REGEN: Reference-Guided Synthetic Multivariate Time Series Generation for Forecasting

2026-06-03

AI Summary

This work addresses the scarcity of real-world data in multivariate time series forecasting and the inability of existing synthetic data generation methods to preserve periodicity, local variability, and dynamic inter-variable dependencies. To overcome these limitations, the authors propose a reference-guided generative framework that decomposes a small set of observed sequences into three components: a periodic backbone, stochastic residuals, and cross-variable causal dependencies. Innovatively treating reference sequences as structural scaffolds rather than imitation targets, the method explicitly controls the shape, stochasticity, and causality of synthetic data through phase-aligned periodic modeling, deep kernel Gaussian processes for residual fitting, and a structural causal model with coupling coefficients to capture lagged dependencies. Experiments demonstrate that models trained on the generated data incur negligible performance loss compared to those trained on real data—and even outperform them in strongly periodic scenarios such as traffic forecasting—while foundation models pretrained with this approach significantly surpass existing synthetic data methods.

Abstract

Training robust multivariate time series forecasting models requires large, diverse corpora, yet many real-world domains provide only a handful of observed sequences. Existing generators fail to resolve this mismatch: prior-based approaches (e.g., CauKer, TimePFN) produce domain-agnostic samples, while data-driven methods (e.g., TimeGAN) treat references as black-box supervision, forfeiting explicit control over periodic structure, local variability, and cross-variable dynamics. We propose ReGeN, a reference-guided generative pipeline that treats observed sequences not as examples to imitate, but as structural scaffolds for controllable synthesis. ReGeN decomposes each reference into three interpretable components: a phase-aligned periodic backbone capturing dominant domain morphology; per-variable stochastic residuals modeled with a deep-kernel Gaussian process; and lag-aware cross-variable dependencies injected through a structural causal model with fitted coupling coefficients. Sampling these components at controllable temperature broadens distributional coverage while preserving domain-grounded structure. We show that ReGeN-generated data consistently substitutes for real sibling data with minimal forecasting degradation, and in strongly periodic domains such as traffic, can outperform the real source itself. We further show that a foundation model pretrained on ReGeN corpora outperforms those pretrained on prior-based and data-driven synthetic alternatives. This suggests that in low-data regimes, how reference data is structurally exploited can matter as much as how much data is available.

Problem

• multivariate time series
• synthetic data generation
• data scarcity
• forecasting
• reference-guided generation

Innovation

• reference-guided generation
• multivariate time series
• structural decomposition
• controllable synthesis
• foundation model pretraining