Universität zu Köln

Weibels, Sebastian (2026). Three Essays on Machine Learning in Empirical Finance. PhD thesis, Universität zu Köln.

PDF Dissertation_Weibels_KUPS.pdf Download (4MB)

Abstract

This thesis consists of three studies that examine how machine learning methods can be applied in empirical financial market research. The studies address three central topics, namely portfolio optimization, the forecasting of corporate earnings, and the role of information-processing frictions in asset pricing. The first study develops and empirically evaluates a machine learning approach that directly optimizes stock portfolio weights conditional on firm characteristics. The results indicate that this approach generates substantial utility gains for investors relative to less complex linear alternatives, and that these gains are robust across different investor types and portfolio restrictions. The study further introduces the concept of economic regularization and demonstrates that the effective complexity of the machine learning approach declines as risk aversion or the degree of restrictions increases. The second study forecasts corporate earnings using machine learning models trained on a comprehensive set of financial statement variables. Its central contribution lies in the interpretation of the underlying model. Among the main findings is that income statement variables play a particularly important role for short-term forecasts, while balance sheet information gains relevance as the forecast horizon extends. The analysis also reveals that the relationship between fundamentals and future earnings is largely linear, with interactions and nonlinearities contributing a small but growing share for longer horizons. The third study introduces a new data-driven measure of firm atypicality based on an autoencoder and investigates its relation to the cross-section of expected stock returns. The findings show that atypical firms earn significantly lower subsequent returns, and that this premium reflects mispricing rather than compensation for risk, being concentrated among firms with high limits to arbitrage and low investor attention.

Item Type:	Thesis (PhD thesis)
Creators:	Creators Email ORCID ORCID Put Code Weibels, Sebastian weibelssebastian@gmail.com UNSPECIFIED UNSPECIFIED
URN:	urn:nbn:de:hbz:38-803021
Date:	2026
Language:	English
Faculty:	Faculty of Management, Economy and Social Sciences
Divisions:	Faculty of Management, Economics and Social Sciences > Economics > Econometrics and Statistics > Professur in Data Analytics
Subjects:	Management and auxiliary services
Uncontrolled Keywords:	Keywords Language Machine Learning English Portfolio Optimization English Earnings Forecasting English Asset Pricing English Information-Processing Frictions English
Date of oral exam:	27 April 2026
Referee:	Name Academic Title Zimmermann, Tom Prof. Dr. Hess, Dieter Prof. Dr.
Refereed:	Yes
URI:	http://kups.ub.uni-koeln.de/id/eprint/80302