Dissertations, Theses, and Capstone Projects

Date of Degree

9-2025

Document Type

Doctoral Dissertation

Degree Name

Doctor of Philosophy

Program

Economics

Advisor

Christos Giannikos

Committee Members

Wim Vijverberg

Sebastiano Manzan

Subject Categories

Econometrics | Economics | Finance

Keywords

Asset pricing, Behavioral finance, Institutional investors

Abstract

This dissertation consists of three chapters that investigate the relationship between asset pricing in a high-dimensional data setting and how investors make portfolio choices in reaction to the choices of other investors.

Chapter 1 introduces the existing work in asset pricing that inspired Chapters 2 and 3. I survey the most relevant literature that has structured the study of return prediction and investor choice, starting from the CAPM market excess return as the ‘first’ and original factor, to the ‘veritable zoo’ of firm-specific variables, which continues to grow as financial data have become more detailed and accessible. The recent use of very large datasets has complemented traditional factor models, opening the door to more complex, high-dimensional tools.

Chapter 2 uses supervised machine learning applications for return prediction. Although research has focused on estimating models and evaluating their predictive performance, the discussion of selection methods (hyperparameter tuning) could be more robust. I estimate machine learning regression models with different hyperparameter tuning methods to predict U.S. stock monthly returns and identify which firm-level characteristics provide higher explanatory power. I find that penalty selection methods for lasso regression (adaptive lasso, plugin estimator, and BIC criterion) make a marginal contribution to predictive performance compared to cross-validation.

Lastly, Chapter 3 challenges the idea that investors observe and incorporate only market fundamentals and financial indicators to forecast returns and allocate assets. I propose a new measure, Commonality, that makes investors comparable based on their holdings, and I use it to estimate the influence that U.S. institutional investors have on each other's trades by using SEC 13F data from 1980-2023. I build an empirical framework that allows for heterogeneity across investors who select portfolios based on the lagged information of their competitors' holdings. The findings show that the overlapping of assets is low and that investors influence each other, but not in the same magnitude. In particular, passive investors exercise influence over active investors, regardless of size.

This work is embargoed and will be available for download on Wednesday, September 15, 2027

Share

COinS