Celebrating the 2025 Computer Science and Economics (CSEC) Senior Prize Winners

Senior Essay: Assessing whether FOMC meeting minutes improve economic forecasts 

I used a large language model, GPT-4o, to judge the sentiment of each set of minutes from regularly scheduled FOMC meetings. I assessed whether the sentiment of these meetings contained information that was not included in the forecasts from the Federal Reserve’s Tealbook and the Survey of Professional Forecasters (SPF). I created regressions of actual data against forecast and sentiment data and used the coefficients to determine whether the sentiment of the FOMC meeting minutes could improve the forecasts. The results showed that the sentiment only affected the Tealbook forecasts. I created additional regressions and I found that the date range and stability of the economy both impact how effective sentiment is at improving the forecasts.

Senior Essay: Framing Effects in Digital Health Insurance Marketplaces: A Computational and Experimental Analysis of Consumer Choice

This thesis examines how digital interface design elements, including premium-sorting algorithms and recommendation labels, influence consumer decision-making in health insurance marketplaces. Using a custom-built experimental platform, the study replicates real-world insurance exchanges to evaluate behavioral responses across different interface framings. The results show that these common design features significantly reduce optimal plan selection, particularly among participants with lower insurance literacy. These findings underscore the broader implications of digital choice architecture for consumer welfare, regulatory policy, and models of bounded rationality.

Senior Essay: Automated Discovery of Tax Code Vulnerabilities: A Hybrid Approach Using LLMs and SMT Solvers

This project explores using large language models and SMT solvers to automatically discover legal loopholes, demonstrated through a case study of the "grain glitch" in Section 199A of the US tax code. By using an LLM to translate a complex statute into symbolic logic that an SMT solver could then analyze, this hybrid approach successfully rediscovered the known tax minimization strategy, validating the potential for automated legal vulnerability detection.

Senior Essay: Bridging the Gap Between Nash Equilibria and Coarse-Correlated Equilibria

In this paper, we investigate a class of equilibria lying between Nash equilibria and coarse-correlated equilibria (CCEs) called rank-k CCEs. To that end, we first explore the region of rank-2 CCEs in Rock Paper Scissors and show that the unique Nash equilibrium occurs precisely at a singular point in a non-convex region. We then explore the relationship between rank-k CCEs and no-regret online learning algorithms and provide evidence to support the conjecture that there is a tight logarithmic upper bound on the number of periods it takes to converge to a CCE. Finally, we present our initial attempt at proving that finding rank-2 CCEs is computationally hard before concluding with a discussion of challenges encountered while constructing such a proof.

Senior Essay: Engine Design and Reinforcement Learning for Cooperative Deck-Builder Games

I developed a light-weight, configuration-driven engine written in D for cooperative deck-building card games. The engine exposes an API that can plug directly into Python RL libraries, enabling large-scale, high-performance simulation of complex games. Using this platform, I trained and benchmarked Proximal Policy Optimization (PPO) agents in a prototype for Harry Potter: Hogwarts Battle, revealing a cooperative strategy plateau that highlights both the engine’s diagnostic power and the genre’s inherent game balancing mechanics.