A few weeks ago, we did a deep dive into the factors versus characteristics debate.

One of the reasons we’ve brought up this debate is due to the fact that “factor” loadings (from regressions) are arguably not as helpful as portfolio characteristics. In other words, knowing a portfolio P/E ratio is more informative for forecasting expected returns than knowing the HML factor loading is .6.

This sentiment is reflected in the paper at hand (“The Cross-Section of Expected Stock Returns” by Jonathon Lewellen. The paper can be found here.):

My paper also relates to Fama and French (1997), Simin (2008), and Levi and Welch (2014), who show that the CAPM and Fama-French (1993) three-factor model do not provide reliable estimates of expected returns. My results suggest that forecasts from characteristic-based regressions have better out-of-sample predictive power than either of the asset-pricing models.

The specific question this paper answer is as follows: Are current firm characteristics predictive of future returns?

Turns out the answer is, “yes.” (at least it was in the past)

This paper uses the Fama-MacBeth (FM) two-stage regression to achieve the average slopes on a variety of commonly-used factor investing strategies. From the average slopes, the paper then goes to examine how well these can be used, with the current characteristics of the firm, to predict future stock returns.

From the abstract, the paper finds that “Empirically, the forecasts vary substantially across stocks and have strong predictive power for actual returns.”

So, given the analysis in the paper, characteristics had predictive power in the past.

Below we dig into the paper.

The Factors, Data, and FM Results

Most are familiar with the numerous factors used by managers/advisors/individuals to sort/screen securities. This paper examines 15 such factors, listed in the paper and described below.