Predicting Out-of-Sample Decay in Stock Anomalies

Summary

The paper studies which characteristics of published stock anomalies predict a decline in risk-adjusted performance outside the original sample. It tests explanations associated with arbitrage capital entering a strategy after publication and with in-sample overfitting from multiple hypothesis testing. Publication year alone explains a substantial share of the variation in Sharpe decay: the paper reports that decay for newly published factors rises by five percentage points per year.

Measures associated with overfitting, including how many operations are needed to calculate a signal and how sensitive its in-sample Sharpe ratio is to outliers, add explanatory power. Some variables associated with arbitrage are statistically significant, but contribute little predictive power. The findings offer warning signs for evaluating research claims, rather than a reliable way to forecast the future performance of any individual strategy. The analysis concerns a large set of academic stock anomalies, so its results may not transfer directly to other markets or strategy types.

Key ideas

• Publication year predicts a meaningful share of the observed variation in stock factor Sharpe decay.
• Signal complexity and sensitivity of in-sample Sharpe ratios to outliers are associated with overfitting risk.
• The study considers both post-publication arbitrage and multiple-testing explanations for weaker out-of-sample performance.
• Some arbitrage-related measures are statistically significant but add little predictive power.

Tags

# Why and how systematic strategies decay


# Why and how systematic strategies decay









In this paper, we propose ex-ante characteristics that predict the drop in risk-adjusted performance out-of-sample for a large set of stock anomalies published in finance and accounting academic journals. Our set of predictors is generated by hypotheses of OOS decay put forward by McLean and Pontiff (2016): arbitrage capital flowing into newly published strategies and in-sample overfitting linked to multiple hypothesis testing. The year of publication alone - compatible with both hypotheses - explains 30% of the variance of Sharpe decay across factors: Every year, the Sharpe decay of newly-published factors increases by 5ppt. The other important variables are directly related to overfitting: the number of operations required to calculate the signal and two measures of sensitivity of in-sample Sharpe to outliers together add another 15% of explanatory power. Some arbitrage-related variables are statistically significant, but their predictive power is marginal.

Shown in full with attribution under the source's licence. Licence: abstract CC0

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.