Official Code for the Paper "Meta-Learning for Cross-Sectional Return Prediction in Financial Markets"

In this paper, we propose the Financial Prior-Data Fitted Network (FinPFN), a meta-learning framework utilizing a Transformer architecture for cross-sectional stock return prediction.

Getting Started

This is a Python project, we used Python 3.10 in development and recommend to use a virtualenv or conda. To use our code, clone the project with

git clone [email protected]:wangy8989/FinPFN.git

install all dependencies with

pip install -r requirements.txt

The TabPFN version we used is 2.0.8.

The code is forked from finetune_tabpfn_v2.

Training a model

data_utils.py provides the dataloader of financial data prior for the model.

main.py provides methods to finetune a TabPFN model.

Evaluating Models

finpfn.ipynb provides a workflow to train and evaluate models.

Downloading Data

Data link with Password: finpfn12345

Cite

When using, please cite FinPFN <-- free link for 50 days

@article{wang2025finpfn,
 title = {Meta-learning for return prediction in shifting market regimes},
 journal = {Journal of Financial Markets},
 pages = {101042},
 year = {2025},
 issn = {1386-4181},
 doi = {https://doi.org/10.1016/j.finmar.2025.101042},
 url = {https://www.sciencedirect.com/science/article/pii/S1386418125000825},
 author = {Yicheng Wang and Sandro Claudio Lera},
 keywords = {Financial machine learning, Return prediction, Regime shifts, Meta-learning},
 abstract = {We propose a meta-learning framework for cross-sectional return prediction that adapts to regime-dependent dynamics. Instead of learning a fixed mapping from features to returns, we condition our model forecasts on recent feature-return relationships. This allows it to adjust to evolving market states without explicit regime labels or frequent re-estimation. We implement the framework with a Transformer-based Bayesian predictor, the Financial Prior-data Fitted Network (FinPFN), and evaluate it on daily Chinese A-shares and monthly U.S. equities. During regime changes, proxied by large volatility shifts, our method significantly outperforms benchmarks, offering a practical tool for dynamic return prediction.}
}

TabPFNs were from

@article{hollmann2025tabpfn,
 title={Accurate predictions on small data with a tabular foundation model},
 author={Hollmann, Noah and M{\"u}ller, Samuel and Purucker, Lennart and
         Krishnakumar, Arjun and K{\"o}rfer, Max and Hoo, Shi Bin and
         Schirrmeister, Robin Tibor and Hutter, Frank},
 journal={Nature},
 year={2025},
 month={01},
 day={09},
 doi={10.1038/s41586-024-08328-6},
 publisher={Springer Nature},
 url={https://www.nature.com/articles/s41586-024-08328-6},
}