Stock forecasting with large language models often treats news as independent signals, but equity-relevant news actually evolves through temporally dependent corporate events. The authors of a new arXiv paper argue that historical information is only useful when event continuity, information availability, and transition reliability are explicitly modeled. Existing LLM-based approaches, they suggest, do not fully account for these dependencies.
The paper introduces RICE-Alpha, a reliability-informed correction framework that leverages event graphs. While the abstract is brief, the title and framing indicate that the method corrects LLM-agent forecasts by building a structured representation of how corporate events unfold over time. This allows the model to weigh which past events remain relevant and how reliably they connect to the present.
The significance lies in addressing a known weakness in LLM-based forecasting: the tendency to ignore the causal and temporal structure behind market-moving news. By modeling event continuity and transition reliability, RICE-Alpha aims to make stock predictions more robust, though the abstract does not yet provide quantitative results.