Can ChatGPT Forecast Stock Price Movements? Return Predictability and Large Language Models (2304.07619v5)

Abstract: We document the capability of LLMs like ChatGPT to predict stock price movements using news headlines, even without direct financial training. ChatGPT scores significantly predict out-of-sample daily stock returns, subsuming traditional methods, and predictability is stronger among smaller stocks and following negative news. To explain these findings, we develop a theoretical model incorporating information capacity constraints, underreaction, limits-to-arbitrage, and LLMs. The model generates several key predictions, which we empirically test: (i) it establishes a critical threshold in AI capabilities necessary for profitable predictions, (ii) it demonstrates that only advanced LLMs can effectively interpret complex information, and (iii) it predicts that widespread LLM adoption can enhance market efficiency. Our results suggest that sophisticated return forecasting is an emerging capability of AI systems and that these technologies can alter information diffusion and decision-making processes in financial markets. Finally, we introduce an interpretability framework to evaluate LLMs' reasoning, contributing to AI transparency and economic decision-making.

• The paper demonstrates that ChatGPT’s sentiment analysis produces scores highly correlated with daily stock returns, outperforming traditional methods.
• The methodology uses news headline analysis to compute a unique ChatGPT score, offering a rigorous quantitative framework for market predictions.
• The findings imply that integrating ChatGPT into trading strategies can enhance predictive accuracy and financial decision-making in the markets.

Can ChatGPT Forecast Stock Price Movements? Return Predictability and LLMs

The paper by Alejandro Lopez-Lira and Yuehua Tang from the University of Florida investigates the predictive capabilities of LLMs with a specific focus on ChatGPT in the domain of financial markets. This paper addresses the potential of ChatGPT in forecasting stock price movements through sentiment analysis of news headlines. By assigning sentiment scores to headlines, the researchers aim to analyze the correlation between these scores and daily returns, exploring the viability of ChatGPT in enhancing quantitative trading strategies.

The preliminary findings from this research reveal that ChatGPT demonstrates superior accuracy in predicting stock market returns compared to existing sentiment analysis methods like those from RavenPack. Their methodology involves computing a "ChatGPT score" based on the sentiment analysis of news headlines related to firms' stock prices. The results strongly correlate these scores with everyday stock market returns, indicating that ChatGPT can provide more precise predictions and improve the performance of trading strategies.

A significant outcome of the paper is the comparative analysis of ChatGPT against earlier versions of LLMs such as GPT-1, GPT-2, and BERT, highlighting ChatGPT's enhanced capability in return predictability. Notably, it was observed that when the effect of ChatGPT sentiment scores is controlled, the impact of traditional sentiment analysis methods on predicting daily returns is nullified. This showcases the emerging capacity of more complex models like ChatGPT in the domain of financial forecasting, suggesting an evolution in the capabilities of AI models.

The proposed research design includes several key steps: collecting a comprehensive dataset of relevant news headlines, employing ChatGPT for sentiment analysis using specific prompts, computing sentiment scores, quantitatively analyzing the relationship between these scores and stock returns, and understanding the factors enabling more accurate predictions through textual analysis. Such a rigorous approach is aimed at establishing the robustness and generalizability of results across various sources and timeframes.

The implications of this research are multifaceted. Practically, it can significantly enhance financial technologies, particularly in improving the accuracy of quantitative trading strategies. For policymakers, regulators, and institutional investors, the paper provides insights into the advantages and challenges posed by the increasing integration of LLMs in financial marketplaces. Theoretically, it enriches the academic discourse concerning AI applications in finance, setting the stage for further investigation into developing advanced LLMs tailored for financial decision-making.

Overall, the research posits a significant contribution to the understanding and implementation of LLMs in finance. The exploration of ChatGPT as a tool for stock market prediction paves the way for more efficient decision-making processes, potentially influencing employment and productivity within the financial sector. As AI continues to integrate into various industries, such studies remain critical in delineating the boundaries and potentials of advanced computational models in economic contexts. Future investigations could focus on scaling these models and exploring their broader application across different financial instruments and market conditions.