Estimation of risk premia from option data and using machine learning methods: comparison, forecast quality and potential of hybrid strategies

• Applicants: Professor Dr. Joachim Grammig; Professor Dr. Christian Schlag
• Subject Area: Statistics and Econometrics ; Accounting and Finance
• Term: from 2020 to 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 440957921

• Final Report Year: 2025

Final Report Abstract

Empirical finance or financial econometrics is one of the methodologically most developed areas of economics. Nobel Prizes for the prominent scholars, and a high practical relevance testify to the importance of the field. In recent times, new developments have emerged that address the core topic of empirical financial market analysis - the quantification of risk premiums - in various directions. On the one hand, there is the integration of machine learning methods (MML), and on the other hand the deepening of the analysis of option data, not so much from a statistical perspective, but strictly theory-based. The classic methods of financial market econometrics are thus given new perspectives and impetus. At the time our grant application was written, it seemed that the two paths evolved in parallel and without explicit reference to each other, although the same scientific question was being addressed. In our project, an attempt was made to work out the advantages and disadvantages of the methodologically very different approaches. Thanks to the excellent data access at the University of Frankfurt and the SAFE institute, a comprehensive database could be created that facilitated a comprehensive comparison of the alternative methods on common empirical grounds. Another aim of the project was to identify synergies, i.e. whether these very different paradigms could be combined in a meaningful way. On the one hand, it was examined how theoretically based variables quantified with alternative data sources can improve the explanatory power of agnostic MML. On the other hand, priority was given to the theory-based approach and it was examined what support can be provided by MML to increase the quality of the risk premia quantification. These analyses are part of an extensive working paper that has been presented at top international conferences and published in the leading field journal of financial econometrics. A second working paper was produced in the later phase of the project, combining classical financial econometrics with the option-based approach and using MML for the notoriously difficult test of the conditional CAPM (Capital Asset Pricing Model). Both studies show the usefulness of hybrid models that combine theory/optionbased approaches with machine learning methods for empirical asset pricing.

Publications

• Diverging Roads: Theory-Based vs. Machine Learning-Implied Stock Risk Premia. SSRN Electronic Journal.
  Grammig, Joachim; Hanenberg, Constantin; Schlag, Christian & S.önksen, Jantje
  
• Quantifying Equity Risk Premia: Financial Economic Theory and High-Dimensional Statistical Methods. Dissertation an der Universität Tübingen, Fakultät für Wirtschafts- und Sozialwissenschaften
  Hanenberg, Constantin
  
• Essays in Econometrics and Finance. Unveröffentlichte Habilitationsschrift. Fakultät für Wirtschafts- und Sozialwissenschaften an der Universität Tübingen.
  Sönksen, Jantje
  
• Testing the Conditional CAPM using Cross-sectional Regressions: A Multi-task Learning Approach. SSRN Electronic Journal.
  Grammig, Joachim; Hanenberg, Constantin; Schlag, Christian & S.önksen, Jantje