Chronos-2: From Univariate to Universal Forecasting

TL;DR

Chronos-2 is presented, a pretrained model capable of handling univariate, multivariate, and covariate-informed forecasting tasks in a zero-shot manner and established as a general-purpose forecasting model that can be used in real-world forecasting pipelines.

Abstract

Pretrained time series models have enabled inference-only forecasting systems that produce accurate predictions without task-specific training. However, existing approaches largely focus on univariate forecasting, limiting their applicability in real-world scenarios where multivariate data and covariates play a crucial role. We present Chronos-2, a pretrained model capable of handling univariate, multivariate, and covariate-informed forecasting tasks in a zero-shot manner. Chronos-2 employs a group attention mechanism that facilitates in-context learning (ICL) through efficient information sharing across multiple time series within a group, which may represent sets of related series, variates of a multivariate series, or targets and covariates in a forecasting task. These general capabilities are achieved through training on synthetic datasets that impose diverse multivariate structures on univariate series. Chronos-2 delivers state-of-the-art performance across three comprehensive benchmarks: fev-bench, GIFT-Eval, and Chronos Benchmark II. On fev-bench, which emphasizes multivariate and covariate-informed forecasting, Chronos-2's universal ICL capabilities lead to substantial improvements over existing models. On tasks involving covariates, it consistently outperforms baselines by a wide margin. Case studies in the energy and retail domains further highlight its practical advantages. The in-context learning capabilities of Chronos-2 establish it as a general-purpose forecasting model that can be used "as is" in real-world forecasting pipelines.

Figures (19)

• Figure 1: The complete Chronos-2 pipeline. Input time series (targets and covariates) are first normalized using a robust scaling scheme, after which time index and mask meta features are added. The resulting sequences are split into non-overlapping patches and mapped to high-dimensional embeddings via a residual network. The core transformer stack operates on these patch embeddings and produces multi-patch quantile outputs corresponding to the masked future patches provided as input. Each transformer block alternates between time and group attention layers: the time attention layer aggregates information across patches within a single time series, while the group attention layer aggregates information across all series within a group at each patch index. A group is a flexible notion of relatedness and may correspond to a single time series, multiple series sharing a source or metadata, variates of a multivariate series, or targets along with associated covariates. The figure illustrates two multivariate time series with one known covariate each, with corresponding groups highlighted in blue and red. This example is for illustration only; Chronos-2 supports arbitrary numbers of targets and optional covariates.
• Figure 2: The pairwise win rates (a) and skill scores (b) of the top-4 pretrained models on fev-bench with 95% confidence intervals (CIs) obtained through bootstrapping. Chronos-2 outperforms the next best models (TiRex and TimesFM) by a statistically significant margin on both metrics. The complete plot and results for other forecasting metrics can be found in \ref{['fig:fev-pairwise-win-sql', 'fig:fev-pairwise-win-wql', 'fig:fev-pairwise-win-mase', 'fig:fev-pairwise-win-wape', 'fig:fev-pairwise-skill-sql', 'fig:fev-pairwise-skill-wql', 'fig:fev-pairwise-skill-mase', 'fig:fev-pairwise-skill-wape']} (Appendix).
• Figure 3: Chronos-2's probabilistic forecasting results in univariate mode and the corresponding improvements from in-context learning (ICL), shown as stacked bars on (a) the univariate subset of fev-bench, (b) GIFT-Eval, and (c) Chronos Benchmark II. For these univariate benchmarks, ICL enables cross-learning, allowing the model to share information across items within a batch and thereby generate more accurate forecasts than univariate inference alone. Results for point forecasting metrics are available in \ref{['fig:univar-icl-improve-point']} (Appendix).
• Figure 4: Chronos-2's probabilistic forecasting results in univariate mode and the corresponding gains from in-context learning (ICL), shown as stacked bars on the multivariate and covariates subsets of fev-bench. On multivariate tasks, ICL provides only modest improvements, though Chronos-2 in univariate mode already surpasses the multivariate Toto-1.0 model. On the covariates subset, however, ICL delivers the largest gains, demonstrating Chronos-2's ability to effectively use covariates. Besides Chronos-2, only TabPFN-TS and COSMIC support covariates, and Chronos-2 outperforms all baselines (including TabPFN-TS and COSMIC) by a wide margin. Results for point forecasting metrics are available in \ref{['fig:multi-icl-improve-point', 'fig:covariates-icl-improve-point']} (Appendix).
• Figure 5: Comparison of Chronos-2 against baselines on tasks which include dynamic covariates from the energy and retail domains. Chronos-2 outperforms all baselines by a wide margin, including TabPFN-TS and TiRex, the strongest baselines on the covariates subset of fev-bench (\ref{['fig:covariates-icl-improve']}). For retail, we consider the domain-appropriate WQL metric. Results for point forecasting metrics are available in \ref{['fig:quant-energy-point', 'fig:quant-retail-point']} (Appendix).