Post-Publication Decay of Published Anomalies

A well-documented pattern across decades of academic finance: when researchers compare an anomaly's returns before publication to its returns after publication, the after-publication returns are reliably weaker — one influential study estimated an average decline of roughly 30–50% in an anomaly's out-of-sample profitability once it appears in print. This isn't a fluke of one or two famous cases; it shows up across a wide range of published return predictors.

Two explanations compete, and both are probably partly true. The first is statistical: even honest, careful research is drawn from a noisy process, and a published result was, almost by construction, one of the stronger draws from that noise — some of its original strength was luck that won't repeat. The second is behavioral and structural: once a pattern is public, traders and funds start exploiting it, and that crowding pushes prices toward eliminating the very mispricing the anomaly relied on — the market becoming more efficient exactly where a inefficiency was just pointed out. Distinguishing the two matters for how much decay to expect: a purely statistical artifact should decay once, immediately, and then behave normally, while genuine crowding-driven decay can continue gradually as more capital arrives.

The practical implication for anyone building a strategy off a published anomaly is to discount the in-paper Sharpe ratio meaningfully before deploying capital, and to check the anomaly's performance specifically in the years since publication rather than trusting the full-sample number the paper reports, since that full sample is dominated by the stronger pre-publication years.

The size of the decline also varies with how well-known and how easy to trade the anomaly is. A pattern that requires exotic data or heavy leverage to exploit decays more slowly, because fewer participants are actually positioned to arbitrage it away, while a simple, liquid signal that any desk can implement with standard data tends to decay faster once it's public. That difference is itself useful information when deciding how much of a published finding's original edge is likely to still be available.