Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Generalisation Bounds of Zero-Shot Economic Forecasting using Time Series Foundation Models

Jittarin Jetwiriyanon, Teo Susnjak, Surangika Ranathunga

TL;DR

This paper assesses zero-shot forecasting with Time Series Foundation Models (TSFMs) for macroeconomic indicators, using New Zealand data as a challenging testbed. By evaluating TimeGPT-1, Chronos, and Moirai in a purely univariate, pre-trained setting against classical baselines (Persistence, ARIMA, LSBoost, and Factor models), the study demonstrates that Moirai-based architectures can match or exceed traditional benchmarks during stable periods and exhibit notable resilience during shocks, though performance can degrade without domain covariates. Key findings show Moirai often delivering the lowest RMSE and good calibration across horizons, with Chronos also performing strongly, while TimeGPT exhibits mixed results. The results offer practical guidance on when zero-shot deployments are viable for macro monitoring and highlight the need for multivariate extensions, modest fine-tuning, and calibrated predictive distributions to handle regime changes and crises more robustly.

Abstract

This study investigates zero-shot forecasting capabilities of Time Series Foundation Models (TSFMs) for macroeconomic indicators. We apply TSFMs to forecasting economic indicators under univariate conditions, bypassing the need for train bespoke econometric models using and extensive training datasets. Our experiments were conducted on a case study dataset, without additional customisation. We rigorously back-tested three state-of-the-art TSFMs (Chronos, TimeGPT and Moirai) under data-scarce conditions and structural breaks. Our results demonstrate that appropriately engineered TSFMs can internalise rich economic dynamics, accommodate regime shifts, and deliver well-behaved uncertainty estimates out of the box, while matching state-of-the-art multivariate models on this domain. Our findings suggest that, without any fine-tuning, TSFMs can match or exceed classical models during stable economic conditions. However, they are vulnerable to degradation in performances during periods of rapid shocks. The findings offer guidance to practitioners on when zero-shot deployments are viable for macroeconomic monitoring and strategic planning.

Generalisation Bounds of Zero-Shot Economic Forecasting using Time Series Foundation Models

TL;DR

This paper assesses zero-shot forecasting with Time Series Foundation Models (TSFMs) for macroeconomic indicators, using New Zealand data as a challenging testbed. By evaluating TimeGPT-1, Chronos, and Moirai in a purely univariate, pre-trained setting against classical baselines (Persistence, ARIMA, LSBoost, and Factor models), the study demonstrates that Moirai-based architectures can match or exceed traditional benchmarks during stable periods and exhibit notable resilience during shocks, though performance can degrade without domain covariates. Key findings show Moirai often delivering the lowest RMSE and good calibration across horizons, with Chronos also performing strongly, while TimeGPT exhibits mixed results. The results offer practical guidance on when zero-shot deployments are viable for macro monitoring and highlight the need for multivariate extensions, modest fine-tuning, and calibrated predictive distributions to handle regime changes and crises more robustly.

Abstract

This study investigates zero-shot forecasting capabilities of Time Series Foundation Models (TSFMs) for macroeconomic indicators. We apply TSFMs to forecasting economic indicators under univariate conditions, bypassing the need for train bespoke econometric models using and extensive training datasets. Our experiments were conducted on a case study dataset, without additional customisation. We rigorously back-tested three state-of-the-art TSFMs (Chronos, TimeGPT and Moirai) under data-scarce conditions and structural breaks. Our results demonstrate that appropriately engineered TSFMs can internalise rich economic dynamics, accommodate regime shifts, and deliver well-behaved uncertainty estimates out of the box, while matching state-of-the-art multivariate models on this domain. Our findings suggest that, without any fine-tuning, TSFMs can match or exceed classical models during stable economic conditions. However, they are vulnerable to degradation in performances during periods of rapid shocks. The findings offer guidance to practitioners on when zero-shot deployments are viable for macroeconomic monitoring and strategic planning.

Paper Structure

This paper contains 35 sections, 16 equations, 6 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related works
  3. Forecasting Difficulty for Macroeconomic Indicators
  4. Modern and Emerging Forecasting Approaches
  5. Zero-Shot Transfer Learning for Macroeconomic Forecasting
  6. Zero-Shot TSFMs in Economic Forecasting
  7. Summary and Research Questions
  8. Methodology
  9. Dataset
  10. RBNZ Operational Dataset
  11. Baseline Models
  12. Persistence model
  13. ARIMA model
  14. LSBoost(Least-Squares Boosting)
  15. Time Series Foundation Models
  16. ...and 20 more sections

Figures (6)

  • Figure 1: The transformer architecture. vaswani2017attention
  • Figure 2: The experiment workflow
  • Figure 3: Actual and forecasting values for the quarterly annual percentage change from 2017Q1 to 2024Q4.
  • Figure 4: A regression plot of actual GDP variance against MSE over 26 years.
  • Figure 5: Comparisons of the Moirai Base Model with LSboost and Factor Model using Reserve Bank Nowcasts.
  • ...and 1 more figures