MadEvolve: Evolutionary Optimization of Trading Systems with Large Language Models
Researchers are developing new frameworks to evaluate and improve the use of Large Language Models (LLMs) in quantitative finance. One approach, AlphaForgeBench, reframes LLMs as researchers to generate alpha factors and strategies, addressing the instability and inconsistency issues seen when LLMs act as direct trading agents. Another method proposes generating a portfolio of optimization models using LLMs, leveraging their roles as both generators and evaluators to ensure robustness and provide decision-makers with multiple high-quality candidates. Additionally, an evolutionary optimization framework called MadEvolve has shown success in optimizing trading strategies and alpha generation for tasks like Bitcoin trading. AI
IMPACT New frameworks aim to improve LLM reliability and robustness in financial strategy generation and optimization.