Separating Stock Universe Filters from Factor Ranking
Summary
The note distinguishes selecting a stock universe from testing whether a factor can rank future returns within that universe. It describes factor analysis as sorting stocks by a factor, dividing them into equal-sized groups, and comparing group returns. Results can change across universes: the author says some financial factors show separation among large-cap stocks but not among small-cap stocks, which they attribute to differences in investor behavior and company characteristics.
The discussion warns that choosing filters based on historical returns can overfit the sample, especially in small-cap strategies that then rank selected stocks by market capitalization. It attributes such strategies’ returns to both the size factor and the selected universe, and argues that the latter may weaken live. Suggested research practice is to prefer intuitive filters, such as index membership or simple financial thresholds, then test factors within the resulting universe. The examples and conclusions are conceptual rather than supported by quantified tests; they do not establish that any particular filter or factor will work out of sample. The author also suggests studying specialized universes with a sound rationale, potentially using higher-frequency data.
Key ideas
- Factor sorting tests whether a factor differentiates future returns within a chosen stock universe.
- Factor performance can differ across universes, as illustrated by financial factors in large-cap and small-cap stocks.
- Historical tuning of universe filters can overfit and may make backtests look stronger than live results.
- Small-cap strategies may combine returns from market-cap ranking with returns from their selected stock universe.
- Research on specialized, logically motivated universes can provide an alternative to broad-market factor studies.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.