Dépendance de court et de long terme des rendements de taux de change

In this paper, we estimate ARFIMA-GARCH models introduced by Baillie et al. (1996) for the four major daily exchange rates against the USD (DEM, FRF, YEN and the GBP). We extend the former contributions by accounting for the observed conditional heteroskedasticity and kurtosis respectively through a GARCH process and a Student-t based maximum likelihood estimation. Our estimations suggest that most of exchange rate returns exhibit a short - and long-memory behavior. We can conclude that the exchange rate dynamics are more complex than implied by a random walk. These results have important economic and statistical implications. First, the returns forecasts should be improved. Second, differencing the data in order to have stationnarity is not statistically appropriate.