The TimesFM model

This document describes BigQuery ML's built-inTimesFM time series forecasting model.

The built-in TimesFM univariate model is an implementation of Google Research'sopen sourceTimesFM model. The Google ResearchTimesFM model is a foundation model for time-series forecasting that has beenpre-trained on billions of time-points from many real-world datasets, so youcan apply it to new forecasting datasets across many domains. The TimesFM model is available in all BigQuery supported regions.

Using BigQuery ML's built-in TimesFM model with theAI.FORECAST functionlets you performforecasting without having to create and train your own model, so you canavoid the need for model management.The forecast results from the TimesFM model are comparable toconventional statistical methods such as ARIMA. If you want moremodel tuning options than the TimesFM model offers, you can create anARIMA_PLUSorARIMA_PLUS_XREGmodel and use it with theML.FORECAST functioninstead.

To try using a TimesFM model with theAI.FORECAST function, seeForecast multiple time series with a TimesFM univariate model.

To use the TimesFM model to detect anomalies in time series data, use theAI.DETECT_ANOMALIES function(Preview).

To evaluate forecasted values from the TimesFM model against the actual values,use theAI.EVALUATE function.

To learn more about the Google Research TimesFM model, use the followingresources:

• Google Research blog
• GitHub repository
• Hugging Face page