Nonparametric tests for economic models

点击翻译按钮获取中文摘要

This report investigated three topics. The first paper compares conventional GMM estimator to Empirical Likelihood(EL) and Exponential Tilting estimators when the number of moment conditions increases with the number of observations. The main findings of the experiments show the following. Small sample biases of EL and ET are considerably smaller than GMM if the number of moment conditions is less than 10% of the number of observations. When the number of the moment conditions exceed 10% of the number of observations, small sample biases of EL and ET increase almost linearly as the number of moment conditions increase and the slopes are greater than the slope of GMM. The small sample bias of bias corrected GMM (Newey and Smith 2001) is always the same or smaller than the biases of other estimators. The standard deviations of GMM and bias corrected GMM estimators are decreased as the number of moment conditions is increased. The standard deviations of EL and ET estimators, however, are increased when the number of moment conditions exceeds 10% of the number of observations.The second paper investigated Granger causality. We developed root-n consisitent nonparametric Granger causality test.Third paper analyzed a paradoxical phenomenon of semiparametric model that some semiparametric estimators are more efficient when infinite dimensional nuisance parameters are unknown. This paper examined the structure of the paradox. The necessary and sufficient condition of the paradox is presented and a simple sufficient condition is derived. In addition, two examples of semiparametric estimators are examined.