Time Series Foundation Model
Definition
A base model for time series uses extensive pretraining as a starting point for various forecasting and analysis tasks.
A model pretrained on extensive time series corpora intended to be transferable to multiple forecasting or analysis tasks.
Simply explained
The model has already encountered many temporal sequences. This prior knowledge is meant to help it examine or predict new measurement series.
Why it matters in practice
The term helps to select, develop, evaluate AI systems in a well-founded manner and realistically assess their technical risks.
Example from business
A medium-sized company compares a pretrained model with seasonal baselines for product forecasts.
Limits and pitfalls
- The term is not standardized; data frequency, horizon, covariates, and structural breaks limit transferability.
Origin & history
From 2023, works like TimesFM transferred the foundation model paradigm to time series.
Use in mid-sized companies
- Possible application: A medium-sized company compares a pretrained model with seasonal baselines for product forecasts.
Frequently asked questions
What does base model for time series mean?
A model pretrained on extensive time series corpora intended to be transferable to multiple forecasting or analysis tasks. What should you watch out for? The term is not standardized; data frequency, horizon, covariates, and structural breaks limit transferability.
Sources
- A Decoder-only Foundation Model for Time-series Forecasting · Das et al., A Decoder-only Foundation Model for Time-series Forecasting
Source checked on 10.10.2026. Editorial application examples are hypothetical; no evidence of real customer results.
Editorially reviewed on: