MetaARIMA: Configuration-Level Transfer Learning for ARIMA Models

Abstract

Time Series Foundation Models have demonstrated that pretraining on diverse datasets enables effective transfer to new forecasting tasks. Classical methods like ARIMA, however, remain configured and trained independently for each series, discarding knowledge from previous experiments. Consequently, selecting and training an ARIMA configuration remains challenging and often relies on heuristic approaches based on in-sample goodnessof-fit measures. We propose MetaARIMA, a pretrained configuration policy based on metalearning that brings transfer learning to classical statistical forecasting at the configuration level rather than the model level. The policy learns from historical experiments which ARIMA configurations achieve top-quantile performance, and recommends promising candidates for new series while preserving interpretability and local estimation. At inference, it selects a diverse set of candidates via maximal marginal relevance and efficiently identifies the final configuration using successive halving. Experiments in a transfer learning setting, training on M4 and evaluating on M1, M3, and Tourism, show that MetaARIMA outperforms AutoARIMA and other classical methods while being more dataefficient. In a comparison with recent zero-shot foundation models, MetaARIMA remains competitive with Chronos2, Moirai2, and TimesFM. The experiments are publicly available at https://github.com/vcerqueira/experiments-metaarima.

Beyond Source-Level Transfer: Sample-Level Learning for High-Dimensional Quantile Regression
Xiangyu Xing, Yingxuan Wang, Wangli Xu
arXiv7 Oct 2026PreprintarXiv licenceFull textStatisticsarXiv
Residual Deep Learning for Realized Volatility Using a Hybrid HAR-LSTM Model with ARIMA Features and SHAP Interpretability
Cristi Spulbar, Cezar Cătălin Ene
Mathematics9 Oct 2026Peer-reviewedCC BY 4.0General mathematicsCrossref
Machine Learning and ARIMA Model Averaging for Adaptive Public Health Forecasting: Comparative Evaluation and an Ontario COVID-19 Case Study
Yushu Zou, Ye Li, Johra Moosa +3
arXiv31 Aug 2026PreprintarXiv licenceFull textStatisticsarXiv
Transfer learning in bayesian optimization for aircraft design
Ali Tfaily, Youssef Diouane, Nathalie Bartoli +1
Structural and Multidisciplinary Optimization30 Sep 2026Peer-reviewedCC BY-NC-SAFull textGeneral mathematics General physicsHAL
Transfer Learning with Heterogeneous Feature Spaces in Linear Regression
Zejing Zheng, Rui Huang, Junlong Zhao
arXiv8 Sep 2026PreprintCC BY 4.0Full textStatisticsarXiv

Cite

BibTeX RIS

BibTeX

@misc{cerqueira2026metaarima,
  title = {{MetaARIMA: Configuration-Level Transfer Learning for ARIMA Models}},
  author = {Vitor Cerqueira and Ricardo Inácio and Carlos Soares},
  howpublished = {arXiv},
  year = {2026},
  month = oct,
  doi = {10.48550/arxiv.2610.34171},
  eprint = {2610.34171},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.34171},
  note = {CC BY 4.0},
}

RIS

TY  - UNPB
TI  - MetaARIMA: Configuration-Level Transfer Learning for ARIMA Models
AU  - Vitor Cerqueira
AU  - Ricardo Inácio
AU  - Carlos Soares
PY  - 2026
DA  - 2026/10/25
PB  - arXiv
DO  - 10.48550/arxiv.2610.34171
UR  - https://arxiv.org/abs/2610.34171
AB  - Time Series Foundation Models have demonstrated that pretraining on diverse datasets enables effective transfer to new forecasting tasks. Classical methods like ARIMA, however, remain configured and trained independently for each series, discarding knowledge from previous experiments. Consequently, selecting and training an ARIMA configuration remains challenging and often relies on heuristic approaches based on in-sample goodnessof-fit measures. We propose MetaARIMA, a pretrained configuration policy based on metalearning that brings transfer learning to classical statistical forecasting at the configuration level rather than the model level. The policy learns from historical experiments which ARIMA configurations achieve top-quantile performance, and recommends promising candidates for new series while preserving interpretability and local estimation. At inference, it selects a diverse set of candidates via maximal marginal relevance and efficiently identifies the final configuration using successive halving. Experiments in a transfer learning setting, training on M4 and evaluating on M1, M3, and Tourism, show that MetaARIMA outperforms AutoARIMA and other classical methods while being more dataefficient. In a comparison with recent zero-shot foundation models, MetaARIMA remains competitive with Chronos2, Moirai2, and TimesFM. The experiments are publicly available at https://github.com/vcerqueira/experiments-metaarima.
ER  -

Abstract

Related

Cite