Are Time-Indexed Foundation Models the Future of Time Series Imputation?
This work addresses the problem of missing value recovery in time series data for real-world applications, offering a zero-shot approach that is incremental but broadens the scope of foundation models in this domain.
The paper conducted the first large-scale empirical study of time-indexed foundation models for zero-shot time series imputation, evaluating them on 33 out-of-domain datasets (approximately 1.3M imputation windows) and showing they are a powerful and practical step toward general-purpose imputation without retraining.
Foundation models for time series imputation remain largely unexplored. Recently, two such models, TabPFN-TS and MoTM, have emerged. These models share a common philosophy that places them within the family of time-indexed foundation models. This paper presents the first large-scale empirical study of these models for zero-shot imputation, which enables missing value recovery without retraining across a wide range of scenarios. We conduct extensive univariate experiments across 33 out-of-domain datasets (approximately 1.3M imputation windows) and evaluate their ability to integrate covariates at inference time to improve accuracy without fine-tuning. Our results demonstrate that time-indexed foundation models are a powerful and practical step toward achieving general-purpose, zero-shot imputation for real-world time series.