Estimating Stock Market Betas via Machine Learning

Aug 2, 2026

Uploaded by Theodore

Machine learning-based stock market beta estimators outperform established benchmark models in both statistical and economic terms. Using data on the predictability of time-varying market betas for U.S. stocks, the study finds the lowest forecast and hedging errors from machine learning estimators. It shows improved performance for market-neutral anomaly strategies and minimum-variance portfolios. Random forests lead overall, while model complexity varies strongly over time; historical betas, turnover, and size are key predictors.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages37

Reading Time1 h 33 min

AccessFree to download

Introduction

Motivation and objectives

Benchmarking approach

Contributions and empirical setup

JOURNAL OF FINANCIAL AND QUANTITATIVE ANALYSIS © The Author(s), 2024 . Published by Cambridge University Press on behalf of the Michael G. Foster School of Business, University of Washington. This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence ([http://creativecommons.org/licenses/by/4.0](http://creativecommons.org/licenses/by/4.0)), which permits unrestricted re-use, distribution and reproduction, provided the original article is properly cited. doi:10.1017/S0022109024000036 Estimating Stock Market Betas via Machine Learning Wolfgang Drobetz University of Hamburg, Faculty of Business Administration [wolfgang.drobetz@uni-hamburg.de](wolfgang.drobetz@uni-hamburg.de) (corresponding author) Fabian Hollstein Saarland University, School of Human and Business Sciences [fabian.hollstein@uni-saarland.de](fabian.hollstein@uni-saarland.de) Tizian Otto University of Hamburg, Faculty of Business Administration [tizian.otto@uni-hamburg.de](tizian.otto@uni-hamburg.de) Marcel Prokopczuk Leibniz University, Hannover School of Economics and Management [prokopczuk@fcm.uni-hannover.de](prokopczuk@fcm.uni-hannover.de) Abstract Machine learning-based stock market beta estimators outperform established benchmark models both statistically and economically. Analyzing the predictability of time-varying market betas ofU.S. stocks, we document that machine learning-based estimators produce the lowest forecast and hedging errors. They also help to create better market-neutral anomaly strategies and minimum variance portfolios. Among the various techniques, random forests perform the best overall. Model complexity is highly time-varying. Historical stock market betas, turnover, and size are the most important predictors. Compared to linear regressions, allowing for nonlinearity and interactions significantly improves predictive performance. I. Introduction In single-factor asset pricing models, such as the capital asset pricing model (CAPM) introduced by Sharpe (1964), Lintner (1965), and Mossin (1966), the expected return of a stock in equilibrium is determined solely by its sensitivity to market risk. While multifactor models that include additional factors can explain the cross-sectional variation in expected returns somewhat better than the CAPM (see, e.g., Fama and French (2008), Harvey, Liu, and Zhu (2016), for extensive evidence), it explains the time series variation in returns well. Moreover, as shown We thank two anonymous referees, Yakov Amihud, Michael Bauer, Wolfgang Bessler, Axel Cabrol, Mikhail Chernov, Hubert Dichtl, Thierry Foucault (the editor), Kay Giesecke, Lisa Goldberg, Valentin Haddad, Barney Hartman-Glaser, Bernard Herskovic, Bryan Kelly, Serhiy Kozak, Markus Leippold, Martin Lettau, LarsA. Lochstoer, Harald Lohre, Tyler Muir, Andreas Neuhierl, Terrance Odean, Stavros Panageas, Markus Pelger, Tatjana Puhan, Carsten Rother, Boris Vallee, Michael Weber, and Ivo Welch for helpful comments and suggestions. [https://doi.org/10.1017/S0022109024000036 Published online by Cambridge University Press](https://doi.org/10.1017/S0022109024000036 Published online by Cambridge University Press) 2 Journal of Financial and Quantitative Analysis by Graham and Harvey (2001), Jacobs and Shivdasani (2012), and Graham (2022), the CAPM is widely used in the industry. The vast majority of chief financial officers of large U.S. firms rely on a 1-factor market model to estimate their cost of equity capital. For this application, firms typically estimate market betas as the main component and treat the market risk premium as an almost free parameter (Cochrane (2011), Jacobs and Shivdasani (2012)) . Investors, in turn, use the market betas for capital allocation decisions and portfolio risk management (Barber, Huang, and Odean (2016), Berk and van Binsbergen (2016), and Daniel, Mota, Rottke, and Santos (2020)) . However, there are two main problems with using the

FAQs

What is the main goal of estimating time-varying market betas in this study?

The study examines whether machine learning-based models outperform established approaches when estimating time-varying market betas, and explains why they add value.

How do machine learning estimators compare with benchmark beta models?

Machine learning-based estimators produce lower forecast and hedging errors than benchmark models and improve market-neutral anomaly strategies and minimum-variance portfolios.

Which modeling approach and predictors perform best according to the abstract?

Random forests perform best overall. Historical stock market betas, turnover, and size are the most important predictors, and modeling nonlinearity and interactions improves predictive performance versus linear regressions.

Aug 2, 2026

Uploaded by Theodore