부산대학교 도서관

This paper proposes a prompt-based method utilizing a large language model (LLM) to predict changes in foreign exchange rates based on limit order information. While traditional deep learning models for prediction utilize numerical values as input and output, LLMs use sentences and prompts. To address this, we design prompts that incorporate the numerical values. GPT-2, a widely adopted LLM, is employed and fine-tuned using a training dataset. The effectiveness of our proposed method is demonstrated through empirical analysis using actual time series data.