This paper introduces a novel framework that combines ChatGPT's graph inference capabilities with Graph Neural Networks (GNN) to enhance stock movement prediction. The research addresses an unexplored frontier by leveraging ChatGPT to extract dynamic network structures from temporal textual data, specifically financial news, and integrating these networks into GNNs for predictive tasks. This approach aims to improve the accuracy and robustness of financial forecasting models by incorporating evolving relational information derived from textual sources.

The experimental results demonstrate that the proposed model consistently outperforms state-of-the-art Deep Learning benchmarks in stock movement forecasting. Furthermore, portfolios constructed based on the model's outputs show superior performance, including higher annualized cumulative returns, reduced volatility, and lower maximum drawdown. These findings underscore the potential of ChatGPT for text-based network inferences and highlight its promising implications for the financial sector, suggesting a new direction for leveraging advanced NLP techniques in quantitative finance.