A Comparative Study on How Data Normalization Affects Zero-Shot Generalization in Time Series Foundation Models
This addresses a critical but overlooked issue in TSFMs for researchers and practitioners, offering a practical solution to enhance generalization across diverse time-series domains, though it is incremental as it builds on existing normalization methods.
The study tackled the problem of input normalization for Time-Series Foundation Models (TSFMs) to improve zero-shot generalization, finding that REVIN normalization reduced zero-shot MASE by 89% compared to an un-normalized baseline and by 44% versus other methods, while maintaining in-domain accuracy at 0.84 MASE.
We investigate input normalization methods for Time-Series Foundation Models (TSFMs). While normalization is well-studied in dataset-specific time-series models, it remains overlooked in TSFMs where generalization is critical. Time-series data, unlike text or images, exhibits significant scale variation across domains and channels, coupled with non-stationarity, can undermine TSFM performance regardless of architectural complexity. Through systematic evaluation across four architecturally diverse TSFMs, we empirically establish REVIN as the most efficient approach, reducing zero-shot MASE by 89\% relative to an un-normalized baseline and by 44\% versus other normalization methods, while matching the best in-domain accuracy (0.84 MASE) without any dataset-level preprocessing -- yielding the highest accuracy-efficiency trade-off. Yet its effect utilization depends on architectural design choices and optimization objective, particularly with respect to training loss scale sensitivity and model type (probabilistic, point-forecast, or LLM-based models).