Synthetic data is essential for training foundation models for time series (FMTS), but most generators assume static correlations, and are typically missing realistic inter-channel dependencies. We introduce DynLMC, a Dynamic Linear Model of Coregionalization, that incorporates time-varying, regime-switching correlations and cross-channel lag structures. Our approach produces synthetic multivariate time series with correlation dynamics that closely resemble real data. Fine-tuning three foundational models on DynLMC-generated data yields consistent zero-shot forecasting improvements across nine benchmarks. Our results demonstrate that modeling dynamic inter-channel correlations enhances FMTS transferability, highlighting the importance of data-centric pretraining.
@article{arxiv.2604.05064,
title = {Dynamic Linear Coregionalization for Realistic Synthetic Multivariate Time Series},
author = {Annita Vapsi and Penghang Liu and Saheed Obitayo and Aakriti and Manoj Cherukumalli and Prathamesh Patil and Amit Varshney and Nicolas Marchesotti and Elizabeth Fons and Vamsi K. Potluru and Manuela Veloso},
journal= {arXiv preprint arXiv:2604.05064},
year = {2026}
}
Comments
ICLR 2026 Workshop on Time Series in the Age of Large Models