A Unifying View of Some Nonparametric Predictability Tests
35 Pages Posted: 14 Sep 2006
Date Written: September 6, 2006
We show that many existing tests for time-series predictability are special cases of a general nonparametric test based on the OLS estimator of the slope coefficient in a bivariate linear regression of certain type. By manipulating the features of this regression one can construct numerous new predictability tests. It turns out that some of the tests existing in the literature are asymptotically equivalent, and differ only by what kind of pivotization is applied to the core statistic. In addition, we show that the same tests may be constructed via reverse regressions. We also provide an extension to multiple null hypotheses and, respectively, tests based on multiple regressions. Among other things, we pay special attention to the issue of correct pivotization, discuss interpretation of regression-based tests, and argue against some widespread misconceptions.
Keywords: Testing, time series, mean predictability, sign predictability, market timing
JEL Classification: C12, C22, C32, C53
Suggested Citation: Suggested Citation