Abstract
This paper investigates the behavior of autocorrelation in EMS exchange rates during volatile and tranquil periods. Both daily and weekly frequencies are examined. The methodology used is an extension of he Generalized Autoregressive Conditionally Heteroskedastic model (GARCH). The extension of the model allows fust order autocorrelations to be lime dependent. Specifically, the autocorrelation coefficient is an exponential funchon of the conditional variance. The hypothesized density function for exchange rate changes is the Generalized Error Distribution (GED) which can accommodate the leptdrurtosis observed in almost all speculative returns. The results show that the magnitude of fust order autocorrelation is larger during periods of low volatility (i.e., volatility and autocorrelation are inversely related). This inverse relationship is significant in daily changes of EMS exchange rates but not in weekly changes. This finding suggests hat short term prediction of future changes in EMS exchange rates is possible, if the assumption of constant autocorrelation is relaxed. Similar results are reported by LeBaron (1992) for changes in stock prices for the U.S. market.