Three Essays on Machine Learning and Nonlinearity in Asset Pricing

Abstract

The complexity and nonlinearity of financial data, particularly asset prices, have been widely documented in the literature. Traditional linear measures often struggle to accurately capture these features. Machine learning methods have emerged as a popular alternative to traditional statistical methods for studying complex and nonlinear financial data. Machine learning models, including robust principal component analysis, random forests, and gradient-boosted machines, can handle large datasets and capture complex interactions between variables without relying on specific distributional assumptions. As such, the use of machine learning techniques in finance has opened new avenues for exploring financial data’s complexities and nonlinearities, leading to more accurate and insightful analysis. Our research interest generally lies in uncovering the complexities and nonlinearities in financial data and its empirical implications for asset pricing. Chapters 2 and 3 focus on the nonlinearities in the dependence structure of financial asset returns, while Chapter 4 shifts the focus to the nonlinearities in the dependence structure of financial and macroeconomic variables with predictive power for future returns. To address each problem, we apply both advanced statistical methods and machine learning models to capture the nonlinearities and measure their asset pricing implications. This study's findings highlight the efficacy of integrating machine learning techniques in financial research, providing a new lens through which to view complex market dynamics. Our results underscore the importance of moving beyond traditional models to embrace more adaptive and sophisticated analytical frameworks to better capture nonlinearity in finance.