Abstract:
The asymptotic relative efficiency of efficient method of moments when implemented with a seminonparametric auxiliary model is compared to that of conventional method of moments when implemented with polynomial moment functions. Because the expectations required by these estimators can be computed by simulation, these two methods are commonly used to estimate the parameters of nonlinear latent variables models. The comparison is for the models in the Marron-Wand test suite, a scale mixture of normals, and the second largest order statistic of the lognormal distribution. The latter models are representative of financial market data and auction data, respectively, which are the two most common applications of simulation estimators. Efficient method of moments dominates conventional method of moments over these models.
