EVERMINE: DISSECTING THE SELF-EVOLUTION OF RESEARCH CAPABILITIES IN LONG-HORIZON ALPHA RESEARCH
By Siyuan Li, Jiangfeng Zhang, Rui Yao, Weihua Qiu, Mingyang Xu, Zixuan Yuan
Rating
1500
Battle Count: 0
Relevance
9/10
Highly relevant as it directly addresses the automation of alpha discovery, a core task in quantitative trading. It evaluates how AI agents can improve factor portfolios over time, which is critical for strategy development.
Implementation Complexity
8/10
High complexity due to the need for a custom agent harness, sandboxed environments, specific LLM deployments, and a sophisticated evaluation loop involving factor pool updates and state replays.
Reproducibility
4/5
The paper provides detailed experimental setups, data sources (Binance Spot), and evaluation protocols. Code is promised upon acceptance. The use of specific model versions and detailed resource budgets aids reproducibility.
About this paper
Methodology: EverMine Framework. Problem types: Alpha Discovery, Portfolio Optimization, Reinforcement Learning, Self-Evolving Agents.
The interactive Everscope explorer (charts, battles, favorites) loads below.