Foreign Exchange Rate Forecast by a Large Language Model Integrated with Trend Description

In this paper, we propose a large language model to forecast the direction of change in a foreign exchange rate. The input of the proposed model is textual information as a prompt, whereas that of conventional forecast models is numerical information. A recent trend in the exchange rate is added to input textual information to enhance forecast accuracy. GPT-2 is adopted as our large language model and is fine-tuned using training data. The effectiveness of the proposed model is empirically examined using actual data.