Machine Learning and Asset Pricing Models
- Author(s): Porsani, Rafael Amaral
- Advisor(s): Subrahmanyam, Avanidhar;
- Roll, Richard W
- et al.
Even though statistical-learning techniques have become increasingly popular in many scientific areas, few studies in the field of cross-sectional asset pricing have incorporated these in their essence. In the first chapter of this dissertation, we suggest a framework for testing the empirical performance of linear asset-pricing factor models, and for investigating anomalies, which employs an array of such techniques, bringing artificial intelligence and asset-pricing a step closer. The methodology utilized in our work combines a range of supervised learning algorithms with the model testing strategies of Avramov and Chordia (2006) and Brennan et al. (1998).
Chapter 2 presents results generated by applying our framework to multiple asset pricing models. While simple in nature, the estimation procedure we use can have implications for risk management, the study of anomalies, the creation of optimal investment policies, and the general study of expected returns. Some of the concepts explored herein may take an added role in future studies which investigate these subjects, helping to reshape the way we think about asset prices and financial-market anomalies.
The title of this dissertation is given after its first two chapters. Chapter 3 is titled "The Building Blocks of Employment: A Signal Processing Analysis". As argued by Hawking (2016), artificial intelligence and growing automation have decimated jobs in traditional manufacturing, and may engender further job destruction into the middle classes, promoting a widening of wealth inequality in their wake. In this chapter, we contribute to the general study of employment, a theme of critical importance today, by utilizing signal processing techniques to decompose into a myriad of building blocks the employment-to-population time series during the years 1975 to 2000 - a prolonged period where strong job gains were registered and recoveries from recessions were quick. An analysis of the main resulting components is presented. The components of employment produced by our signal-processing modeling approach are made available to researchers interested in better comprehending the employment rate during this period, and forces tied to job gains then.
Chapter 4 is co-authored with Mahyar Kargar, and it is titled "The Evolution of Global Financial Integration: A Multivariate Analysis of Currencies and Equities". In this study, we rely on principal component
regressions and canonical correlation analyses to show that not only currencies became more integrated
with each other from the mid-nineties through the early years of the twenty-first century, but also different assets classes -- currencies and equities -- became more closely associated throughout the same period. Our framework suggests that a common set of latent factors was, during these years, ever more capable of explaining
returns from disparate assets; although such buoyant trend in integration subsequently faced strong headwinds, no longer being present in recent times.