51 Pages Posted: 22 Nov 2014 Last revised: 9 Jul 2020
Date Written: July 7, 2020
Identifying the factors that drive the cross-section of expected returns is challenging for at least three reasons. First, the choice of testing approach (time-series versus cross-sectional) will deliver different sets of factors. Second, varying test portfolio sorts changes the importance of candidate factors. Finally, given the hundreds of factors that have been proposed, test multiplicity must be dealt with. We propose a new method that makes measured progress in addressing these key challenges. We apply our method in a panel regression setting and shed some light on the puzzling empirical result that the market factor drives the bulk of the variance of stock returns, but is often knocked out in cross-sectional tests. In our setup, the market factor is not eliminated. Further, we bypass arbitrary portfolio sorts and instead execute our tests on individual stocks | with no loss in power. Finally, our bootstrap implementation, which allows us to impose the null hypothesis of no cross-sectional explanatory power, naturally controls for the multiple testing problem.
Note: This paper was formerly circulated under the title "How Many Factors?"
Keywords: Factors, Variable selection, Bootstrap, Data mining, Orthogonalization, Multiple testing, Predictive regressions, Fama-MacBeth, GRS, Performance evaluation, Return prediction
JEL Classification: G12, G14, C12, C21, C22, C31, C32
Suggested Citation: Suggested Citation