NONNESTED TESTING IN MODELS ESTIMATED VIA GENERALIZED METHOD OF MOMENTS

We analyze the limiting distribution of the Rivers and Vuong (2002, <italic>Econometrics Journal</italic> 5, 1–39) statistic for choosing between two competing dynamic models based on a comparison of generalized method of moments minimands. It is shown that (i) if both models are misspecified then the statistic has a standard normal distribution under the null hypothesis of equal fit but the ranking could be determined by the choice of the weighting matrix; (ii) if both models are correctly specified or locally misspecified then the limiting distribution of the test statistic is nonstandard under the null.