EVOQUANT Automates Robust Quantitative Trading Strategy Optimization.

Quantitative trading strategy optimization typically involves extensive manual effort from domain experts, who must identify weak signals, fine-tune risk controls, and repeatedly validate iterative revisions. While large language models (LLMs) offer potential for acceleration, directly using them to rewrite strategies often leads to issues like hallucinations, strategy drift, and overfitting during backtesting. To overcome these challenges, researchers developed EVOQUANT, a self-evolving, verifier-guided framework for optimizing quantitative trading strategies. EVOQUANT leverages LLMs to deeply analyze performance bottlenecks and generate semantically controlled candidate edits. These edits then pass through a multi-stage verification pipeline to select the best strategy. Crucially, the framework distills optimization experience into reusable knowledge, enabling continuous self-improvement. Evaluations across seven diverse strategies in both A-share and Crypto markets demonstrated significant improvements, with the average test Sharpe ratio increasing from -0.298 to 0.538, and the best strategy showing a 199% relative improvement. This work shifts quantitative strategy optimization from a costly manual process to an automated, verifiable paradigm.