The working paper Stock Market Anomalies and Machine Learning Across the Globe (a joint project of Vitor Azevedo with Sebastian Müller and Sebastian Kaiser) is on June 20th among the top 10 most downloaded recent papers in the Economics Research Network from SSRN. The paper had 638 downloads from April 21st, 2022, to June 20th, 2022.
In this paper, we examine the out-of-sample performance of 240 stock market anomalies enhanced by 49 machine learning algorithms and over 260 individually trained models across an international data sample of nearly 1.9 billion stock-month-anomaly observations from 1980 to 2019. We demonstrate significant monthly returns of around 1.8-2.2%, while more than 85% of the tested models show superiority over the linearly composed baseline factor benchmark. Apart from a reliable post-publication decline exclusively in the United States, the results are risk-adjusted by allowing for transaction costs up to 300 basis points and avoiding any forward-looking bias with composite predictors based on the tested machine learning approaches. The results of our non-linear models are significant across several classical asset pricing models and uncover market inefficiencies that challenge current international asset pricing theories.