Time series forecasting improves by correcting past errors automatically
Look Back, Correct Forward: Error Feedback Calibration for Time Series Forecasting
Artificial Intelligence
Summary
Time series forecasting models make predictions about the future based on past data, but they often make mistakes as conditions change. The authors propose a method called Error Feedback Calibration (EFC) that uses recent prediction errors to adjust future forecasts without changing the original model itself. This approach looks back at overlapping past forecast mistakes to understand error patterns and then corrects upcoming predictions accordingly. Their experiments show this method improves accuracy across various models and datasets while avoiding excessive corrections.
What this means in practice
- •For energy demand planners: Improve grid load forecasting accuracy by automatically correcting recent forecast biases without retraining forecasting models.
- •For supply chain managers: Reduce forecasting errors in inventory planning by adjusting future demand estimates based on recent observed prediction mistakes.
Authors
Haonan Yang, Yulin He, Zhuo Li
Abstract
Time series forecasting models are increasingly deployed in continuously evolving environments. After a forecast is issued, future observations gradually arrive and reveal how the model has failed. Existing methods mainly treat such observations as supervision for parameter updates. However, it remains unclear whether verified forecasting errors can themselves provide causal feedback for correcting future predictions. To address this, we introduce Error Feedback Calibration (EFC), a fully test-time framework for deterministic post-hoc correction of time series forecasting. EFC constructs a Causal Error Diagonal (CED) from revealed errors of overlapping historical forecasts, providing a fresh view of the frozen forecaster's recent failure behavior. Then Cumulative Error Transport (CET) maps this error state to future residuals using cumulative estimation, while temporal out-of-sample validation controls the correction strength. Temporal Residual Learning (TRL) further predicts only the residual structure left after CET from recent CED trajectories, rather than relearning the entire forecasting error. Extensive experiments across forecasting backbones and datasets show that EFC broadly improves forecasting performance while substantially controlling harmful over correction.