Graduation Year

2026

Date of Submission

4-2026

Document Type

Campus Only Senior Thesis

Degree Name

Bachelor of Arts

Department

Economics

Reader 1

Ben Gillen

Rights Information

2026 Blaise J Heher

Abstract

Within this paper, I evaluate whether increasing complexity within expected return models delivers improved net portfolio performance under realistic implementation conditions. This is documented using a ladder of six expected return models increasing in complexity from a naive equally weighted benchmark to static and macro-conditioned factor models and finally to high-dimensional machine learning approaches. The portfolio construction infrastructure is held constant throughout with the intent of isolating forecast complexity as the sole source of variation. The universe used to evaluate these portfolios is made up of U.S. mid-to-large cap equities from 2005 to 2024. The primary evaluation metric is out-of-sample Sharpe ratios net of transaction costs. The paper’s headline result is that complexity does not pay uniformly. The non- linear machine learning approach, Random Forest, offers performance in-line with 1/N on a net basis while Elastic Net cannot match 1/N on even a gross basis. Both models see significant portions of gross returns eroded by elevated turnover. The relationship between complexity and net performance is non-monotonic. There are meaningful improvements at each factor model specification, but the implementation costs associated with high-dimensional models can offset the statistical improvements achieved. Overall, the net-of-transaction-cost results suggest that the effective dimension of the portfolio optimization problem is lower than what would be concluded from gross performance evidence alone. This finding is meaningful to asset management practitioners attempting to employ high-dimensional modeling strategies.

Download

This thesis is restricted to the Claremont Colleges current faculty, students, and staff.

Share

COinS