A joint test of predictability and structural break in predictive regressions

Abstract

This paper explores a joint test of predictability and one-time structural break, both of which are assumed to be absent under the null hypothesis. The test combines IVX estimator with a sup-Wald-type statistic. The limiting distribution of the test statistic is expected to be non-pivotal under (near-)integration. Nevertheless, for univariate cases, the distribution is highly insensitive to the variation of unestimable nuisance parameter. We hence propose to use critical values from the pivotal distribution derived under stationarity for empirical study. Simulation results suggest that this approach delivers satisfactory and robust inference in finite sample. An empirical application to the predictability of US stock returns is provided.