We present a new heteroskedastic conditional variance model using NonLinear Moving Average as the basis for this specification [NLMACH(q)]. The typical problem of this class of models-i.e., noninvertibility—is solved by means of an intuitive parametric restriction; this allows us to use...