From Linear to Hierarchical: Evolving Tree-structured Thoughts for Efficient Alpha Mining

2025-08-22

AI Summary

Existing Alpha mining approaches suffer from either low computational efficiency (symbolic methods) or failure to preserve hierarchical structure (direct LLM application). To address these limitations, we propose the Tree-Structured Thought Evolution (TSTE) framework, which explicitly models Alpha’s inherent tree-like logical structure as an evolvable reasoning path. TSTE synergistically integrates large language models’ generative capabilities with symbolic regression principles and introduces dedicated evolutionary operators to optimize high-level reasoning architectures—without requiring full code execution. This work is the first to incorporate tree-structured reasoning into Alpha mining, substantially reducing reliance on domain expertise and computational resources. Empirical evaluation across four real-world financial market datasets demonstrates that TSTE generates superior Alpha factors in significantly less time, outperforming conventional linear evolutionary methods in both efficiency and factor performance.

Technology Category

• Machine Learning: Evolutionary Learning
• Search and Optimization: Evolutionary Computation
• Data Mining & Knowledge Management: Representing, Reasoning, and Using Provenance, Trust

Application Category

• Web Mining and Content Analysis: Models for Web evolution
• Economics, Online Markets and Human Computation: Architectures and workflows that use LLMs for crowd work
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Alpha mining, which discovers signals that predict asset returns, has long been attractive for automatic quantitative investment. This problem is typically formulated as a tree-based symbolic regression with handcrafted market data features and arithmetic operators. Unfortunately, existing symbolic methods are concerned with computational inefficiency and dependence on prior knowledge. Recent implementation of Large Language Models (LLMs) show that they can automatically generate executable codes for various tasks efficiently, thus can be considered as a new promising way for alpha mining. Specifically, LLMs-driven methods evolve a set of heuristics, including thoughts and codes, where the thoughts are usually represented as plain-text prompts of codes. Unfortunately, trivially adopting them in alpha mining ignores the fact that alphas are with hierarchical tree structures. This paper introduces Tree-structured thought Evolution (TreEvo), which evolves hierarchical reasoning ideas solely at the thought level. Experiments on four real-market datasets demonstrate that TreEvo can obtain better alphas with much less computational time and human expert efforts. And this superiority hardly holds without the tree-structured thoughts and the compatible evolutionary operators.

Problem

• Evolving hierarchical tree-structured thoughts for alpha mining
• Addressing computational inefficiency in symbolic regression methods
• Reducing dependence on prior knowledge and human effort

Innovation

• Tree-structured thought evolution for hierarchical reasoning
• Evolving hierarchical thoughts without code generation
• Compatible evolutionary operators for tree structures