Timbre improves time series forecasting using contextual reports
TIMBRE: Teaching Time Series Forecasters to Read, Remember, and Reconcile
Machine Learning
Summary
Forecasting future numbers, like demand or weather, can be helped by understanding related reports and past reactions. The authors propose TIMBRE, a method that tries to better combine this information to improve predictions. In tests on 13 tasks, it reduced errors in some cases but not consistently better than a strong existing method. The key benefit comes from how TIMBRE learns to transfer past responses, but more testing is needed to pinpoint what parts help most.
What this means in practice
- •For supply chain planners: Incorporate event reports and past responses to improve demand forecasting accuracy in logistics operations.
- •For energy grid operators: Enhance numerical forecasts of energy usage by integrating event-based information with historical data.
Authors
Xinyu Guan, Zhirong Zhang, Hongyuan Liu, Pengcheng Xu, Yu Sun, Chen Song, Qianyang Zhao
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.