Exchange Rate Models
All the news that fits

It has often been claimed that the best predictor of next period's spot exchange rate is the current spot rate. Researchers have found that such a random walk model outperforms 'structural' models of the exchange rate, in which the exchange rate is explained by other economic variables, even when the structural forecasts are performed using actual (rather than expected) future values of the explanatory variables.

In recent years economic theory has emphasized the distinction between anticipated and unanticipated movements in economic variables. The exchange rate, for example, can be viewed as an asset price which is highly sensitive to new information. In Discussion Paper No. 188, Research Fellow Christian Wolff applies this distinction to structural exchange rate models, emphasizing the relationship between unanticipated movements (or 'innovations') in the exchange rate and innovations in the model's explanatory variables.

Wolff begins with a 'structural' model in which the spot exchange rate depends on the expected change in the exchange rate and on a set of explanatory variables which reflect market fundamentals, such as foreign and domestic money supplies, real income, interest rates and inflation rates. The expected change in the exchange rate is included because domestic and foreign residents may be induced to move into or out of the currency as they expect its value to rise or fall.

This model can be rewritten so that the innovation in the spot rate depends on the innovations in the explanatory variables. Wolff estimates these (unobservable) innovations in the explanatory variables using a system of vector autoregressions (VARS), in which each explanatory variable is regressed on lagged values of itself and the other explanatory variables. The innovations, or 'news', are the residuals from these regressions. Wolff calculates the innovations for the expected change in the exchange rate using the current spot rate as a proxy for the expected future spot rate. This assumes that changes in the spot rate are entirely unpredictable, but Wolff argues that this is borne out by the empirical experience of floating rates.

Wolff then estimates the 'news' model by regressing the spot exchange rate on its own lagged value and the innovations in the explanatory variables, using monthly data from the period 1973- 84. He finds that for the dollar/yen and dollar/sterling exchange rates, forecasts from the structural 'news' model compare favourably with those from a random walk model for forecast horizons of up to six or twelve months, although the structural model performs less well compared to the random walk as the forecast horizon recedes.

Wolff notes that his results may be sensitive to the method used to calculate the innovations. But adding other variables that might be thought to convey news about the US economy did not improve the forecasts. Wolff also notes that estimation of the news model depends on the validity of the empirical measures of news that are employed: information subsequently available to the researcher may not have been contemporaneously observable by the market.

Nevertheless, Wolff concludes from these results that structural models of exchange rate determination have failed to outperform the random walk in previous research because these models were not properly tested in a news framework.


Exchange Rates, Innovations and Forecasting
Christian Wolff


Discussion Paper No. 188, July 1987 (ATE)