A NOTE ON INEQUALITY CONSTRAINTS IN THE GARCH MODEL

We consider the parameter restrictions that need to be imposed to ensure that the conditional variance process of a GARCH(<italic>p</italic>,<italic>q</italic>) model remains nonnegative. Previously, Nelson and Cao (1992, <italic>Journal of Business ’ Economic Statistics</italic> 10, 229–235) provided a set of necessary and sufficient conditions for the aforementioned nonnegativity property for GARCH(<italic>p</italic>,<italic>q</italic>) models with <italic>p</italic> ≤ 2 and derived a sufficient condition for the general case of GARCH(<italic>p</italic>,<italic>q</italic>) models with <italic>p</italic> ≥ 3. In this paper, we show that the sufficient condition of Nelson and Cao (1992) for <italic>p</italic> ≥ 3 actually is also a necessary condition. In addition, we point out the linkage between the absolute monotonicity of the generalized autoregressive conditional heteroskedastic (GARCH) generating function and the nonnegativity of the GARCH kernel, and we use it to provide examples of sufficient conditions for this nonnegativity property to hold.