RMSE Reduction for GMM Estimators of Linear Time Series Models

In this paper we analyze GMM estimators for time series models as advocated by Hayashi and Sims, and Hansen and Singleton. It is well known that these estimators achieve efficiency bounds if the number of lagged observations in the instrument set goes to infinity. <br> A new version of the GMM estimator based on kernel weighted moment conditions is proposed. Higher order asymptotic expansions are used to obtain optimal rates of expansions for the number of instruments to minimize the asymptotic MSE of the estimator. <br> Estimates of optimal bandwidth parameters are then used to construct a fully feasible GMM estimator where the number of lagged instruments are endogenously determined by the data. <br> Expressions for the asymptotic bias of kernel weighted GMM estimators are obtained. It is shown that standard GMM procedures have larger asymptotic biases than kernel weighted GMM. A bias correction for the estimator is proposed. It is shown that the bias corrected version achieves a faster rate of convergence of the higher order terms of the MSE than the uncorrected estimator.