Investigating simple target-covariate relationships for Chronos-2 and TabPFN-TS

2026-05-12

AI Summary

This study investigates the effectiveness of time series foundation models (TSFMs) in capturing simple relationships between target variables and covariates. Through carefully controlled experiments under a zero-shot setting, the authors compare Chronos-2 and TabPFN-TS in their ability to incorporate predefined covariate information. The results demonstrate that TabPFN-TS significantly outperforms Chronos-2 in modeling such straightforward dependencies, particularly at short forecast horizons. By introducing structured, interpretable relationships, this work provides the first systematic evidence of disparities in covariate modeling capabilities among TSFMs. These findings challenge the prevailing reliance on standard benchmark datasets alone for model evaluation and offer new insights into the interpretability and practical applicability boundaries of time series foundation models.

Abstract

Time Series Foundation Models (TSFMs) have recently achieved state-of-the-art performance, often outperforming supervised models in zero-shot settings. Recent TSFM architectures, such as Chronos-2 and TabPFN-TS, aim to integrate covariates. In this paper, we design controlled experiments based on simple target-covariate relationships to assess this integration capability. Our results show that TabPFN-TS captures these relationships more effectively than Chronos-2, especially for short horizons, suggesting that the strong benchmark performance of Chronos-2 does not automatically translate into optimal modeling of simple covariate-target dependencies.

Problem

• Time Series Foundation Models
• covariate integration
• target-covariate relationships
• Chronos-2
• TabPFN-TS

Innovation

• Time Series Foundation Models
• covariate integration
• controlled experiments
• TabPFN-TS
• Chronos-2