BAYESIAN SEMIPARAMETRIC ARCH MODELS

A Bayesian semi-nonparametric approach to ARCH models is developed with the advantage that small sample results are obtained even when the likelihood function is subject to nonlinear inequality constraints (as in the ARCH models used in this paper). The semi-nonparametric nature of the approach allows for the relaxation of the assumption of normal errors. An application and a small Monte Carlo study indicate that the methods we advocate are both feasible and necessary.