Non Linear Moving-Average Conditional Heteroskedasticity

Ever since the appearance of the ARCH model (Engle 1982a), an impressive array of variance specifications belonging to the same class of models has emerged. Despite numerous succesful developments, several studies seem to show their performance is not always satisfactory see Boulier (1994). In this paper a new alternative to model conditional heteroskedastic variance is proposed: the Non-Linear Moving Average Conditional Heteroskedasticity: (NLMACH). While NLMACH properties are similar to those of the ARCH-class specifications this new proposal represents a convenient alternative to modeling conditional volatility through a non-linear moving average process. The NLMACH performance is investigated using a Monte Carlo experiment and modeling exchange rate returns. It is found that NLMACH outperforms GARCHs forecasts whereas the application to exchange rates provides mixed evidence.