Real Exchange Rate Movements

One aim of this work is to investigate the causes of fluctuations in the mark/dollar, pound/dollar, and yen/dollar real exchange rates. These real exchange rates have had been fluctuating to a large extent since the beginning of the period of flexible exchange rates in 1973 (see figure 1). Generally, it is assumed that these movements had bad effects on employment, inflation, allocation and growth in the industrialized countries and that they initiated protectionist measures¹. Therefore several proposals have been made to reduce these fluctuations.

In this paper, a two-country version of the Dornbusch (1976) open macroeconomics model with a free-floating exchange rate will be estimated. The model stems from Buiter (1988, p. 122) and is not reported here. Buiter did not derive an estimable equation for his two-country model. To get an estimable equation it was proceeded as follows: Buiter (1988, p. 124) derived the following equation for the long-run equilibrium real exchange rate:

In this section, a test of an autoregressive unit root by Elliott, Rothenberg, and Stock (1992) will be performed with respect to the variables in equation (4) from section B. The tests of an autoregressive unit root are necessary in order to decide whether a cointegration analysis is useful. If all variables were I(0) processes, the effort to specify an error correction model would not be necessary and “classical” methods could be applied. The tests of unit roots will also show which of the real exchange rates are I(1) or I(0) processes. The time series properties of real exchange rates are important to know for foreign exchange rate modelling. To model real exchange rates correctly their stochastic properties should be known. The results of these tests will also decide which error correction model should be used. For example, if several variables turn out to be trend-stationary, an error correction model (ECM) should be applied which allows for this kind of variables or more precisely which allows for stochastic cointegration^l.

To do the cointegration analysis the number of lagged differences k - 1 = k₁ in the ECM

To judge the quality of the overshooting model from section B. the forecasting results of this model were compared with the out-of-sample forecasting results of the naive random walk model. Since the exchange rate is probably driven by “news” and different causes for excessive exchange rate movements, one cannot expect to get good forecasting results for structural exchange rate models. Despite these theoretical considerations, the out-ofsample forecasting results of the structural models were considered disappointing because the structural models did not perform better than the simple random walk model. Especially Meese and Rogoff (1983ab, 1988) showed this for different versions of the monetary model. This paper tries to improve upon these results by estimating the modified monetary model from section B. with Johansen’s ML procedure.

Gonzalo’s and Granger’s decomposition of a cointegrated vector X_t into a permanent component \( {A_1}\alpha _ \bot ^/{X_t} \) and transitory component \( ({A_2}{\beta ^/}{X_t}) \) is as follows:

One aim of this work was to explain the causes of the fluctuations in the mark/dollar, pound/dollar, and yen/dollar real exchange rate and to draw some policy conclusions from them. For every ECM, the fundamentals of the two-country overshooting model had at least some transitory influence on the corresponding real exchange rates. Changes in some fundamentals led also to changes of real exchange rates in the long-run. Moreover, the overshooting model also seems to have some power to explain movements of the three real exchange rates because the forecasting performance of the overshooting model in error correction form was satisfactory for the three exchange rate models. Since “news” with respect to fundamentals had little explanatory power to explain the movements of the considered real exchange rates, a large part of the transitory components had been due to causes of covariance-stationary excessive real exchange rate movements. Because of the stationarity of the transitory component nonstationary excessive real exchange movements can be excluded from the transitory components. These are real exchange rate movements caused by announced changes of stochastic processes of fundamentals (process switching or regime shifts), rational speculative bubbles, and learning processes. The weak evidence with respect to the last two causes of real exchange movements coincides with results of other empirical studies in the literature.