It’s TIME : Towards the Next Generation of Time Series Forecasting Benchmarks
Zhongzheng Qiao, Sheng Pan, Anni Wang, Viktoriya Zhukova, Yong Liu, Xudong Jiang, Qingsong Wen, Mingsheng Long, Ming Jin, Chenghao Liu · 2026
Read It’s TIME : Towards the Next Generation of Time Series Forecasting Benchmarks by Zhongzheng Qiao, Sheng Pan, Anni Wang, Viktoriya Zhukova, Yong Liu, Xudong Jiang, Qingsong Wen, Mingsheng Long, Ming Jin, Chenghao Liu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-12147.txt.
Description
Forecasting models that claim to work on new data are often tested on old, reused sets, so their scores can flatter them. Zhongzheng Qiao led this benchmark work with nine co-authors, building TIME from 50 fresh data sets and 98 forecasting tasks. The tasks are set up for strict zero-shot tests, with no leaked test data.
People and large language models worked together to check data quality and to shape tasks around real work needs and what can sensibly be predicted. Rather than rating whole data sets, the team scores models by time pattern, using the shape of each series. Twelve foundation models were tested, and the ranked results are shared online for closer study.
Catalog record
| CREATOR | Zhongzheng Qiao, Sheng Pan, Anni Wang, Viktoriya Zhukova, Yong Liu, Xudong Jiang, Qingsong Wen, Mingsheng Long, Ming Jin, Chenghao Liu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 117 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-12147