Open Source Asset Pricing (Chen-Zimmermann)
A public, standardized library of hundreds of stock-return predictors, built so researchers stop re-implementing each other's factors slightly differently and calling the discrepancies "findings."
Academic asset-pricing papers each claim to find a new return predictor, but two researchers coding up "the same" accruals anomaly from scratch often get noticeably different numbers, because tiny choices — which Compustat field, how to lag it, how to handle missing values — pile up. Chen and Zimmermann built the Open Source Asset Pricing project to remove that noise: a shared, versioned codebase and dataset that implements around 200 published predictors identically, with the exact construction logic published alongside each one.
The project's headline finding is itself useful: when factors are built consistently, a large share of published "anomalies" turn out to be far weaker after 2004 than in the original sample, and many correlate heavily with each other rather than being independent discoveries.
Chen-Zimmermann gives researchers a common, auditable baseline for known return predictors — so a claim that a new factor is genuinely novel can be checked against roughly 200 existing ones built the same way, rather than against whatever ad hoc replication someone else happened to code up.
Worked example. A researcher claims a new "inventory growth" signal predicts returns with a monthly long-short spread of 0.6%. Pulling the same period's accruals and asset-growth factors from the Chen-Zimmermann library shows a 0.71 correlation with existing signals — meaning the "new" factor's return is largely explained by ones already catalogued, not a fresh source of alpha.
Practice in interviews
Further reading
- Chen & Zimmermann, 'Open Source Cross-Sectional Asset Pricing'