Cross-channel importance varies by forecasting method and horizon
Rethinking Cross-Channel Importance in Time-Series Forecasting
Machine LearningInformation Retrieval
Summary
Forecasting multiple time series often relies on understanding how different data channels relate and contribute to predictions. The authors found that channels statistically related to the target can differ from those actually useful or used by forecasting models, especially depending on how far ahead predictions are made. By adapting which channels to focus on for each forecast horizon, some models improved significantly, while others showed little benefit. This means that which data sources matter depends strongly on the forecasting method and the specific prediction question.
What this means in practice
- •For financial analysts: Improve short- and long-term market predictions by selecting relevant financial indicators tailored to forecast horizon.
- •For energy demand planners: Enhance load forecasting accuracy by adapting input signals based on their horizon-dependent usefulness to existing forecasting tools.
Authors
Yong-Hoon Choi, Kwang-Hyun Park, Youngjin Cho
Abstract
Cross-channel modeling is central to multivariate time-series forecasting, yet channels that are statistically related, predictively useful, and actually used by a trained forecaster are often treated as if they defined the same notion of importance. We show that they need not coincide. Cross-channel dependency structures change substantially across future offsets, and horizon-adaptive source selection improves a controlled Ridge predictor in 21 of 32 dataset--prediction-length conditions, with a mean gain of $5.16\%$. This selected-set signal also transfers to a matched nonlinear predictor. Yet imposing the same horizon-specific source logic on iTransformer yields only 11 of 20 wins and a mean gain of $0.208\%$, with little alignment between controlled and neural gains. Functional interventions further show that strong forecasters use cross-channel information, while their source-reliance rankings agree little with controlled utility or with one another across iTransformer, TimesNet, and a cross-channel TimeMixer. As a constructive consequence, bounded post-hoc support improves a frozen channel-independent forecaster in 12 of 16 dataset--horizon conditions, with a positive aggregate bootstrap interval. Cross-channel importance should therefore be interpreted relative to the forecasting mechanism and question that define it: related $\neq$ useful $\neq$ used.