Guido Baltussen, Bart Van Vliet and Pim Van Vliet at Robeco, in collaboration with Erasmus University of Rotterdam, have published a new research paper, titled: The Cross-Section of Stock Returns before 1926 (and beyond).
This new paper studies a cross section of stock returns using a novel constructed database of U.S. stocks covering 61 years of additional and independent data. The database contains data on stock prices, dividends and hand-collected market capitalizations for 1,488 major stocks between 1866-1926. Results over this ‘pre-CRSP’ era reveal a flat relation between market beta and returns, an insignificant size premium, and significant momentum, value and low-risk premiums that are of similar size over the post-1926 period. Overall, stock characteristics can explain over 25% of variation in stock returns. Further, recent machine learning methods are successful in predicting cross-sectional returns out-of-sample. These results show strong out-of-sample robustness of traditional factor models and novel machine learning methods.
Highlights:
- For this large project, the Robeco Quant team, in collaboration with Erasmus University of Rotterdam, built up a hand-collected database of American stocks between 1866 and 1926
- These stocks have been used to study factor premiums in an independent sample, in order to gain more insight into their expected performance. In addition, machine learning methods have been applied
- The findings are clear: value, momentum and low-risk are very robust and attractive factors, as shown in the figures below
Research abstract:
We study the cross-section of stock returns using a novel constructed database of U.S. stocks covering 61 years of additional and independent data. Our database contains data on stock prices, dividends and hand-collected market capitalizations for 1,488 major stocks between 1866-1926. Results over this ‘pre-CRSP’ era reveal a flat relation between market beta and returns, an insignificant size premium, and significant momentum, value and low-risk premiums that are of similar size as over the post-1926 period. Overall, stock characteristics can explain over 25% of variation in stock returns. Further, recent machine learning methods are successful in predicting cross-sectional returns out-of-sample. These results show strong out-of-sample robustness of traditional factor models and novel machine learning methods.
