演講摘要 : In this paper, we propose a simple approach that utilizes the sample mean and variance of the standardized residuals to eliminate the effect of estimating the conditional mean and variance models on the moment tests for the standardized error distributions. The idea is different from the seminal Newey-Tauchen and Neyman-Wooldridge approaches. The resulting tests are asymptotically valid in the presence of estimation uncertainty, are robust to all T1/2-consistent estimators of the conditional mean and variance parameters, and are applicable to the case where the distribution being tested is not completely specified, such as testing symmetry. More importantly, the proposed tests are much simpler to implement and are invariant to various correctly specified conditional mean and variance models. As demonstrative examples, we further use this approach to propose a class of skewness-kurtosis-based tests and a class of characteristic-function-based tests for various distribution assumptions. This approach can also be easily applied to other sensible highermoment restrictions. We also assess our approach and compare various moment tests using a Monte Carlo simulation.
