Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Alpha-R1: Alpha Screening with LLM Reasoning via Reinforcement Learning

Zuoyou Jiang, Li Zhao, Rui Sun, Ruohan Sun, Zhongjian Li, Jing Li, Daxin Jiang, Zuo Bai, Cheng Hua

TL;DR

Alpha-R1 tackles non-stationarity and alpha decay in factor investing by introducing a semantics-driven reasoning core trained with GRPO. The model combines long-term market memory with real-time textual and price signals to perform context-aware alpha screening, activating or deactivating factors as regimes shift. Empirical results show Alpha-R1 outperforms traditional baselines and generic LLMs across multiple asset pools, with strong zero-shot generalization to unseen universes. This framework offers interpretable, regime-aware portfolio construction that integrates unstructured information with quantitative backtesting to mitigate structural fragility in dynamic markets.

Abstract

Signal decay and regime shifts pose recurring challenges for data-driven investment strategies in non-stationary markets. Conventional time-series and machine learning approaches, which rely primarily on historical correlations, often struggle to generalize when the economic environment changes. While large language models (LLMs) offer strong capabilities for processing unstructured information, their potential to support quantitative factor screening through explicit economic reasoning remains underexplored. Existing factor-based methods typically reduce alphas to numerical time series, overlooking the semantic rationale that determines when a factor is economically relevant. We propose Alpha-R1, an 8B-parameter reasoning model trained via reinforcement learning for context-aware alpha screening. Alpha-R1 reasons over factor logic and real-time news to evaluate alpha relevance under changing market conditions, selectively activating or deactivating factors based on contextual consistency. Empirical results across multiple asset pools show that Alpha-R1 consistently outperforms benchmark strategies and exhibits improved robustness to alpha decay. The full implementation and resources are available at https://github.com/FinStep-AI/Alpha-R1.

Alpha-R1: Alpha Screening with LLM Reasoning via Reinforcement Learning

TL;DR

Alpha-R1 tackles non-stationarity and alpha decay in factor investing by introducing a semantics-driven reasoning core trained with GRPO. The model combines long-term market memory with real-time textual and price signals to perform context-aware alpha screening, activating or deactivating factors as regimes shift. Empirical results show Alpha-R1 outperforms traditional baselines and generic LLMs across multiple asset pools, with strong zero-shot generalization to unseen universes. This framework offers interpretable, regime-aware portfolio construction that integrates unstructured information with quantitative backtesting to mitigate structural fragility in dynamic markets.

Abstract

Signal decay and regime shifts pose recurring challenges for data-driven investment strategies in non-stationary markets. Conventional time-series and machine learning approaches, which rely primarily on historical correlations, often struggle to generalize when the economic environment changes. While large language models (LLMs) offer strong capabilities for processing unstructured information, their potential to support quantitative factor screening through explicit economic reasoning remains underexplored. Existing factor-based methods typically reduce alphas to numerical time series, overlooking the semantic rationale that determines when a factor is economically relevant. We propose Alpha-R1, an 8B-parameter reasoning model trained via reinforcement learning for context-aware alpha screening. Alpha-R1 reasons over factor logic and real-time news to evaluate alpha relevance under changing market conditions, selectively activating or deactivating factors based on contextual consistency. Empirical results across multiple asset pools show that Alpha-R1 consistently outperforms benchmark strategies and exhibits improved robustness to alpha decay. The full implementation and resources are available at https://github.com/FinStep-AI/Alpha-R1.
Paper Structure (49 sections, 11 equations, 3 figures, 3 tables)

This paper contains 49 sections, 11 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. LLMs in Quantitative Trading
  4. Financial Foundation Models and Adaptation
  5. LLMs for Alpha Factor Generation
  6. LLM-based Quant Agents
  7. Methods for Screening Alpha Factors
  8. Machine Learning in Regularized and Sparsity-Driven Selection
  9. Machine Learning in Tree-Based and Non-Linear Screening
  10. Deep Learning in Factor Compression
  11. Reinforcement Learning for LLMs
  12. Methodology
  13. Data, Memory, and Factor Baselines
  14. Raw Data Abstraction
  15. Iterative Memory Construction
  16. ...and 34 more sections

Figures (3)

  • Figure 1: Alpha-R1 Framework Overview. The pipeline follows a sequential logic: (a) Preparation: Abstracting raw technical indicators and financial news into atomic textual units to construct a global historical memory ($M_{global}$), coupled with systematic factor backtesting; (b) Semantic Construction: Mapping quantitative performance metrics into structured semantic factor profiles ($\alpha_{des}$) and synthesizing dynamic market states ($S_t$); (c) Reasoning & Optimization: Performing context-aware alpha screening via a reasoning core that evaluates $\alpha_{des}$ against $S_t$, with the policy iteratively refined through GRPO.
  • Figure 2:
  • Figure 3: Parameter Sensitivity and Generalization Analysis. The heatmaps illustrate the impact of varying TopN and HoldingDays. Green regions indicate favorable performance (High Sharpe/Return, Low Drawdown), while Red regions indicate poor performance. The broad Green clusters across both asset pools confirm the strategy's robustness.