Seer improves forecasting by reasoning about events in time series
SEER: Self-Evolving Event Reasoning and Retrieval for Time Series Forecasting
Machine LearningArtificial IntelligenceComputation and Language
Summary
Time series forecasting tries to predict future values based on past data, but often real-world changes come from events not shown in the numbers alone. The authors introduce SEER, a system that learns from its own mistakes to better find and understand relevant events affecting predictions. It uses two feedback loops: one to improve how it searches for events and another to build knowledge on cause-and-effect over time. SEER also avoids cheating by only using past information for predictions. It outperforms previous methods on multiple challenging forecasting tasks.
What this means in practice
- •For financial analysts: Enhance stock price or economic indicator forecasts by integrating dynamically learned event impacts from external news sources.$Commercial implications: Enables building forecasting tools that adapt to event-driven market changes better than standard methods.
- •For supply chain planners: Improve demand forecasting by capturing the causal effects of external disruptions and news on time series data.
Authors
Mingtian Tan, Palash Goyal, Mihir Parmar, Sarkar Snigdha Sarathi Das, Chun-Liang Li, Nanyun Peng, Thomas Hartvigsen, Jinsung Yoon, Tomas Pfister
Abstract
Real-world time series are frequently driven by exogenous events and structural shifts, rendering conventional forecasting based solely on historical numerical observations insufficient. While language models can retrieve external news, standard retrieval-augmented approaches struggle with high noise, missing signals, and an inability to reason causally about event impacts. We propose SEER (Self-Evolving Event Reasoning and Retrieval), a closed-loop framework that dynamically optimizes event conditioning for time series forecasting. SEER translates prediction errors into two decoupled feedback mechanisms: (i) a reflective retrieval memory that refines subsequent search queries and filters spurious noise, and (ii) a persistent causal knowledge base that distills transferable domain dynamics. SEER enforces strict chronological boundaries across both event retrieval and reflection, preventing look-ahead bias and data leakage. Across six volatile time-series benchmarks, SEER consistently outperforms state-of-the-art time series foundation models and language model baselines.