linked to monetary policy. The signaling literature the views central intervention bank s future as working monetary either policy or by by operating via a portfolio-balance effect. The latter approach is motivated by mean-variance optimization, where agents are concerned with terminal wealth composed of domestic and foreign currencies and bonds. Sterilized intervention will alter the relative supplies of domestic money and bonds. With risk averse investors who view domestic and foreign bonds as imperfect substitutes, the impact of intervention will adjust the relative rate of return by changing the exchange rate. However, the portfolio balance theory implies no impact of intervention on the exchange rate when there is perfect substitutability of bonds and/or Ricardian equivalence, so that consumers exactly anticipate future taxes associated with government debt. Any test of the theory requires information on the relative supplies of the assets. The alternative view of intervention as a signal of the central bank's future monetary policy implies that a sterilized purchase of foreign currency is expected to lead to a depreciation of the exchange rate if the foreign currency purchase is assumed to signal a more expansionary domestic monetary policy. Klein and Rosengren (1991) find no consistent relationship between intervention and monetary policy and Kaminsky and Lewis (1996) report that the impact of intervention on exchange rates has sometimes been inconsistent with the implied monetary policy. Humpage (1997) concludes that the US authorities in the 1990s had no information superior to the market so that intervention could not be viewed as signaling new information about monetary policy. Dominguez and Frankel (1993) found inconclusive evidence on the 5 signalling and portfolio balance transmission mechanisms, while Ghosh (1992) found that variables associated with the portfolio balance approach appear to have little effect on spot exchange rates. Definitions For subsequent analysis, we define St as spot exchange rate in terms of DM-$ at time t, Ft is the forward exchange rate at time t, for delivery at time t+k and Pt is the domestic price level. The UIRP condition is, (1) Et[(Ft - St+k)/Pt+k] = 0, since expected real returns in the forward market are zero. taking a Taylor series expansion to second order terms, (2) Etst+k - ft = -(1/2)Vart(st+1) + Covt(st+1pt+1), (3) Et∆st+1 - (it* - it) = -(1/2)Vart(st+1) + Covt(st+1pt+1), On where lower case variables denote the logarithms of variables in levels, and it is the dollar return on a risk free $ denominated bond, and it* is the foreign currency return on a risk free bond denominated in terms of the foreign currency. Usually, the ex post deviation from UIRP is expressed as, yt = ∆st+k - (it* - it), where the two terms on the right hand side of (2) are neglected. Hence the country with the higher rate of interest is expected to have the depreciating currency. The forward premium anomaly found in many studies is that a regression of the form of, st+k - st = + β(it* - it) + ut+k, is found to have a negative slope coefficient, β, which implies that the country with the higher rate of interest is expected to 6 have a currency appreciation. A generalization of equation (2) is to specify real returns over the current and future consumption stream, so that equation (3) is modified to the discrete time asset pricing approach of Lucas (1978) to be, Et[(Ft - St+k)/Pt+k]U/(Ct+k)/U/(Ct) = 0, (4) where U/(Ct+1)/U/(Ct) is the marginal rate of substitution in terms of utility derived from current and future consumption. On assuming a logarithmic utility function with a constant coefficient of relative risk aversion, (CRRA), denoted by γ, then equation (4) can be expressed as, (5) = -(1/2)Vart(st+1) + Covt(st+1pt+1) + γCovt(st+1ct+1). Etst+1 - ft The last term, will be denoted by ρt = γCovt(st+1ct+1) and is known as a time dependent risk premium. There are several possible theoretical developments of this term. For example, if each country's consumption growth is assumed to be proportional to world income growth so that Ct Ytγ, the CRRA assumption, the risk premium term in (6) is ρt = Covt(st+1yt+1γ). More generally, ρt the ρCovt(st+1qt+1), risk where premium qt is is the = then frequently logarithm of under expressed the as, = intertemporal marginal rate of substitution. It should be noted that in many previous studies of time varying risk premium, e.g. Domowitz and Hakkio (1985), Hodrick (1987, 1989) and Kaminsky and Peruga (1990), the xt variables typically contain conditional variances and covariances of the asset price vis-%-vis the forcing variables. However, the fact that goods market variables needed for the Lucas-Breeden model such as relative prices, consumption levels, money supplies, etc 7 are generally only observed monthly and the fact that goods market prices are very smooth compared to asset market prices makes estimation and testing these relationships problematic. Also, as noted by Baillie and Bollerslev (1989), exchange rates typically possess little ARCH effects at the monthly level, so that studies such as Kaminsky and Peruga (1990) have been unsuccessful in collaborating the Lucas Breeden asset pricing model. Consequently, it is attractive to estimate the model from daily data where the effects of changes in the intervention variables may be more clearly apparent. Data This study uses data provided by the Board of Governors of the Federal Reserve System and the German Bundesbank from January 3, 1987 through January 22, 1993. This includes the periods around the Louvre Accord and the stock market crash in October, 1987. For each country, the data record in 100 million US dollar units the actual net purchases of US dollars vis a vis the German mark from close of business on day t-2 to close of business on day t-1. The exchange rates are recorded at 9:30am Paris time as supplied to us by Olsen & Associates of Zurich, Switzerland. For each bilateral exchange rate there are four intervention variables: the purchases and sales of dollars by each country. The intervention variables are aligned so as to be predetermined with respect to the change in the exchange rate from day t to t+1. We use a unique data set on overnight Eurocurrency deposit rates obtained from the Paris market through DRIFACS, with the ultimate source being Credit Lyonnais, Paris. The use of this data, which are essentially interest rates of one day maturity time, allows us to avoid many of the econometric problems associated with forward rates with overlapping contracts, which reduce the efficiency in tests of unbiased expectations of the forward rate. With one month forward contracts it is important to elaborately match in accordance with the settlement conventions as 8 described by Riehl and Rogriguez (1977), Levine (1989) and Bekaert and Hodrick (1993). Figure 1 shows the movement in the DM/$ rate and the overnight interest rates during the sample period. The U.S. rate exceeded that of Germany until Fall 1990. UIRP requires that higher(lower) U.S. rates be offset by an expected decline(rise) in the DM/$. However, causality could run in either direction. The DM/$ began a steady decline from over 2 in June 1989. Figure 2 shows the logarithmic analogues which together comprise the deviation from UIRP. Clearly, and not surprisingly, the volatility in the exchange rate exceeds that of the interest rates. Figure 3 shows the logarithmic change in the exchange rate together with U.S. intervention vis-%-vis the DM, with positive intervention indicating purchases of dollars. It is not clear from Figure 3 whether U.S. intervention was intended to affect either the level or the volatility of the DM/$. There were only two possibly distinct intervals of intervention activity-the period around the Louvre Accord of February 20-21, 1987 and the equity market crash of October 1987. An Econometric Model for UIRP An econometric model for the ex post deviations from uncovered interest rate parity is given by, (6) yt =(∆st+1 - it* + it) = b/xt + εt+1, (7) εt = (8) σ2t = ω + βσ2t-1 + (1 - βL - (1 - ϕL)(1 - L)δ)ε2t, ξtσt, where ξt is i.i.d.(0,1) process, b are a k dimensional vector of predetermined variables, and σ2t is a time-varying, positive and measurable function of the information set at time t-1. 9 Hence the conditional variance σ2t, is represented by a FIGARCH (Fractionally Integrated AutoRegressive Conditional Heteroskedastic) models as developed by Baillie, Bollerslev and Mikkelsen (1996). The above model is the FIGARCH(1,δ,1) process and generates the type of very slow decay which are frequently observed in the autocorrelations of squared returns, absolute returns and the power transformations of returns; see Granger (1996). (9) Ding, Granger and Engle (1993) and Ding and The general FIGARCH(p,δ,q) process is given by, [1 - β(L)]σ2t = ω + [1 - β(L) - ϕ(L)(1 - L)δ]y2t, where β(L) = β1L + .... + βpLp, ϕ(L) = 1 - α(L) - β(L), and α(L) = q 1L + .....+ qL ; while # denotes the long memory, or fractional parameter and is defined for 0 < δ < 1. There are some further important restrictions on the parameters; namely that ω > 0 and that all the roots of [1 - β(L)] and ϕ(L) must lie outside the unit circle. By straightforward algebra, the process can also be expanded as the infinite order ARCH model, 2 2 (10) σ t = ω/[1 - β(1)] + λ(L)y t, where (11) λ(L) = {1 - [1 - β(L)]-1ϕ(L)(1 - L)δ}. The key property of the above FIGARCH model, which distinguishes it from alternatives is that it again implies very slow hyperbolic rate of decay on the impulse response weights λk kd-1 in equations (10) and (11), which is essentially the "long memory" property, or "Hurst effect". Many well known ARCH models are special cases of the FIGARCH representation. For example, when d = 0, p = q = 1, then equation (8) reduces to the GARCH(1,1) model. When d = p = q 10 = 1, in equation (8) realizes the Integrated GARCH, or IGARCH(1,1) model, and implies complete persistence of the conditional variance to a shock in squared returns. The attraction of the flexibility of the FIGARCH process is that intermediate ranges of persistence can be introduced by having d in the range, 0 < d < 1. In many practical situations quite low order models are adequate, such as the FIGARCH(1,d,1) model. An even simpler model to be applied later is the FIGARCH(1,d,0) process, (1 - βL)σ2t = ω + [1 - βL - (1 - L)d]y2t. (12) For this model the impulse response weights in (6) are σ2t = ω/(1 β) + 2 λ(L)y t, and it can be shown that λk = Γ(k+d- 1)/{Γ(k)Γ(d)}[(1-β)-(1-d)/k], where Γ(.) is the gamma function. For large lag k, d-1 λk = [(1-β)/Γ(d)]k , which generates the same slow hyperbolic rate of decay on the impulse response weights of the conditional variance σ2t. (6) In this study the estimation of the system of equations in through (8) is facilitated by means of Quasi Maximum Likelihood Estimation (QMLE); see Bollerslev and Wooldridge (1988). The procedure uses conventional non-linear procedures to maximize the Gaussian log likelihood function, (13) log(θ) = -(T/2)log(2Π) - (1/2)Σt=1,T[log(σ2t + ε2tσ-2t], with respect to a specified vector of parameters, θ. It should be noted that the numerical procedures are quite general and can be readily extended to models such as the regression model with ARMA disturbances and FIGARCH volatility process. Since most return series are not well described by the conditional normal density in (14); subsequent inference using robust standard errors, is based upon noting that the limiting distribution of the QMLE are given 11 by, (15) where T1/2(θT - θ0) < N{0, A(θ0)-1B(θ0)A(θ0)-1}, A(.) and B(.) represent the Hessian and outer product gradient respectively; and θ0 denotes the true parameter values. Results Some tests of uncovered interest rate parity are obtained by regressing the spot rate returns on the interest rate differential and also the same menu of predetermined variables as in equation (6). Table 1 presents the results estimating the model when four intervention variables (buying and selling of dollars vis-%-vis the DM by both countries) are in the conditional mean and when a GARCH(1,1) formulation is employed for the conditional variance. Results for equation (6) are the same with a FIGARCH specification or when day-of-the-week dummies are included in the conditional mean as is discussed later. In all cases the estimate of the coefficient associated with the interest rate differential was negative in accord with the average value of -0.88 found in 75 separate studies by Froot and Thaler (1990). However, with the overnight Euro deposit rate data, the .95 percentile confidence intervals around the estimated value of b were sufficient to include the value of b = 1. Since the main focus in this study is that of the risk premium the value b = 1 is maintained throughout so that the dependent variable is the ex post return over uncovered interest rate parity, yt = (∆st+1 - it* + it). Since the sampling interval of one day exactly matches the maturity time of the forward contract, the yt series appears uncorrelated. Consequently, model, table 2 reports estimates of the following (6 ) [st+1-(it* -it)]= µ + γ1UStb + γ2USts + γ3Gtb + γ4Gts + Σj=1,4λjDjt + 12 εt, (7) εt = (8) σ2t = ω + βσ2t-1 + (1 - βL - (1 - ϕL)(1 - L)δ)ε2t, ξtσt, where ξt is the i.i.d.(0,1) process, σ2t is the conditional variance process, Djt denote day of the week dummy variables. Baillie and Bollersev (1989), Hsieh (1989), and McFarland, et al. (1982) have discussed the possibility of day-of-the-week effects in the conditional mean and conditional volatility of daily exchange rate returns. UStb denotes the Federal Reserve Bank buying dollars, USts denotes the Federal Reserve Bank selling dollars, and Gtb and Gts denotes corresponding actions by the German Bundesbank. The introduction of intervention in this manner is consistent with viewing intervention as providing a signal of policy. U.S. purchases of dollars, by this reasoning, should signal that the authorities have information, presumably about policy, which once known, would boost the DM/$. This applies to the impact of Bundesbank dollar purchase as well. Thus we would expect estimates of both γ1 and γ3 to be positive and for estimated γ2 and γ4 to be negative. Table presents results from estimating the system (6 ), (7), and (8). Diagnostic tests indicate the success of the FIGARCH(1,d,1) specification in modeling the conditional variance of the deviations from UIRP and in offering a significant improvement over alternatives such as GARCH. None of the daily dummies (Monday through Thursday) were significant in the conditional mean. Of the four intervention variables, only German buying has a significant impact with the opposite sign from that implied by the signaling hypothesis. This is consistent with German dollar purchases reducing DM/$. The second column of Table 2 indicates the results for a subperiod that spans just after the October 1987 crash through the 13 end of 1989. The coefficient on German purchases continues to be negative but U.S. buying has a positive impact. Conclusion This paper has presented investigation into the impact preliminary results of U.S. and German of an dollar intervention vis-%-vis the DM on the deviation from uncovered interest rate parity (UIRP). We view intervention as possibly signaling future policy so that dollar purchases should increase the deviation from UIRP. The approach adopted differs from previous work by utilizing overnight Eurocurrency rate data matched exactly to exchange rates. A FIGARCH formulation of the conditional variance and thus the standard errors is implemented with QMLE. We find that German dollar purchases decreased the deviation from UIRP, a result consistent with our previous finding (Baillie and Osterberg [1997a]) that it decreased the DM/$ during August 6, 1985 through March 1, 1990. This result holds up in an examination of October 20, 1987 through 1989 subperiod during which, however, we find a positive impact of U.S. dollar purchases. This latter result is consistent with our analysis of the risk in the forward market (Baillie and Osterberg[1997b]). The same conclusion results when we account for possible differences between coordinated and unilateral interventions. In general, these results are negative for the signalling hypothesis. However, additional results have confirmed similar findings for analyzing the impact of cumulative intervention. Further work might take account of market conditions and policy intentions during specific subperiods. For example, intervention at times might be intended to !lean against the wind , or to reduce volatility. 14 Table 1: Estimation of the model: ∆st+1 = µ + b(it* - it) + γ1UStb + γ2USts + γ3Gtb + γ4Gts + εt, εt = ξtσt, σ2t = ω + αε2t-1 + βσ2t-1 Parameters µ -0.256(0.203) b -1.813(2.210) γ1 0.009(0.007) γ2 -0.003(0.007) γ3 -0.011(0.005) γ4 0.005(0.005) ω 1.728(0.626) α 0.081(0.015) β 0.893(0.019) m3 m4 Q20 Q220 0.232 4.430 14.88 19.14 ln(j)-5029.86 Key: There are T = 1,463 observations from January 5, 1987 through January 22, 1993. 15 Table 2: Estimation of the model: [∆st+1-(it*-it)] = µ + γ1UStb + γ2USts + γ3Gtb + γ4Gts + Σj=1,4λjDjt + εt, εt = ξtσt, σ t = ω + βσ2t-1 + [1 - βL - (1 - ϕL)1 - L)δ]ε2t, 2 Parameters Full sample Oct 20,1987 -Dec 31, 1989. --------------------------------------------------------µ -0.408(0.391) 0.640(0.690) γ1 0.009(0.007) 0.022(0.009) γ2 -0.004(0.007) -0.010(0.005) γ3 -.010(.005) -0.014(0.005) γ4 0.005(0.005) λ1 0.382(0.681) -1.826(.926) λ2 0.080(0.558) -0.935(0.904) λ3 0.287(0.545) 0.863(0.885) λ4 0.323(0.526) -1.647(0.979) δ 0.525(0.144) 0.624(0.290) ω 1.682(0.804) 1.815(1.544) β 0.671(0.103) 0.713(0.176) ϕ 0.182(0.069) 0.042(0.140) m3 m4 Q20 Q202 ln(j) 0.17 4.21 15.43 19.37 0.007(0.004) 0.12 3.74 15.16 14.93 -5028.89 -1783.370 Key: There are T = 1,463 observations for January 5, 1987 through January 22, 1993. In the reduced sample, T = 530. Asymptotic robust standard errors are in parentheses. The Qm statistic is the Ljung-Box test for autocorrelation based on the first m autocorrelations of the standardized residuals. Qm2 is the 16 Ljung-Box test for ARCH effects based on the first m lags of the autocorrelations of the squared standardized residuals. 17 References Almekinders, G J (1995), Foreign exchange intervention: and evidence, Brookfield, Vt.: Elgar. Theory Baillie, R T and T Bollerslev (1989), "The message in daily exchnage rates: a conditonal variance tale", Journal of Business and Economic Statistics, Baillie, R T and W P Osterberg (1997a), Why do central banks intervene?, Journal of International Money and Finance, 16, 909-919. Baillie, R T and W P Osterberg (1997b), Central bank intervention and risk in the forward market, Journal of International Economics, 43, 483-497. Baillie, R.T., T. Bollerslev and H.-O. Mikkelsen, (1996), "Fractionally Integrated Generalized Autoregressive Conditional Heteroskedasticity", Journal of Econometrics , 74, 3-30. Bekaert, G and R J Hodrick (1993), "on biases in the measurement of foreign exchange risk premiums", Journal of International Money and Finance, 12, 115-138. Bollersev, T. P. , and J. M. Wooldridge (1988), $Quasi-Maximum Likelihood Estimation of Dynamic Models with Time/Varying Covariances,# Massachusetts Institute of Technology Department of Economics Working Paper, November. Ding, Z. and C.W.J. Granger, (1996), "Modeling volatility persistence of speculative returns", Journal of Econometrics, 73, 185-215. Ding, Z., C.W.J. Granger and R.F. Engle, (1993), "A long memory property of stock returns and a new model", Journal of Empirical Finance, Vol. 1, pp. 83-106. Dominguez, K (1990), "Market responses to coordinated central bank intervention", Carnegie Rochester Conference Series on Public Policy, 31, 121-164. Dominguez, K (1992), "The informational role of official foreign exchange intervention policy: The signalling hypothesis", in Exchange Rate Efficiency and the Behavior of International Asset Markets, New York: Garland Publishing, 41-80. 18 Dominguez, K (1993), "Does central bank intervention increase the volatility of foreign exchange rates?", NBER working paper 4532. Dominguez, K M, J A Frankel (1993), $Does Foreign-Exchange Intervention Matter? The Portfolio Effect,# American Economic Review 83, pp.1356-69. Domowitz, I and C S Hakkio (1985), "Conditional variance and the risk premium in the foreign exchange market", Journal of International Economics, 19, 47-66. Edison, H J (1993), $The effectiveness of central-bank intervention: A survey of the literature after 1982,# Special Papers in International Economics, 18, Princeton. Froot, K.A. and R.H. Thaler (1990), $Anomalies: Foreign Exchange, Journal of Economic Perspectives, 4, 179-192. Ghosh, A (1992), "Is it signalling? Exchange intervention and the dollar-deutschemark rate", Journal of International Economics, 32, 201-220 Goodhart, C A E and T Hesse (1993), "Central intervention assessed in continuous time", International Money and Finance, 12, 368-389. bank forex Journal of Hodrick, R J (1987), Empirical Evidence on the Efficiency of Forward and Futures Foreign Exchange Markets, Academic Press. Hodrick, R J (1989), "Risk, uncertainty and exchange Journal of Monetary Economics, 23, 433-459. rates", Hsieh, D A (1989), "Modeling heteroskedasticity in daily foreign exchange rates", Journal of Business and Economic Statistics, 7, 307-317. Humpage, O F (1988), "Intervention and the dollar's decline", Federal Reserve Bank of Cleveland Economic Review, quarter 2,2-16. Humpage, O F (1997), $Recent U.S. Intervention: Is Less More?#, Federal Reserve Bank of Cleveland Economic Review, quarter 3, 2-10. Humpage, O F, and W P Osterberg (1992), $Intervention and the Foreign Exchange Risk Premium: An Empirical Investigation of Daily Effects#, Global Finance Journal, 3, 23-50. 19 Kaminsky, G and K K Lewis (1996), "Does foreign exchnage intervention signal future monetary policy?", Journal of Monetary Economics, 37, 285-312. Kaminsky, G and R Peruga (1990), "Can a time varying risk premium explain excess returns in the forward market for foreign exchange?", Journal of International Economics, 28, 47-70. Klein, M W (1992), $Big Effects on Small Interventions: The Informational Role of Intervention in Exchange Rate Policy,# European Economic Review 36(4), 915-934. Klein, M W and E Rosengren (1991), "Foreign exchange intervention as a signal of monetary policy", New England Economic Review, May/June, 39-50. Levine, R (1989), "The pricing of forward contracts", Journal of International Money and Finance, 8, 163-179. Loopesko, B E (1984), $Relationships among Exchange Rates, Intervention, and Interest Rates: An Empirical Investigation#, Journal of International Money and Finance 3, 257-77. McFarland, J W, R R Pettit and S K Sung (1982), "The distribution of foreign exchange price changes: trading day effects and risk measurement", Journal of Finance, 37, 693-715. Obstfeld, M (1989), "The effectiveness of intervention", NBER working paper, 2796. foreign exchange: Osterberg, W P (1989), $Intervention and the Risk Premium in Foreign Exchange Rates#, Federal Reserve Bank of Cleveland Working Paper 8908. Riehl, H and R Rodriguez (1977), Foreign Exchange Markets, New York, McGraw Hill. 20 1993 1992 1992 1992 1992 1992 1991 1991 1991 1991 German overnight rate 1991 1991 1990 1990 1990 1990 1990 1990 1989 1989 1989 1989 1989 1988 1988 1988 1988 1988 1988 2/23/87 1987 1987 1987 1987 1987 1987 Percent DM/$ DM/$ FIGURE 1: U.S. AND GERMAN OVERNIGHT RATES AND DM/$ EXCHANGE RATE 12 2.2 10 2 8 1.8 6 1.6 4 1.4 U.S. overnight rate 2 10/20/87 1.2 1993 1992 1992 1992 1992 1992 1991 1991 1991 1991 1991 1991 1990 1990 1990 1990 1990 1990 1989 1989 1989 1989 1989 1988 1988 1988 1988 1988 1988 2/23/87 1987 1987 1987 1987 1987 1987 Change in exchange rate 50 40 0.05 10 0 0 -0.05 -10 -20 -0.1 -30 -0.15 10/20/87 -40 -0.2 Interest rate differential FIGURE 2: 1000(∆ln dm/$), 1000(ln(1+German rate/36000)-ln(1+U.S. rate/36000)) 0.2 0.15 30 0.1 20 1992 1992 1992 1992 1992 1992 1991 1991 1991 1991 1991 1991 1990 1990 1990 1990 1990 1990 1989 1989 1989 1989 1989 1989 1988 1988 1988 1988 1988 1988 2/23/87 1987 1987 1987 1987 1987 1987 Change in exchange rate 60 900 40 600 20 300 0 0 -20 -300 -40 -600 -60 10/20/87 -900 Millions of U.S. dollars FIGURE 3: 1000(∆ln dm/$), U.S. Intervention vis-à-vis dm