Boosting Agnostic Fundamental Analysis: Using Machine Learning to Identify Mispricing in European Stock Markets
23 Pages Posted: 14 Dec 2021 Last revised: 5 Apr 2022
Date Written: December 5, 2021
Abstract
Interested in fundamental analysis and inspired by Bartram and Grinblatt (2018 & 2021), we apply linear regression (LR) and tree-based machine learning (ML) methods to estimate monthly peer-implied fair values of European stocks from 21 accounting variables. Comparing LR and ML models, we document substantial heterogeneity in the importance of predictors as measured by SHAP values. Examining trading strategies based on deviations from fair values, we find ML-strategies earn substantially higher risk-adjusted returns (“alpha”) than their LR-counterparts (48–66 vs. 11–36 bp per month for value-weighted portfolios). Our findings document the importance of allowing for non-linearities and interactions in fundamental analysis, as well as substantial non-naïve market inefficiencies.
Keywords: Fundamental analysis, market efficiency, stock return, machine learning, random forest, gradient boo
JEL Classification: C45, C53, G11, G14, G15
Suggested Citation: Suggested Citation