Pure LLM Event-Driven Forecasting

Background: The Event-Driven Reasoning branch generates predictions by employing an LLM-based pipeline that leverages exogenous text and historical error-informed guidance via Historical In-Context Learning (HIC).

Question / Future Work: Exploring the potential for the Event-Driven Reasoning branch to perform end-to-end time series generation or forecasting without the explicit numerical branch, relying solely on LLM capabilities guided by exogenous text and HIC, could offer a unified generative framework for complex time series behavior driven by external events.