Functioning of the International Monetary System

Jacob Frenkel, and Morris Goldstein

• Publisher: INTERNATIONAL MONETARY FUND
• Year: 1996
• ISBN:
  9781557755544

NONE

Exchange Rate Determination, Alternative Exchange Arrangements, Possible Reform of the System

11 Lessons From Empirical Models of Exchange Rates

PETER ISARD*

This paper finds scope for optimism that the empirical modeling of exchange rates will someday lead to significantly better-than-random explanations. The first part of the paper focuses on the relationships among exchange rates, national price levels, interest rates, and international balances of payments and provides perspectives on some elements of truth about these relationships that are consistent with the past decade of modeling failures. It considers some general lessons and open issues for the design of better models. The lessons emphasize the importance of analyzing exchange rates within complete macroeconomic frameworks and of assuming that expectations are formed in ways consistent with the structural models or with information that can be easily extracted from time series of relevant variables.

Changes in exchange rates have received prominent attention during the 1980s, particularly the extraordinary changes in exchange rates between the U.S. dollar and other major currencies. From its average levels in December 1980 to its peaks in February 1985, the dollar appreciated about 25 percent against the Japanese yen, 75 percent against the German mark, and 125 percent against the British pound.¹ After February 1985 the dollar depreciated sharply; by December 1986 the value of the dollar was nearly 25 percent weaker against the yen than it had been in December 1980, was back down to roughly the same exchange rate that had prevailed against the deutsche mark at the end of 1980, and had reversed about half of its peak appreciation against the pound.

Few economists in 1980, however, predicted, or would have been inclined to predict, that the swings in dollar exchange rates would approach the magnitudes of the changes that actually occurred. To some extent the failure of economists to foresee the behavior of exchange rates reflected their imperfect foresight of the stances of monetary and fiscal policies, and of the consequences of those policies for inflation rates, real activity levels, and other economic conditions. But in addition, statistical tests have now indicated formally that the explanatory power of econometric exchange rate models has been extremely poor. Two widely cited papers by Meese and Rogoff (1983a, 1983b) were the first studies to provide extensive and fairly convincing evidence that existing models of systematic exchange rate behavior could not outperform a naive random-walk model or the forward exchange rate in postsample forecasting tests, even when the forecasts of systematic behavior were based on the ex post realized values of the explanatory variables.² Subsequent studies by Backus (1984), Woo (1985), Boughton (1987), Meese and Rogoff (1985), and Schinasi and Swamy (1986), among others, have extended the forecasting comparisons to portfolio-balance models as well as to other varieties of monetary models,³ and have clarified the relative performances of the models of systematic and random behavior.

Backus (1984) concentrated on the exchange rate between the U.S. and Canadian dollars and tested a number of portfolio-balance models in which the regressors included stocks of outside assets and net foreign asset positions; his results were qualitatively identical to those of Meese and Rogoff in revealing that a random walk typically produced the best postsample predictions. Woo (1985) examined an alternative specification of the monetary model (based on combining the classic long-run money demand specification with a short-run partial adjustment hypothesis), which outperformed the random-walk model in predicting the exchange rate between the U.S. dollar and the deutsche mark during a forecasting period from March 1980 through October 1981. Boughton (1987) extended the comparisons to the portfolio-balance models of Artus (1976, 1981, 1984) and Boughton (1984), which assumed static expectations about the real or price-level-adjusted exchange rate (that is, equality between the expected rate of change in the nominal exchange and the expected inflation differential); in addition, Boughton tested the monetary model of Shafer and Loopesko (1983). Boughton found that each of those models could explain only a small portion of observed month-to-month changes in exchange rates, but that the portfolio-balance models in general performed better than a random-walk model in postsample forecasts of the exchange rates of the dollar against both the mark and the SDR currency basket, although the portfolio-balance models did not in general perform better in predicting the dollar-yen rate. Meese and Rogoff (1985) extended the analysis of their first two papers in a number of directions: to nondollar exchange rates (mark-yen and mark-pound) as well as dollar exchange rates (dollar-mark, dollar-yen, and dollar-pound); to real exchange rates as well as nominal exchange rates; and to November 1980–June 1984 as a postsample (Reagan-regime) forecasting period. They characterized the post-sample forecasting results from their third study as “slightly more favorable” to the models of systematic behavior than the results of their earlier studies. Schinasi and Swamy (1986) re-examined the postsample forecasting performances of the models tested by Meese and Rogoff (1983a, 1983b) and found that some of those models could “substantially” outperform forecasts of a random-walk model under certain specification changes that included relaxing the restriction that the coefficients of the models were fixed over time.

Although these subsequent papers have been somewhat more favorable to the models of systematic behavior, the general conclusion from this line of investigation remains sobering: the existing models of systematic behavior explain little of the observed variances of exchange rates during the 1970s and 1980s.

The lessons that might be inferred from the poor empirical performance of exchange rate models provide the subject of this paper, which is organized in two main sections. Section I first attempts to put into perspective some elements of truth abut the relationships between exchange rates, national price levels, interest rates, and international balances of payments. Section II then discusses a number of more general lessons and open issues that deserve and are now receiving serious attention in attempts to build better empirical models of exchange rate determination. Section III provides some concluding remarks.

I. Elements of Truth

It would be incorrect to infer from the performance of existing empirical models that the behavior of exchange rates is completely random. Economists have perceived for centuries that the behavior of exchange rates is not completely independent of the behavior of national price levels, interest rates, or international balances of payments; indeed, were it not for their moral integrity, many risk-averse economists would probably be willing to take large open positions in exchange markets if they knew they had inside information about forthcoming data or policy actions pertaining to inflation rates, interest rates, or international payments balances. Consistently, there is still a basis for some optimism that the empirical modeling of exchange rates will someday lead to significantly better-than-random ex post explanations and will thus also provide an informative framework for ex ante policy analysis or conditional forecasting. In this context, it is useful to try to put the elements of truth into perspectives that are consistent with the past decade of modeling failures.

Exchange Rates and National Price Levels

The concept of purchasing power parity (PPP) has served as a building block for many exchange rate models during the 1970s and 1980s. The term is usually associated with Cassel (1918), but the PPP notion has been traced as far back as the sixteenth century (See Einzig (1970, pp. 145-46)).

The concept of PPP, which has several variants, is basically a notion that the exchange rate between the currencies of any pair of countries should equilibrate to a ratio of aggregate price indices for the two countries (multiplied by an appropriate constant scale factor), or that the percentage change in the exchange rate should equal the difference between the percentage rates of inflation in the two countries.⁴ As a parity or arbitrage condition that characterizes the equilibrium r...

Table of contents

• Cover Page
• Title Page
• Copyright Page
• Contents
• Introduction
• International Monetary Policy Coordination
• European Monetary System, European Economic and Monetary Union, German Unification
• Exchange Rate Determination, Alternative Exchange Arrangements, Possible Reform of the System
• Footnotes