Consistent Nonparametric Entropy-Based Testing

Abstract

The Kullback-Leibler information criterion is used as a basis for one-sided testing of nested hypotheses. No distributional form is assumed, so nonparametric density estimation is used to form the test statistic. In order to obtain a normal null limiting distribution, a form of weighting is employed. The test is also shown to be consistent against a class of alternatives. The exposition focusses on testing for serial independence in time series, with a small application to testing the random walk hypothesis for exchange rate series, and tests of some other hypotheses of econometric interest are briefly described.