Estimating the Anomaly Baserate
46 Pages Posted: 1 Mar 2019
Date Written: February 28, 2019
The academic literature contains literally hundreds of variables that seem to predict the cross-section of expected returns. This so-called ‘anomaly zoo’ has caused many to question whether researchers are using the right tests for statistical significance. But, here’s the thing: even if a researcher is using the right tests, he will still be drawing the wrong conclusions from his analysis if he is starting out with the wrong priors—i.e., if he is starting out with incorrect beliefs about the ex ante probability of discovering a tradable anomaly prior to seeing any test results.
So, what are the right priors to start out with? What is the correct anomaly baserate?
We propose a new statistical approach to answer this question. The key insight is that, under certain conditions, there’s a one-to-one mapping between the ex ante probability of discovering a tradable anomaly and the best-fit tuning parameter in a penalized regression. When we apply our new statistical approach to the cross-section of monthly returns, we find that the anomaly baserate has fluctuated substantially since the start of our sample in May 1973. The ex ante probability of discovering a tradable anomaly was much higher in 2003 than in 1990. As a proof of concept, we construct a trading strategy that invests in previously discovered predictors and show that adjusting this strategy to account for the prevailing anomaly baserate boosts its performance.
Keywords: Return Predictability, Data Mining, Penalized Regression
JEL Classification: C12, C52, G11
Suggested Citation: Suggested Citation