ChatGPT Informed Graph Neural Network for Stock Movement Prediction (2306.03763v4)

Abstract: ChatGPT has demonstrated remarkable capabilities across various NLP tasks. However, its potential for inferring dynamic network structures from temporal textual data, specifically financial news, remains an unexplored frontier. In this research, we introduce a novel framework that leverages ChatGPT's graph inference capabilities to enhance Graph Neural Networks (GNN). Our framework adeptly extracts evolving network structures from textual data, and incorporates these networks into graph neural networks for subsequent predictive tasks. The experimental results from stock movement forecasting indicate our model has consistently outperformed the state-of-the-art Deep Learning-based benchmarks. Furthermore, the portfolios constructed based on our model's outputs demonstrate higher annualized cumulative returns, alongside reduced volatility and maximum drawdown. This superior performance highlights the potential of ChatGPT for text-based network inferences and underscores its promising implications for the financial sector.

ChatGPT Informed Graph Neural Network for Stock Movement Prediction: A Methodological Overview

The research paper presents a sophisticated framework that harnesses the capabilities of ChatGPT, a LLM, to enhance the performance of Graph Neural Networks (GNNs) in predicting stock movement. This innovative approach is centered on the integration of temporal textual data derived from financial news to construct dynamic network structures. These structures are then used to inform GNNs for predictive financial modeling. The primary focus of this paper is to explore the utility of LLMs in financial economics, specifically in the domain of stock market prediction, which has historically been influenced by the efficient market hypothesis that posits stock prices reflect all available information.

Methodology

The framework introduces a three-tiered approach:

1. Network Structure Inference via ChatGPT: ChatGPT is leveraged to process daily news headlines and identify latent inter-dependencies among equities. The LLM infers which companies are affected by specific news events and the sentiment (positive, negative, or neutral) associated with these events. This capability allows the construction of time-evolving graphs, inputted into GNNs to derive relational embeddings of companies.
2. Company Embedding through GNN: Once the network structure is built, GNNs are deployed to encode the embeddings of the affected companies. The GNN considers both the intrinsic attributes of the nodes (companies) and the topological information from the constructed dynamic graphs, which are informed by ChatGPT.
3. Sequential Models and Output Layers: Integrating LSTM with GNN-derived company embeddings and daily stock market data, the model produces forecasts of stock price movements using a classification approach. The model is trained to predict whether a stock will move up, down, or remain neutral in the near future.

Experimental Results

The performance of the framework was evaluated using a dataset comprising companies listed on the DOW 30 index over a two-year period. The results demonstrate the model's superiority over conventional machine learning approaches such as LSTM and ARIMA and a textual embedding technique using transformer-based models. The proposed framework achieved significant improvements in Weighted F1, Micro F1, and Macro F1 metrics, inferring a more nuanced understanding of market dynamics via its ChatGPT-informed network.

Moreover, the financial performance of portfolios constructed based on the model's predictions was evaluated. The resultant portfolios exhibited higher cumulative returns and reduced risk metrics, including lower annualized volatility and maximum drawdown, as compared to those based solely on traditional models and baseline sentiment analysis.

Implications and Future Directions

This paper underscores the potential of LLMs like ChatGPT in the financial sector, especially for enhancing the prediction capabilities of GNNs by leveraging textual data. The results highlight ChatGPT's effectiveness in achieving nuanced text-based network inferences, which can be crucial given the intricate and interdependent nature of stock markets. Nonetheless, the research points to possible enhancements by incorporating more complex network structures, varied data inputs, and exploring other LLM-based models. The prospect of a broader application of this novel integration across various domains of financial engineering presents numerous opportunities for refining predictive accuracy and decision-making in finance.

In conclusion, the paper demonstrates a methodologically robust framework that successfully combines the strengths of LLMs and GNNs, setting a promising precedent for future research in AI-driven financial prediction and the broader applicability of these computational techniques in dynamic environments.