We perform an extensive Monte Carlo study of efficient method of moments (E
MM) estimation of a stochastic volatility model. EMM uses the expectation u
nder the structural model of the score from an auxiliary model as moment co
nditions. We examine the sensitivity to the choice of auxiliary model using
ARCH, GARCH, and EGARCH models for the score as well as nonparametric exte
nsions. EMM efficiency approaches that of maximum likelihood for larger sam
ple sizes. Inference is sensitive to the choice of auxiliary model in small
samples, but robust in larger samples. Specification tests and 't-tests' s
how little size distortion. (C) 1999 Elsevier Science S.A. All rights reser
ved. JEL classtficatiolz: C15; C22.