Universität zu Köln

Simon, Frederik André Heinrich (2024). Essays in Empirical Financial Research. PhD thesis, Universität zu Köln.

PDF DissertationSimon.pdf Download (2MB)

Abstract

This thesis comprises three studies that deal with the application of machine learning methods in empirical financial market research. Two central topics are covered: portfolio optimization and corporate earnings forecasting. In the first study, a machine learning approach for the direct optimization of stock portfolio weights conditional on predictor variables is presented and empirically tested. It is shown that the presented approach leads to significant utility gains for investors compared to alternative less complex linear methods. This improvement is robust across different investor types and portfolio restrictions. Furthermore, the study shows that a higher degree of risk aversion and a stronger degree of restrictions lead to a reduction in the complexity of the machine learning approach. In the second study, company earnings are predicted using machine learning methods based on fundamental data. The central scientific contribution of this study is the interpretation of the machine learning model. One of the key findings is that variables originating from a company’s income statement are particularly important. Further, it is shown that the relationship between fundamentals and future corporate earnings differs for profit and loss companies, respectively, but is somewhat linear in each case. In the third study, the relationship between the accuracy of forecasting models for corporate earnings and an investment strategy frequently used in the literature conditional on earnings is examined in more detail. In contrast to the existing literature, transaction costs are considered explicitly. In addition, a new accuracy measure is introduced that measures systematic distortions of forecasts, such as systematic under- or overestimation of future earnings. Finally, it is shown that transaction costs correlate neither with the standard measure of forecast accuracy nor with the newly introduced measure of systematic distortions.

Item Type:	Thesis (PhD thesis)
Creators:	Creators Email ORCID ORCID Put Code Simon, Frederik André Heinrich frederikahsimon@gmail.com UNSPECIFIED UNSPECIFIED
URN:	urn:nbn:de:hbz:38-743591
Date:	2024
Language:	English
Faculty:	Faculty of Management, Economy and Social Sciences
Divisions:	Faculty of Management, Economics and Social Sciences > Business Administration > Finance > Professorship for Business Administration and Corporate Finance
Subjects:	Management and auxiliary services
Uncontrolled Keywords:	Keywords Language financial machine learning UNSPECIFIED portfolio choice UNSPECIFIED implied cost of capital UNSPECIFIED earnings forecasts UNSPECIFIED
Date of oral exam:	28 October 2024
Referee:	Name Academic Title Hess, Dieter Prof. Dr. Kempf, Alexander Prof. Dr. Nolte, Ingmar Prof. Dr.
Refereed:	Yes
URI:	http://kups.ub.uni-koeln.de/id/eprint/74359