Machine learning techniques for cross-sectional equity returns’ prediction

Abstract We compare the performance of linear regression model, which is current standard in science and practice for cross-sectional stock return forecasting, with that machine learning methods, i.e., penalized models, support vector regression, random forests, gradient boosted trees neural networks. Our analysis based on monthly data nearly 12,000 individual stocks from 16 European economies over almost 30 years 1990 to 2019. find prediction returns can be decisively improved through methods. The outperformance (combined) models benchmark model approximately 0.6% (0.7%) per month full cross-section stocks. Furthermore, we no breakdowns, suggests investors do not incur additional risk using methods compared traditional approach. Additionally, superior due substantially higher portfolio turnover. Further analyses suggest generate their added value particularly bear markets when average investor tends lose money. results indicate future research should make more intensive use techniques respect prediction.