Researcher ORCID Identifier
0009-0003-5326-3541
Graduation Year
2026
Date of Submission
11-2025
Document Type
Campus Only Senior Thesis
Degree Name
Bachelor of Arts
Department
Economics
Reader 1
George Batta, D.B.A.
Terms of Use & License Information
Rights Information
© 2025 Antonis D Pappas
Abstract
This study investigates cross-sectional return predictability in the “private-like” segment of the public equity market, focusing on small-capitalization, high-leverage firms that proxy for private credit investment targets. I employed a comprehensive suite of twelve machine learning and econometric models, including Neural Networks, Gradient-Boosted Trees, and penalized linear regressions. The predictor set combines traditional financial characteristics with specialized textual sentiment scores derived from Loughran-McDonald 10-K summaries, testing the value of qualitative information in opaque environments. The out-of-sample prediction proved highly challenging; complex non-linear models exhibited overfitting. The models successfully identified an asymmetric signal, demonstrating a greater ability to predict underperforming firms than outperforming ones. Exploiting this asymmetry, the Ridge Regression Long-Short portfolio generated the highest economic returns, driven by simple fundamental factors. These findings offer a direct contrast to results from liquid markets, establishing that for high-idiosyncratic risk environments, robustness and parsimony in model choice are overwhelmingly superior to the complexity of deep learning techniques. This result provides critical guidance for asset managers operating in illiquid, opaque markets.
Recommended Citation
Pappas, Antonis, "Predicting Private-Like Equity Returns with Machine Learning and Textual Sentiment" (2026). CMC Senior Theses. 4357.
https://scholarship.claremont.edu/cmc_theses/4357
This thesis is restricted to the Claremont Colleges current faculty, students, and staff.
