ForecastCompass: Guiding Agentic Forecasting with Adaptive Factor Memory

Yurui Chang, Yongkang Du, Yuanpu Cao, Jinghui Chen, Lu Lin

Keywords: Forecasting, Agent

Abstract: Agentic forecasting is important for decision-making in dynamic environments, yet remains difficult because agents must make calibrated predictions from incomplete, time-limited evidence. Memory can transfer lessons from resolved forecasts to future tasks, but existing agent-memory methods rarely capture reusable predictive factors or calibration knowledge. We propose ForecastCompass (FoCo), an adaptive factor-based memory framework for agentic forecasting. FoCo organizes experience with a hierarchical task taxonomy and maintains two complementary memories: factor memory for reusable predictive dimensions and reasoning memory for probability updating, uncertainty handling, and calibration. Through retrospective memory revision, FoCo accumulates transferable forecasting knowledge over time. Experiments on Prophet Arena and FutureX with GPT-5-mini and Gemini-2.5-Flash show improved accuracy and calibration.

Submission Number: 95