Large Language Model-Guided DeepFutures for Accurate Domestic Futures Trading

Qingyi Pan, Pei Yang, Zeyang Liu, Yalin Hu

Abstract: Domestic futures trading has garnered significant attention from both industry and academia, where the inherent non-stationary and highly volatile properties of price movements have imposed new challenges. In this paper, we propose the Large Language Model (LLM)-guided DeepFutures to detect anomalous trading patterns. Specifically, DeepFutures integrates TimeMixer, Long Short-Term Memory (LSTM), and Auto-Regressive (AR) components within a Limit Order Book (LOB) setting to effectively capture both short-term fluctuations and long-term trends. Extensive experiments from qualitative and quantitative perspectives demonstrate its superior predictive performance over various competitive baselines on several real-world domestic futures datasets. To the best of our knowledge, this is the first attempt to leverage LLM's domain knowledge for guiding deep models in domestic futures trading. Our code is included in the Supplementary Materials for examination and will be released upon acceptance.

External IDs:dblp:conf/icic/PanYLH25