More than one reader has asked for a ranking of firm characteristics or factors by predictive power for future stock returns, or for a ranking of trading strategies by alpha.
Reasons why it is likely unproductive to attempt to rank firm characteristics or factors according to predictive power for future stock returns, or strategies according to modeled alpha, include:
As described in “A Rather Unsatisfying Morass of Variables”, it is difficult to compare outputs of various available studies because of differences in:
- Basic methodologies (e.g., regression versus portfolio analysis/ranking)
- Factors used as controls/risk-adjusters
- Ways of accounting for errors/uncertainties in variable measurement (e.g., treatment of “outliers”)
- Intervals of historical measurement and future return (see “The Sensitivity of Stock Market Return Predictability to Predictor Measurement Interval”)
- Methods of analyzing time variation in predictive power
Different sample periods may generate different conclusions, as exemplified by the observations in “Why the Story on Predictability Keeps Changing”.
There is considerable inconsistency among studies in assumptions about trading frictions, ranging from ignoring to hand-waving to conservatively estimating.
Many studies do not account for data snooping bias, whether such bias may be direct through consideration of multiple factors and parameter settings or indirect through building upon the snooping bias embedded in prior research.
Many less formal studies incorporate look-ahead bias by choosing parameter thresholds based on information not available in real time.
Financial markets are arguably adaptive (in a very messy way), as described in “Survival of the Richest: The Adaptive Markets Hypothesis” and discussed in “Persistence of Diversity in Investor/Trader Beliefs”, making historical inference unreliable to an uncalibrated degree.
Many of the statistical techniques used in financial markets research are less meaningful (or not meaningful) if return distributions are “wild,” as discussed in “The Black Swan: The Impact of the Highly Improbable (Chapter-by-Chapter Review)” and “Different Paths to the Same (Disconcerting) Destination?”.
See “THE DEMON’S DRUDGERY” for additional discussion on data snooping bias and other limitations of financial markets research.
Hence, “What Works Best?” offers only vague suggestions, with reservations.
