Factor premia are easy to discover and hard to keep. The references below cover standard benchmarks, model selection in a crowded factor zoo, multiple-testing bias, and open replication — the core reading for anyone running cross-sectional strategies today.
Benchmarks and data
Start with the canonical factor series before testing new signals. Most anomaly papers use these as controls or portfolio sorts.
Kenneth French Data Library
Three-, four-, and five-factor returns plus industry portfolios and international extensions. The standard benchmark for US equity factor research.
Factor zoo and selection
Hundreds of published factors create a selection problem. Feng, Giglio, and Xiu provide a formal test for whether a new factor adds information beyond what is already known.
Taming the Factor Zoo: A Test of New Factors
Model-selection framework for evaluating incremental factor contribution. Run this logic before adding another signal to your stack.
False (and Missed) Discoveries in Financial Economics
Updated multiple-testing thresholds for published finance research. Explains why many factor t-stats look significant by accident.
Replicating Anomalies
Replicates 452 published anomalies with NYSE breakpoints and value-weighted returns. 65% fail a |t| ≥ 1.96 hurdle; 82% fail Harvey-style multiple-testing thresholds. Essential context before treating any new factor as durable alpha. Author PDF
Open replication
Open Source Cross-Sectional Asset Pricing
Large-scale replication of cross-sectional signals with shared code. The closest thing to a ground-truth audit of the factor literature. Project home
Open Asset Pricing Signal Data
Downloadable signal panels and replication scripts. Compare your factor construction against a public baseline.
Market states and mispricing
Whether anomaly alpha is compensation for bad-state risk or slow correction of mispricing matters for hedging and capacity. Global panel evidence favors stronger long–short spreads when markets are weak.
In Good and in Bad Times? The Relation between Anomaly Returns and Market States
133 anomalies across 56 equity markets (1980–2019). Risk-adjusted daily alpha is higher in bad times than good (about 1.7 vs 1.0 bps); ~75% of the bad-time gain comes from the short leg — consistent with arbitrage limits and underreaction, not a classic risk-premium story.
Alpha decay, crowding, and AI adoption
Publication and replication are not the only drains on factor premia. Homogeneous ML adoption can compress signal life and raise systemic correlation when everyone retrains on the same data.
AI-Driven Alpha Decay: Algorithmic Homogenization, Reflexive Signal Erosion, and the Paradox of Intelligent Markets
Theory and calibration on crowding, performative feedback, and Red Queen competition among AI quant funds. Implied mid-frequency signal half-life near 18 months vs multi-year pre-AI norms; 13F-style holdings show rising portfolio similarity and lower return dispersion among adopters. SSRN
Practical checklist
- Regress new signals on standard Fama-French factors before claiming alpha
- Apply Harvey-Liu style significance thresholds, not raw t-stats
- Check whether the signal replicates in open-source data releases
- Monitor post-publication decay and capacity, not just in-sample spreads
See also our backtesting pitfalls guide for overfitting checks before live deployment.