Reproduce “CAMEF: Causal-Augmented Multi-Modality Event-Driven Financial Forecasting by Integrating Time Series Patterns and Salient Macroeconomic Announcements” on OpenResearch

About this paper

Yang Zhang, Wenbo Yang, Jun Wang, Qiang Ma, Jie Xiong

Researchers from SWUFE, Kyoto University, and Kyoto Institute of Technology developed CAMEF, a multi-modal framework for event-driven financial forecasting that integrates textual macroeconomic announcements with high-frequency time-series data. CAMEF explicitly incorporates causal learning using LLM-generated counterfactual events, achieving up to a 62.55% MSE reduction compared to leading baselines and demonstrating particular efficacy for stock market indices.

Key insights

• Explicitly modeling causal relationships through LLM-augmented counterfactual data significantly enhances the robustness and interpretability of financial forecasting models, moving beyond mere correlations.
• Sophisticated multi-modal integration, combining textual and high-frequency time-series data via specialized encoders and a GPT-2 decoder, yields superior predictive performance in event-driven financial forecasting.
• The impact of macroeconomic events on financial markets is not uniform; FOMC Minutes demonstrate paramount predictive importance, while inflation reports (CPI, PPI) exhibit generally weaker predictive power.

Summary by alphaXiv.