Empirical asset pricing via machine learning: evidence from the European stock market

Wolfgang Drobetz,Tizian Otto

doi:10.1057/s41260-021-00237-x

Wolfgang Drobetz, Tizian Otto

Open Access

https://doi.org/10.1057/s41260-021-00237-x

Copy DOI

Abstract

This paper evaluates the predictive performance of machine learning methods in forecasting European stock returns. Compared to a linear benchmark model, interactions and nonlinear effects help improve the predictive performance. But machine learning models must be adequately trained and tuned to overcome the high dimensionality problem and to avoid overfitting. Across all machine learning methods, the most important predictors are based on price trends and fundamental signals from valuation ratios. However, the models exhibit substantial variation in statistical predictive performance that translate into pronounced differences in economic profitability. The return and risk measures of long-only trading strategies indicate that machine learning models produce sizeable gains relative to our benchmark. Neural networks perform best, also after accounting for transaction costs. A classification-based portfolio formation, utilizing a support vector machine that avoids estimating stock-level expected returns, performs even better than the neural network architecture.

Highlights

This paper studies two complementary topics in the empirical asset pricing literature: stock return prediction and machine learning
We find that the classification-based approach is superior to even the best-performing expected return-based portfolio formation because it avoids some of the noise in stock-level returns
Because we find no improvement in predictive performance relative to the elastic net approach, we do not present the results for these penalty functions here

Summary

Introduction

This paper studies two complementary topics in the empirical asset pricing literature: stock return prediction and machine learning. We exploit a set of twenty-two predictors as per the linear FM regressions approach used in Drobetz et al (2019) This is our benchmark model, and it is able to explain a substantial percentage of the cross-sectional variation in European stock returns. Against this established conservative benchmark, we compare the performance of different machine learning methods in forecasting stock returns, from both a statistical and an economic perspective. The return and risk measures of long-only investment strategies, i.e., sorting stocks into decile portfolios based on expected return estimates and buying the top decile portfolio, indicate that machine learning methods can produce predictive gains These gains are attributable to predictor interactions and to nonlinear effects that is overlooked by the linear benchmark model.

Literature review

Findings

Conclusion