StockMem: An Event-Reflection Memory Framework for Stock Forecasting

2025-12-02

AI Summary

Stock price prediction faces challenges including market volatility, noisy financial news, and implicit causal relationships; existing large language model (LLM) approaches suffer from textual ambiguity and insufficient causal reasoning capability due to generic memory architectures. To address these issues, we propose an Event–Reflection dual-layer memory framework: (1) a temporal event repository enables cross-sectional integration and longitudinal tracking for multi-granularity event extraction; (2) a causal experience repository (i.e., reflection knowledge base) explicitly models event–price interaction mechanisms to support interpretable analogical reasoning. Our method integrates event structuring, incremental information extraction, dynamic event–price modeling, and retrieval-augmented inference. Evaluated on real-world financial datasets, the framework significantly outperforms state-of-the-art memory-augmented models in forecasting accuracy while enabling traceable identification of critical event chains influencing prices—thus achieving both performance gains and decision transparency.

Technology Category

• Machine Learning: Causal Learning
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Knowledge Representation and Reasoning: Action, Change, and Causality

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Stock price prediction is challenging due to market volatility and its sensitivity to real-time events. While large language models (LLMs) offer new avenues for text-based forecasting, their application in finance is hindered by noisy news data and the lack of explicit answers in text. General-purpose memory architectures struggle to identify the key drivers of price movements. To address this, we propose StockMem, an event-reflection dual-layer memory framework. It structures news into events and mines them along two dimensions: horizontal consolidation integrates daily events, while longitudinal tracking captures event evolution to extract incremental information reflecting market expectation discrepancies. This builds a temporal event knowledge base. By analyzing event-price dynamics, the framework further forms a reflection knowledge base of causal experiences. For prediction, it retrieves analogous historical scenarios and reasons with current events, incremental data, and past experiences. Experiments show StockMem outperforms existing memory architectures and provides superior, explainable reasoning by tracing the information chain affecting prices, enhancing decision transparency in financial forecasting.

Problem

• Predicts stock prices using event-based memory
• Extracts incremental information from evolving news events
• Enhances explainability in financial forecasting decisions

Innovation

• Event-reflection dual-layer memory framework
• Horizontal consolidation and longitudinal tracking of events
• Retrieves historical scenarios for explainable reasoning