NONPARAMETRIC-TESTS FOR THE INDEPENDENCE OF REGRESSORS AND DISTURBANCES AS SPECIFICATION TESTS

We adapt techniques from the literature on chaos and nonlinear dynamic s to detect misspecification in models of serially independent data by checking for dependence between the regressors and disturbances. Our tests are nonparametric in that they determine whether the distributio n of the disturbances depends on the regressors without identifying a model of dependence or the distribution of the disturbances. In Monte Carlo simulations we find that these tests have good power against dep endence caused by omitted variables, incorrect functional form, hetero skedasticity, and similar problems. We also apply our tests to detect misspecification in models of income imputation.