Factor Exploitability Uncertainty Due to Study Design Choices

Do the choices researchers make when constructing factor portfolios to explain stock returns materially affect their findings? In their June 2022 paper entitled “Mind Your Sorts”, Amar Soebhag, Bart van Vliet and Patrick Verwijmeren examine the extent to which such construction choices affect factor performance. They consider 11 distinct long-short factors as used in widely cited factor models of stock returns. They test 256 ways to assess each factor based on eight portfolio construction choices:

1. Use 70/30 or 80/20 percentile breakpoints to define high and low factor values.
2. Use NYSE or NYSE-AMEX-NASDAQ universes to calculate these breakpoints.
3. Include or exclude microcaps (smaller than the 20th percentile of NYSE market capitalizations).
4. Include or exclude financial firms.
5. Neutralize industry effects, or not.
6. Use value or equal weighting of stocks within portfolios.
7. Sort multiple variables independent or sequentially (nested).
8. Use most recent monthly market capitalizations or only annual June values.

They focus on the standard deviation of Sharpe ratios generated across portfolio construction choices as a measure of non-standard errors, for comparison with the standard error generated by factor return variability (randomness) within each choice. They estimate portfolio turnovers and trading frictions across choices to evaluate net Sharpe ratios. Using data required to calculate factor values and factor portfolio returns for a broad sample of U.S. common stocks spanning January 1972 through December 2021, they find that: (more…)