AlphaSharpe: LLM-Driven Discovery of Robust Risk-Adjusted Metrics

• DOI: 10.48550/arxiv.2502.00029
• Publication Date: 2025-01-23
• Authors: Kamer Ali Yüksel; Hassan Sawaf
• Keywords: FOS: Economics and business; FOS: Computer and information sciences; Computer Science - Computation and Language; Artificial Intelligence (cs.AI); Portfolio Management (q-fin.PM); Computer Science - Artificial Intelligence; Risk Management (q-fin.RM); Computer Science - Neural and Evolutionary Computing; Neural and Evolutionary Computing (cs.NE); Computation and Language (cs.CL); Quantitative Finance - Portfolio Management; Quantitative Finance - Risk Management

Financial metrics like the Sharpe ratio are pivotal in evaluating investment performance by balancing risk and return. However, traditional often struggle with robustness generalization, particularly dynamic volatile market conditions. This paper introduces AlphaSharpe, a novel framework leveraging large language models (LLMs) to iteratively evolve optimize financial metrics. AlphaSharpe generates enhanced risk-return that outperform approaches correlation future employing iterative crossover, mutation, evaluation. Key contributions of this work include: (1) an innovative use LLMs for generating refining inspired domain-specific knowledge, (2) scoring mechanism ensure evolved generalize effectively unseen data, (3) empirical demonstration 3x predictive power forecasting. Experimental results on real-world dataset highlight superiority metrics, making them highly relevant portfolio managers decision-makers. not only addresses limitations existing but also showcases potential advancing analytics, paving way informed robust strategies.