Do accounting-based stock return anomalies exist in samples that precede and follow those in which researchers discover them? In their November 2016 paper entitled “The History of the Cross Section of Stock Returns”, Juhani Linnainmaa and Michael Roberts examine the robustness of 36 accounting-based stock return anomalies, with initial focus on profitability and investment factors. Anomalies tested consists of six profitability measures, four earnings quality measures, five valuation ratios, 10 growth and investment measures, five financing measures, three distress measures and three composite measures. For each anomaly, they compare pre-discovery, in-sample and post-discovery anomaly average returns, Sharpe ratios, 1-factor (market) and 3-factor (market, size, book-to-market) model alphas and information ratios. Key are previously uncollected pre-1963 data. They assume accounting data are available six months after the end of firm fiscal year and generally employ annual anomaly factor portfolio rebalancing. Using firm accounting data and stock returns for a broad sample of U.S. stocks during 1918 through December 2015, they find that:
- Regarding profitability and investment factor premiums:
- During 1963-2014, the profitability and investment factors generate statistically significant average gross monthly returns 0.30% and 0.25%, respectively. But during 1918-1962, average gross monthly returns are statistically insignificant 0.02% and 0.09%.
- A mean-variance efficient strategy based on post-1963 market, size, value, investment and profitability factor premiums does not beat the market portfolio on a gross basis during 1918-1962.
- Just eight of 36 anomalies tested generate average gross monthly returns that are positive and
statistically significant at the 5% level during 1918-1962. Similarly, only eight and 14 of 36 anomalies exhibit statistically significant gross monthly 1-factor and 3-factor alphas during 1918-1962, despite the fact that this early sample is longer than most discovery samples. - On average across all 36 anomalies, all performance metrics decrease sharply and significantly both before and after the discovery sample. Specifically:
- Average anomaly gross monthly 1-factor model alpha during the discovery sample is 0.32%, compared to 0.13% and 0.14% during pre-discovery and post-discovery samples, respectively.
- Anomalies generally exhibit higher volatilities and higher pairwise correlations during pre-discovery samples than during discovery samples.
- Average anomaly Sharpe ratio and factor model information ratios are lower by 60% to 77% compared to discovery samples.
- Results suggest that researchers should test an asset pricing model out-of-sample or, when not feasible, require that a model explain half of the in-sample alpha.
In summary, evidence indicates that most accounting-based stock return anomalies weaken or disappear (average gross returns decrease, and volatilities and pairwise correlations increase) when tested out-of-sample either before or after the discovery sample.
Cautions regarding findings include:
- Anomaly performance data are gross, not net. Periodic factor portfolio reformation generates trading frictions, and shorting generates other costs. Shorting may not be feasible for some stocks. However, annual portfolio reformation mitigates portfolio turnover concern.
- Changes in market/economic conditions may render older data less applicable than recent data.
Compare findings with those summarized in:
- “Effects of In-sample Bias and Market Adaptation on Stock Anomalies”
- “In-sample vs. Out-of-sample Performance of 888 Trading Strategies”
- “Live Performance of Alternative Beta Products”
- “Why Smart Beta Funds Will Disappoint?”
- “Stock Return Anomalies Just Artifacts of Premium Volatility?”
- “Inherent Inhibitors of Inference in Financial Markets”
