On the performance of nonparametric specification tests in regression models

Some recently developed nonparametric specification tests for regression models are described in a unified way. The common characteristic of these tests is that they are consistent against any alternative hypothesis. The performance of the test statistics is compared by means of Monte Carlo simulations, analysing how heteroskedasticity, number of regressors and bandwidth selection influence the results. The statistics which do not use a bandwidth perform slightly better if the regression model has only one regressor; otherwise, some of the statistics which use a bandwidth behave better if the bandwidth is chosen adequately. These statistics are applied to test the specification of three commonly used Mincer-type wage equations with Uruguayan and Spanish data; all of them are rejected. (C) 2002 Elsevier Science B.V. All rights reserved.