Repolode
Submit
microsoft

microsoft/finnts

An R package that automates time series forecasting for financial data, with an AI agent, 25+ models and Azure-based parallel runs.

View on GitHubWebsite
Stars
281
Forks
53
Open issues
12
Last commit
4 days ago

What is it?

Finn (finnts) is an automated forecasting framework from Microsoft, originally built for corporate finance work but usable for any time series forecasting problem. It handles feature engineering, feature selection, back testing and model selection automatically, and covers daily, weekly, monthly, quarterly and yearly horizons.

The package offers access to more than 25 univariate and multivariate models and supports external regressors, whether they only have historical values or both historical and future values. For large portfolios, it can run thousands of time series in parallel in the Azure cloud.

Finn also includes a built-in AI agent that iterates on the forecast toward an accuracy target and can answer questions about the exploratory analysis, accuracy and chosen model. It is installed from CRAN or GitHub, and a coding-agent skill is available for users who prefer not to write R code.

Key features

  • 01Automated feature engineering, feature selection, back testing and model selection.
  • 02Access to more than 25 univariate and multivariate forecasting models.
  • 03Azure integration for running thousands of time series in parallel.
  • 04Supports daily, weekly, monthly, quarterly and yearly forecast frequencies.
  • 05Built-in AI agent that iterates on forecasts and answers questions about results.

Use cases

  • Producing corporate financial forecasts across many products or business units.
  • Running batch forecasts over large collections of time series in the cloud.
  • Comparing many forecasting models on the same data with automated back testing.
  • Getting plain-language explanations of forecast results without writing R code.

Quick start

  1. 1

    Install the CRAN version of finnts in R

    install.packages("finnts")
  2. 2

    Load the package and prepare your historical data, then connect an LLM for the agent

  3. 3

    Set up the project, agent, and run the forecast iteration

Things to know

  • The built-in AI agent requires access to a large language model, such as Azure OpenAI, through the ellmer package.
  • Cloud-scale parallel runs depend on Azure integration.

Topics

#time-series#forecasting#r-package#finance#automl#azure#llm#agent#ai#business

FAQ

What is finnts?
An R package that automates time series forecasting for financial data, with an AI agent, 25+ models and Azure-based parallel runs. Finn (finnts) is an automated forecasting framework from Microsoft, originally built for corporate finance work but usable for any time series forecasting problem. It handles feature engineering, feature selection, back testing and model selection automatically, and covers daily, weekly, monthly, quarterly and yearly horizons.
Is finnts free to use? What license does it have?
finnts is open source under the Other license. Check the license terms before using it in commercial products.
Is finnts still maintained?
Actively maintained. The last commit to finnts was on Nov 2, 2026.
What can I use finnts for?
Producing corporate financial forecasts across many products or business units. Running batch forecasts over large collections of time series in the cloud. Comparing many forecasting models on the same data with automated back testing. Getting plain-language explanations of forecast results without writing R code.
How do I install finnts?
Install the CRAN version of finnts in R: install.packages("finnts") Load the package and prepare your historical data, then connect an LLM for the agent Set up the project, agent, and run the forecast iteration
Where can I get finnts?
The source code and installation instructions are on GitHub: https://github.com/microsoft/finnts

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects