Conditional Leptokurtosis and Non-Linear Dependence in Exchange Rate Returns

Abstract

This paper analyzes exchange rate returns for five currencies (Danish krone, Dutch guilder, French franc, Swiss franc, and U.S. dollar). As in Caporale and Pittis (1994), and unlike most of the empirical literature on exchange rates including Pesaran and Robinson (1993), we take a parametric approach to modeling exchange rate dynamics based on the Student’s t autoregressive model with dynamic heteroskedasticity (STAR) due to Spanos (1992). This model, which is more general than standard autoregressive conditional heteroskedasticity (ARCH)-type formulations, is first postulated on the basis of the probabilistic features of the data, and then shown to provide a parsimonious and statistically adequate representation of the data. The estimation results indicate that the martingale property does not hold, and that, in the case of the EMS currencies, there was a decrease in conditional volatility during ERM membership, whereas the conditional variance of the non-EMS currencies evolved in a more erratic fashion. In all cases, leptokurtosis was found in the marginal and joint distributions. The existence of “noise” traders and the irregular flow of information (and/or peso-type problems) could explain these findings.