Prior research suggests that machine learning factor models of the cross section of stock returns greatly enhance portfolio performance by: (1) expanding the dataset to include more variables; and, (2) allowing more complex (non-linear) variable interactions. Does this finding hold up in a realistic portfolio management scenario? In their November 2025 paper entitled "What Drives the Performance of Machine Learning Factor Strategies?", Mikheil Esakia and Felix Goltz decompose performance contributions from these two enhancements in scenarios ranging from ideal to realistic. The ideal scenario, found in much machine learning research, ignores portfolio management constraints. The realistic scenario excludes microcaps, removes look-ahead bias for yet-to-be-published factors and accounts for trading frictions. They further look at exclusion of shorting. They estimate trading frictions as half the monthly effective bid-ask spread (daily average of closing quoted spreads). Using daily and monthly data for publicly listed U.S. common stocks and monthly data for 94 firm-level characteristics as available during June 1963 and through December 2021, they find that:
Subscribe to Keep Reading
Get the research edge serious investors rely on.
- 1,200+ research articles
- Monthly strategy signals
- 20+ years of backtested analysis
Cancel anytime