Why More Could be Better for Evaluating Models
For decades, researchers have tried to crack the code of summarizing a cross-section of stock returns with only a few characteristics-based factor models. But new research from Maryland Smith explains why that quest is ultimately futile.
Previous research on the topic gleaned stock characteristics that could help predict cross-sectional variation in anticipated returns. Researchers believed that by locking in on a small number of characteristics, they could find a linear stochastic discount factor (SDF) representation.
However, with the emergence of new models, it may be time to go back to the drawing board, says Serhiy Kozak, assistant professor at the University of Maryland’s Robert H. Smith School of Business.
The research, which was produced with Stefan Nagel of the University of Chicago and Shrihari Santosh of the University of Colorado, was published last year in the Journal of Financial Economics.
It argues, Kozak says, that for the SDF to perform well, a larger number of characteristic-based factors must be included. Across previous literature, he says, there’s inconsistency among return predictors to be able to consider it an answer to adequately pricing the cross-section.
There remains a possibility to achieve small SDF approximation, though, says Kozak. Instead of using raw characteristics-sorted portfolio returns, using components of characteristics based returns may help toward achieving out-of-sample performance.
It’s an approach that results in using all factors when creating an optimal SDF, Kozak says. But it also may be a helpful strategy for future research on interpreting the SDF.
The mean-variance efficient portfolio resulting from the estimated SDF within the research may also prove useful in evaluating other models regarding the cross-section of equity returns.
Read more: “Shrinking the Cross-Section” is published in the Journal of Financial Economics.