Re(Visiting) Time Series Foundation Models in Finance

ArXiv ID: 2511.18578 “View on arXiv”

Authors: Eghbal Rahimikia, Hao Ni, Weiguan Wang

Abstract

Financial time series forecasting is central to trading, portfolio optimization, and risk management, yet it remains challenging due to noisy, non-stationary, and heterogeneous data. Recent advances in time series foundation models (TSFMs), inspired by large language models, offer a new paradigm for learning generalizable temporal representations from large and diverse datasets. This paper presents the first comprehensive empirical study of TSFMs in global financial markets. Using a large-scale dataset of daily excess returns across diverse markets, we evaluate zero-shot inference, fine-tuning, and pre-training from scratch against strong benchmark models. We find that off-the-shelf pre-trained TSFMs perform poorly in zero-shot and fine-tuning settings, whereas models pre-trained from scratch on financial data achieve substantial forecasting and economic improvements, underscoring the value of domain-specific adaptation. Increasing the dataset size, incorporating synthetic data augmentation, and applying hyperparameter tuning further enhance performance.

Keywords: Time Series Foundation Models (TSFMs), Zero-Shot Learning, Fine-Tuning, Synthetic Data Augmentation, Global Financial Markets, Equities

Complexity vs Empirical Score

  • Math Complexity: 6.0/10
  • Empirical Rigor: 8.5/10
  • Quadrant: Holy Grail
  • Why: The paper employs advanced machine learning architectures (TSFMs) and complex evaluation methodologies, but its primary strength is the large-scale empirical study using real financial data (2 billion observations) with robust backtesting protocols, economic metrics, and benchmarks.
  flowchart TD
    A["Research Goal: Evaluate TSFMs in Global Financial Markets"] --> B["Methodology: Zero-Shot, Fine-Tuning, Pre-train from Scratch"]
    B --> C["Data: Daily Excess Returns<br>Global Equities Markets"]
    C --> D["Computational Process: Benchmarking vs. TSFMs"]
    D --> E["Key Findings: Domain-Specific Pre-training<br>Significantly Outperforms Off-the-Shelf TSFMs"]
    E --> F["Outcomes: Improved Forecasting & Economics via<br>Data Scaling, Synthetic Augmentation, Hyperparameter Tuning"]