Time Series Foundation Models as Strong Baselines in Transportation Forecasting: A Large-Scale Benchmark Analysis

Authors: Javier Pulido, Filipe Rodrigues Date: 2026-02-27 Paper ID: openalex:2602.24238

Summary

This paper investigates the zero-shot effectiveness of the Chronos-2 time-series foundation model as a general-purpose forecaster for transportation dynamics across ten real-world datasets, including traffic volume, speed, and demand. The evaluation protocol rigorously benchmarks Chronos-2 against traditional statistical methods and specialized deep learning architectures without any task-specific training. Results indicate that the foundation model frequently achieves state-of-the-art or competitive accuracy, especially for longer forecasting horizons, confirming its utility as a strong baseline. Furthermore, the model’s inherent probabilistic outputs are shown to offer reliable uncertainty quantification through prediction interval coverage and sharpness metrics. This work strongly suggests that large, pre-trained time-series models can significantly simplify the baseline establishment process in specialized forecasting domains like transportation.

Key Contributions

• Benchmarked the zero-shot performance of the Chronos-2 time-series foundation model across ten diverse transportation forecasting datasets.
• Demonstrated that Chronos-2 achieves state-of-the-art or competitive accuracy compared to specialized deep learning models and statistical baselines without any fine-tuning.
• Showed that Chronos-2’s native probabilistic outputs provide useful uncertainty quantification (coverage and sharpness) for transportation forecasting tasks.
• Advocated for the adoption of general-purpose time-series foundation models as strong, standardized baselines in transportation forecasting research.

Limitations

The evaluation is strictly zero-shot, leaving open the potential gains from lightweight fine-tuning or prompt engineering on these specialized domains.

Open Questions & Future Work

• ts-fm-fine-tuning-transportation-gains
• homogenization-of-foundation-model-errors
• ts-fm-new-baseline-standard
• integrating-spatial-fm-forecasting

Key Concepts

• Chronos-2: A large-scale, general-purpose time-series foundation model evaluated for transportation forecasting tasks.

Datasets

• highway traffic volume and flow datasets
• urban traffic speed datasets
• bike-sharing demand datasets
• electric vehicle charging station data datasets

Limitations

The evaluation is strictly zero-shot, leaving open the potential gains from lightweight fine-tuning or prompt engineering on these specialized domains.

Links

• Abstract
• PDF