Assessment of Regret-based Decision Making and Portfolio Selection Strategies with Applications to the US Financial Markets

Abstract

This study investigates the effectiveness of regret-based minimization techniques in portfolio construction involving algorithmic trading agents. Focusing on a hypothetical portfolio comprising a simple moving average trader, a momentum trader, and a mean-reverting trader, we employ three regret minimization meta-agents—Exponential Weighting, Fictitious Weighting, and Regret-Based Weighting—to optimize the allocation of net worth. The primary objective is to assess each meta-agent’s regret minimization performance, ensuring it converges to 0 over time. Results demonstrate that Exponential Weighting and Regret-Based Weighting meta-agents exhibit satisfactory performance in terms of regret minimization, while the Fictitious Weighting meta-agent falls short due to its reliance on recent data. The findings emphasize the potential value of incorporating regret minimization techniques, specifically Exponential Weighting and Regret-Based Weighting, in algorithmic trading agent-based portfolio construction for improved performance and regret minimization.

Date of Award	Apr 2023
Original language	American English
Supervisor	KHALED Elbassioni (Supervisor)