The power of ChatGPT in processing text: Evidence from analysis and prediction in the exchange rate markets

Abstract This study investigates the application of large language models in analyzing sentiment features within the exchange rate markets. Traditional natural language processing methods, such as LDA and BERT, are effective in extracting topics from text; however, they fail to assess the relative importance of these topics in relation to target exchange rates. To bridge this gap, this paper employs ChatGPT to extract topics from texts and evaluate their importance scores, further enhancing exchange rate forecasting by integrating topic importance into the sentiment analysis framework. Through empirical analysis, the superiority of ChatGPT over LDA and BERT in both topic extraction and importance assessment is demonstrated. Furthermore, this study utilizes the topic importance scores generated by ChatGPT to develop a novel interval-valued sentiment index (TIS index). This index not only accounts for the relative importance of various events influencing exchange rate fluctuations but also captures the dynamic evolution of market sentiment within an interval. Empirical results highlight that the TIS Index significantly enhances the forecasting accuracy of interval models such as TARI and IMLP for exchange rates. These findings further demonstrate the advantages of ChatGPT in sentiment analysis within the foreign exchange market. These findings offer new insights into the application of ChatGPT in financial text research.