How Predictable are Returns Out-of-Sample? Evidence from Value and Momentum
53 Pages Posted: 5 Aug 2014 Last revised: 6 Jun 2016
Date Written: January 14, 2015
This paper examines the performance of variables that have been suggested in the literature as being good predictors of the returns of value and momentum equity investment strategies out-of-sample. Using simple linear regression models with shrunk estimators at monthly and annual frequency, I find that, value is predicted by book-to-market, earnings-price ratio, smooth earning-price ratio (Asness et al. (2000) and Cohen et al. (2003)), book-to-market or dividend yield together with earnings price ratio growth, and by different forecast combinations. The results are not robust to leverage constraints (with the exception of lagged returns at monthly frequency) and do not hold for constant volatility portfolios. Momentum is predicted at monthly and annual frequency by book-to-market, at monthly frequency by lagged-one-month return, and at annual frequency by book-to-market together with earnings-price ratio growth. Results are robust to targeted volatility portfolios (Daniel et al. (2013)), but are not robust to leverage constraints, finer portfolios, or industry controls. All other predictors perform poorly, are unstable between sub-samples, and are not robust to various robustness tests. The results suggest links of value to the real economy and that the variety of predictors proposed by the literature are of little value for predicting momentum returns out-of-sample.
Keywords: predictability, predictive regressions, value, momentum
JEL Classification: G14, G17
Suggested Citation: Suggested Citation