Tests of Return Predictability: A Comparative Study

Abstract

These days, one can choose from a variety of designated tests for return predictability. Statistical evidence, though, remains inconclusive. This is discomforting given the profound implications this empirical question offers. Our contributions are: (i) we provide a comprehensive review of the tests discussed in the literature; (ii) the test idea presented by Cochrane (2007) remains ambiguous, which is why we propose two variants based on ideas from likelihood estimation; (iii) we give a unified framework to holistically assess the validity of the proposed tests; (iv) the test of Campbell and Yogo (2006) comprises a nested hypothesis that inhibits a straightforward computation of p-values; we therefore propose a constrained optimization approach to overcome this limitation; and (v) in terms of real-world data we confirm the important role of the sample period for the presence of return predictability.