Efficient GMM estimation of weak AR processes

A simple argument is used to derive the optimal GMM estimator of a finite order autoregressive process whose innovation may be conditionally heteroskedastic.