What Useful Alphas?

Battle Count: 100

This paper is directly and critically relevant to quantitative trading. It systematically demonstrates that the ~200 published academic anomalies, when implemented in the investable universe of large-cap stocks (top 90% market cap, ~3,000 stocks) after 2005, yield a median return of only 7 bps per month (~1% per year) before transaction costs. After empirical Bayes shrinkage for selection bias, even the best-performing anomaly earns only ~6 bps per month. The paper provides a clear two-by-two framework (period × universe) for evaluating any published anomaly's practical relevance. For quantitative portfolio managers, the conclusion is stark: standard published anomalies are essentially useless in large-cap equities in the 21st century, and public stock markets are very efficient. The paper also identifies the few surviving categories (profitability, financing, and seasonal momentum variants) and quantifies their limited remaining edge.

The methodology is straightforward: standard quintile/decile long-short portfolio sorts from the Chen-Zimmermann dataset, simple market-cap rank and percentage filters, a binary time-period split, and a one-line empirical Bayes shrinkage formula. No complex optimization, machine learning, or iterative estimation is required. The main practical challenge is obtaining and processing the CRSP data and the anomaly dataset, both of which are publicly available. The two-by-two decomposition is purely descriptive.

All data come from the fully open-source Chen and Zimmermann (2022) dataset at openassetpricing.com. The methodology uses standard quintile/decile portfolio sorts, simple market-cap filters, and a transparent empirical Bayes shrinkage formula. The authors explicitly encourage replication and extension. Code and data are publicly available.