Learning to Generate Explainable Stock Predictions using Self-Reflective Large Language Models

Paper: arXiv 2402.03659

Abstract

Explaining stock predictions is generally a difficult task for traditional non-generative deep learning models, where explanations are limited to visualizing the attention weights on important texts. Today, Large Language Models (LLMs) present a solution to this problem, given their known capabilities to generate human-readable explanations for their decision-making process. However, the task of stock prediction remains challenging for LLMs, as it requires the ability to weigh the varying impacts of chaotic social texts on stock prices. The problem gets progressively harder with the introduction of the explanation component, which requires LLMs to explain verbally why certain factors are more important than the others. On the other hand, to fine-tune LLMs for such a task, one would need expert-annotated samples of explanation for every stock movement in the training set, which is expensive and impractical to scale. To tackle these issues, we propose our Summarize-Explain-Predict (SEP) framework, which utilizes a self-reflective agent and Proximal Policy Optimization (PPO) to let a LLM teach itself how to generate explainable stock predictions in a fully autonomous manner. The reflective agent learns how to explain past stock movements through self-reasoning, while the PPO trainer trains the model to generate the most likely explanations from input texts. The training samples for the PPO trainer are also the responses generated during the reflective process, which eliminates the need for human annotators. Using our SEP framework, we fine-tune a LLM that can outperform both traditional deep-learning and LLM methods in prediction accuracy and Matthews correlation coefficient for the stock classification task. To justify the generalization capability of our framework, we further test it on the portfolio construction task, and demonstrate its effectiveness through various portfolio metrics.

Complexity vs Empirical Score

• Math Complexity: 3.0/10
• Empirical Rigor: 8.0/10
• Quadrant: Street Traders — practical and empirical, lighter on theory

Why this score: The paper’s primary innovation lies in a novel RL/agent-based framework (SEP) rather than dense mathematical theory, focusing on LLM application and training dynamics. It is highly empirical, featuring a specific dataset, real-world portfolio construction task, backtest-ready metrics like Sharpe Ratio and MCC, and a public GitHub repository for code.

Research Flowchart

  flowchart TD
  A["Research Goal: Generate Explainable Stock Predictions<br>without expensive human annotation"] --> B["Key Methodology: SEP Framework<br>Summarize-Explain-Predict"]
  
  B --> C["Data Input: Chaotic Social Texts<br>+ Historical Stock Prices"]
  
  C --> D["Process 1: Self-Reflective Agent<br>Generates reasoning samples via self-reasoning"]
  C --> E["Process 2: Proximal Policy Optimization PPO<br>Trains LLM on self-generated explanations"]
  
  D --> F["Computational Core: Fine-tuned LLM"]
  E --> F
  
  F --> G["Outcome: High-Performance Explainable Predictor<br>Outperforms traditional models & LLMs in accuracy"]
  F --> H["Application: Portfolio Construction<br>Validated via effective portfolio metrics"]

Cited References (99)

1. Efficient capital markets: A review of theory and empirical work (1970)
2. Using news articles to predict stock price movements (2001)
3. Hybrid learning to rank for financial event ranking (2021)
4. Deep learning for event-driven stock prediction (2015)
5. Diffusion Variational Autoencoder for Tackling Stochasticity in Multi-Step Regression Stock Price Prediction — → 2309.00073
6. Human-Centric Justification of Machine Learning Predictions. (2017)
7. Listening to chaotic whispers: A deep learning framework for news-oriented stock trend prediction (2018)
8. Stock movement prediction from tweets and historical prices (2018)

6 of 99 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.