TIMBRE: Teaching Time Series Forecasters to Read, Remember, and Reconcile

2026-10-03

AI Summary

This study addresses the challenge of effectively transforming textual reports and historical responses into numerical forecast revisions in event prediction. To this end, it proposes the TIMBRE framework, which integrates multi-source evidence under a frozen forecasting head by employing source-aware representations, state-conditioned response shifting, and reliability-guided fusion to handle heterogeneous information. Furthermore, an independent reading module is introduced to dynamically adjust prediction interval widths, focusing on learning response shift sensitivity. Empirical evaluations across thirteen tasks demonstrate that TIMBRE outperforms standard fusion baselines in eight cases. Ablation studies confirm the effectiveness of the core components, revealing that disabling response shifting significantly increases task-specific errors.

Abstract

Event-informed forecasting requires translating reports and historical responses into changes to a numerical forecast. We propose TIMBRE (Temporal Integration of Memory-Based Responses and Evidence), which combines source-aware representation, state-conditioned response transfer, and reliability-guided fusion before a frozen forecast head. A separate readout adjusts interval widths while preserving the median. In a single-seed, one-epoch development study of 13 tasks, TIMBRE improves MAE over ordinary fusion on eight tasks but over native Chronos-2 on only two. Disabling response transfer in the trained model reduces BTC and AULF MAE by 47.04% and 6.81%, respectively. These findings identify sensitivity to learned response transfer rather than a general forecasting advantage. Missing development-set scores and the absence of retrained ablations limit attribution to individual evidence mechanisms.

Problem

• time series forecasting
• event-informed forecasting
• numerical forecast adjustment
• text-to-forecast translation

Innovation

• Time Series Forecasting
• Event-informed Forecasting
• Response Transfer
• Reliability-guided Fusion
• TIMBRE