Date of Award

Spring 5-13-2024

Author's School

McKelvey School of Engineering

Author's Department

Electrical & Systems Engineering

Degree Name

Master of Science (MS)

Degree Type

Thesis

Abstract

Understanding cross-sectional and time series variation of asset returns is fundamental in finance, particularly in asset pricing. This thesis explores the integration of factor theory with machine learning to deepen our comprehension of these dynamics. Characteristics based factor models offer a systematic framework for quantifying an asset's underlying risk-return structure, leveraging time-varying conditional information on model parameters carried by firm-specific characteristics. These models serve as valuable tools for discerning the driving components of an asset's expected excess return. Recent research established a novel methodology for consistent parameter estimation within this framework, only requiring a large cross-section but not a long time span. The procedure is heavily reliant on machine learning. Formal results and analyses have been carried out for neural networks, but the theory can be extended to many forms of projection. The primary objective of this thesis is to conduct a comparative examination of different estimation procedures within characteristics based factor models, elucidating their trade-offs and robustness across diverse settings. Additionally, we aim to demonstrate how the integration of factor theory with machine learning can enhance the interpretability of machine learning prediction outcomes within empirical asset pricing. Through the decomposition of excess return predictions into underlying quantities such as compensation for risk and mispricing, our goal is to identify which of these components are most relevant for ultimate excess return prediction and why. By gaining a deeper understanding of the relevance of these components, we aim to refine the robustness of out-of-sample predictions. This research seeks to make a meaningful contribution to the ongoing discourse in asset pricing by shedding light on the integration of factor theory and machine learning techniques. By providing nuanced insights into asset dynamics and informing the development of more resilient estimation procedures within characteristics based factor models, we aim to advance the understanding and practice of asset pricing methodologies.

Language

English (en)

Chair

Joseph A. O’Sullivan

Committee Members

Andreas Neuhierl, Jinsong Zhang

Share

COinS