Paper: arXiv 2505.13521

Authors: Gabor Petnehazi, Laith Al Shaggah, Jozsef Gall, Bernadett Aradi

Abstract

This study explores the potential of zero-shot time series forecasting, an innovative approach leveraging pre-trained foundation models, to forecast mortality rates without task-specific fine-tuning. We evaluate two state-of-the-art foundation models, TimesFM and CHRONOS, alongside traditional and machine learning-based methods across three forecasting horizons (5, 10, and 20 years) using data from 50 countries and 111 age groups. In our investigations, zero-shot models showed varying results: while CHRONOS delivered competitive shorter-term forecasts, outperforming traditional methods like ARIMA and the Lee-Carter model, TimesFM consistently underperformed. Fine-tuning CHRONOS on mortality data significantly improved long-term accuracy. A Random Forest model, trained on mortality data, achieved the best overall performance. These findings underscore the potential of zero-shot forecasting while highlighting the need for careful model selection and domain-specific adaptation.

Complexity vs Empirical Score

  • Math Complexity: 3.0/10
  • Empirical Rigor: 7.5/10
  • Quadrant: Street Traders — practical and empirical, lighter on theory

Why this score: The paper provides an empirical evaluation of zero-shot time series foundation models (TimesFM, CHRONOS) applied to global mortality rate forecasting across 50 countries. While mathematically straightforward, the empirical design is rigorous, evaluating multiple benchmarks across 5, 10, and 20-year horizons using the Human Mortality Database.

Research Flowchart

  flowchart TD
    A[Evaluate Zero-Shot Foundation Models for Global Mortality Forecasting] --> B[Data Prep: 50 Countries, 111 Age Groups]
    B --> C[Model Testing: 5, 10, 20-Year Horizons]
    C --> D1[Zero-Shot Models: CHRONOS & TimesFM]
    C --> D2[Baselines: ARIMA, Lee-Carter & Random Forest]
    C --> D3[Fine-Tuned CHRONOS]
    D1 & D2 & D3 --> E[Comparative Evaluation]
    E --> F[Findings: CHRONOS beats traditional methods in short-term; Fine-Tuned CHRONOS improves long-term; Random Forest achieves best overall]

Cited References (37)

  1. Time series analysis: forecasting and control (2015)
  2. Forecasting: principles and practice (2018)
  3. Long Short-term Memory (1997)
  4. Random forests (2001)
  5. Modeling and forecasting US mortality (1992)
  6. Age-time interactions in mortality projection: Applying Lee-Carter to Australia (2001)
  7. A decoder-only foundation model for time-series forecasting (2023) — link
  8. A time series is worth 64 words: Long-term forecasting with transformers (2022) — link
and 29 more
  1. Chronos: Learning the language of time series (2024) — link
  2. Language models are unsupervised multitask learners (2019)
  3. Exploring the limits of transfer learning with a unified text-to-text transformer (2020)
  4. Human Mortality Database (2024)
  5. TimeGPT-1 (2023) — link
  6. Unified training of universal time series forecasting transformers (2024) — link
  7. Large language models are zero-shot time series forecasters (2024)
  8. Forecastpfn: Synthetically-trained zero-shot forecasting (2024)
  9. Lag-llama: Towards foundation models for time series forecasting (2023)
  10. Application of machine learning to mortality modeling and forecasting (2019)
  11. Mortality rate forecasting: can recurrent neural networks beat the Lee-Carter model? (2019) — link
  12. A neural network extension of the Lee–Carter model to multiple populations (2021)
  13. A deep learning integrated Lee–Carter model (2019)
  14. Calibrating the lee-carter and the poisson lee-carter models via neural networks (2022)
  15. Quantile mortality modelling of multiple populations via neural networks (2024)
  16. Time-series forecasting of mortality rates using transformer (2024)
  17. Transformer self-attention network for forecasting mortality rates (2022)
  18. Mortality Forecasting Using Temporal Fusion Transformers (2024)
  19. Temporal fusion transformers for interpretable multi-horizon time series forecasting (2021)
  20. Attention is all you need (2017)
  21. Improving language understanding by generative pre-training (2018)
  22. A neural-network analyzer for mortality forecast (2018)
  23. Point and interval forecasts of death rates using neural networks (2022)
  24. Automatic time series forecasting: the forecast package for R (2008)
  25. pmdarima : ARIMA estimators for Python (2017–) — link
  26. statsmodels: Econometric and statistical modeling with python (2010)
  27. Keras (2015)
  28. Scikit-learn: Machine Learning in P ython (2011)
  29. Forecasting Hungarian mortality rates using the Lee-Carter method (2007)

0 of 37 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.