Hugging Face's logo

Spaces:
FinText
/
README
Running

App Files Community
1
README / README.md
Eghbal's picture
Eghbal
Update README.md
f8c208a verified
preview code
|
raw
history blame
4.42 kB
metadata 
title: FinText-TSFM
emoji: 📈
colorFrom: gray
colorTo: blue
sdk: static
pinned: false

FinText-TSFM 📈

Time Series Foundation Models for Finance

🚀 TSFMs Release

We are pleased to introduce FinText-TSFM, a comprehensive suite of time series foundation models (TSFMs) developed for financial forecasting and quantitative research. This release accompanies the paper : Re(Visiting) Time Series Foundation Models in Finance by Eghbal Rahimikia, Hao Ni, and Weiguan Wang (2025).

💡 Key Highlights

  • Finance-Native Pre-training:
    Models are pre-trained from scratch on large-scale financial time series datasets — including daily excess returns across 89 markets and over 2 billion observations — to ensure full temporal and domain alignment.

  • Bias-Free Design:
    Training strictly follows a chronological expanding-window setup, avoiding any look-ahead bias or information leakage.

  • Model Families:
    This release includes variants of Chronos and TimesFM architectures adapted for financial time series:

    • Chronos-Tiny / Mini / Small
    • TimesFM-8M / 20M

  • Performance Insights:
    Our findings show that off-the-shelf TSFMs underperform in zero-shot forecasting, while finance-pretrained models achieve large gains in both predictive accuracy and portfolio Sharpe ratios.

  • Evaluation Scope:
    Models are benchmarked across U.S. and international markets, using rolling windows (5, 21, 252, 512 days) and 18M+ out-of-sample forecasts.

🧠 Technical Overview

  • Architecture: Transformer-based TSFMs (Chronos & TimesFM)
  • Training Regime: Pre-training from scratch, fine-tuning, and zero-shot evaluation
  • Compute: >50,000 GPU hours on NVIDIA GH200 Grace Hopper clusters

📚 Citation

Please cite the accompanying paper if you use these models:

Re(Visiting) Time Series Foundation Models in Finance.
Rahimikia, Eghbal; Ni, Hao; Wang, Weiguan.
SSRN: https://ssrn.com/abstract=4963618
DOI: 10.2139/ssrn.4963618

🔋 Acknowledgments

This project was made possible through computational and institutional support from:

  • Isambard-AI National AI Research Resource (AIRR)
  • The University of Manchester (Research IT & Computational Shared Facility)
  • Alliance Manchester Business School (AMBS), University of Manchester
  • University College London (UCL)
  • Shanghai University
  • N8 Centre of Excellence in Computationally Intensive Research (N8 CIR)
  • The Alan Turing Institute

Developed by:

<!-- Left side: developed by logos -->
<div style="display: flex; align-items: center; justify-content: center; gap: 25px; flex-wrap: wrap;">
  <img src="https://fintext.ai/UoM-logo.svg" alt="University of Manchester Logo" style="width: 210px; height: auto; margin: 0;">
  <img src="https://fintext.ai/UCL-logo.jpg" alt="UCL Logo" style="width: 100px; height: auto; margin: 0;">
</div>

<!-- Right side: powered by BriCS -->
<div style="display: flex; flex-direction: column; align-items: center; justify-content: center;">
  <p style="font-weight: bold; font-size: 1em; margin: 0 0 4px 0;">Powered by:</p>
  <img src="https://fintext.ai/BriCS-logo.svg" alt="BriCS Logo" style="width: 150px; height: auto; margin-bottom: 2px;">
  <p style="font-size: 0.8em; margin: 0;">Bristol Supercomputing</p>
</div>

Alliance Manchester Business School, The University of Manchester
Department of Mathematics, University College London