Re(Visiting) Time Series Foundation Models in Finance
Organizations: Alliance Manchester Business School, University of Manchester · Department of Mathematics, University College London (UCL) · School of Economics, Shanghai University
Abstract
Financial time series forecasting is vital for trading, portfolio optimization, and risk management but is challenged by noisy, nonstationary, and heterogeneous data. Time series foundation models (TSFMs), inspired by large language models, aim to learn transferable temporal representations. Whether pre-training on massive, diverse datasets enables similar generalization to finance remains an open question. This study provides a comprehensive empirical evaluation of TSFMs in global financial markets using large-scale daily excess-return data. We assess zero-shot inference, fine-tuning, and pre-training from scratch against strong benchmarks. Off-the-shelf TSFMs generally perform poorly, while finance-native pre-training improves long-window gross portfolio performance. Larger pre-training datasets and synthetic augmentation improve selected specifications, particularly at longer input windows. Configuration changes have mixed effects across forecasting and portfolio metrics, and TSFMs do not consistently outperform strong benchmarks. These findings caution against uncritical AI adoption in finance, where insufficient validation can lead to unreliable forecasts and substantial financial losses.
Figures & tables
Appendix figures & tables17 assets
Supplementary material from the paper’s appendix.
Appendix
| Model | Hyperparameters |
| OLS+H | Tuned: – Regularization strength ( ): searched over values from to on a logarithmic scale (8 values) Fixed: – Robustness parameter ( ): 1.35 |
| LASSO+H | Tuned: – Regularization strength ( ): searched over values from to on a logarithmic scale (8 values) Fixed: – Huber threshold ( ): 1.35 – Penalty: L1 (Lasso) |
| RIDGE+H | Tuned: – Regularization strength ( ): searched over values from to on a logarithmic scale (8 values) Fixed: – Huber threshold ( ): 1.35 – Penalty: L2 (Ridge) |
| Enet+H | Tuned: – Regularization strength ( ): searched over values from to on a logarithmic scale (8 values) – Elastic Net mixing parameter ( ): Fixed: – Huber threshold ( ): 1.35 – Penalty: Elastic Net |
| PCR | Tuned: – Number of principal components: candidate sets by input-window length are as follows: : ; : ; : ; : Fixed: – PCA algorithm: Incremental PCA – Regression: Linear Regression (OLS) |
| XGBoost | Tuned: – Maximum tree depth: – Learning rate ( ): Fixed: – Booster: gradient boosted trees |
Time Series Forecasting
Time series forecasting predicts future values from ordered observations using statistical, neural, and probabilistic models to support decisions under uncertainty.
See all 939 papers