GMM Estimation of a Stochastic Volatility Model: A Monte Carlo Study

We examine the properties of alternative GMM procedures for estimation of the lognormal stochastic autoregressive volatility model through a large scale Monte Carlo study. We demonstrate that there is a fundamental trade-off between the number of moments, or information, included in estimation and the quality, or precision, of the objective function used for estimation. We provide a fairly transparent characterization of the trade-off in the present model. Furthermore, a large sample approximation to the optimal weighting matrix is utilized to explore the impact of the weighting matrix for estimation, specification testing and inference procedures, and to obtain practical efficiency bounds for the given class of GMM estimators. The results provide guidelines for obtaining desirable finite sample properties through the choice of the appropriate estimation design, and although the findings are specific to the model, the conclusions are likely to apply to a wide range of settings characterized by strong conditional heteroskedasticity and correlation among the moments.