Tests for m-dependence Based on Sample Splitting Methods

This paper develops new inference methods for m-dependent data. Our approach is based on sample splitting by regular sampling of original data at lower frequencies, so that standard techniques can be used for independent data in individual subsamples. We then explore several alternatives of aggregation across subsample statistics and investigate their asymptotic and finite sample properties. We apply our methods to nonparametric tests of the predictability of excess returns in the presence of m-dependence. We also illustrate how our serial dependence tests can provide valid information for identifying particular economic alternatives when testing the expectations hypothesis in foreign exchange markets.