Actionable insights, from research-grade forecasting.

OpenForecast builds open-source forecasting tools and provides consulting and training in demand forecasting and inventory management. Our packages are the rigorous engine behind that work: grounded in peer-reviewed research, refined over a decade, and used by analysts worldwide.

The core of the toolkit: forecasting built on the ADAM framework, which unifies exponential smoothing (ETS), ARIMA, and regression in a single state-space model. It produces full probabilistic forecasts rather than single point predictions, handles multiple seasonalities and short histories, and models the intermittent demand that standard tools get badly wrong.

# R
install.packages("smooth")

# Python
pip install smooth

Repository · CRAN · PyPI

The companion toolkit that brings the real drivers of demand into the model: regression with information-criteria-based variable selection, explanatory forecasting, and honest forecast evaluation. This is how you explain why demand moves (promotions, pricing, seasonality) instead of extrapolating from history alone. The pacakge also includes tools for demand classification tehniques.

# R
install.packages("greybox")

# Python
pip install greybox

Repository · CRAN · PyPI

Models for forecasting several related series together via the vector exponential smoothing. Useful when products, regions, or channels move together and when shared dynamics carry information a univariate model throws away.

install.packages("legion")

Repository · CRAN

Research-grade implementation of the MSOE state-space framework — the multiple-source-of-error counterpart to the single-source models. Where the frontier of our methodology work lives.

Repository

The M1, M3, and tourism competition datasets, packaged for Python — for benchmarking your methods against the same data the field uses.

Repository

• Forecasts you can plan around — honest uncertainty around every forecast, so safety stock can be set deliberately instead of guessed.
• The hard cases, handled — intermittent and slow-moving items, spare parts, and short data histories where bigger models fail.
• Demand explained, not just extrapolated — bring promotions, pricing, and seasonality into the model as real drivers.
• Glass-box, not black-box — transparent, interpretable models you can explain to finance and stand behind.
• Built to scale — automated model selection across thousands of series, in minutes.

The packages are freely available, and that transparency is the point. Our value is in applying them: turning your demand data and ERP into working forecasting and replenishment decisions: correcting for stockout-censored sales, capturing your demand drivers, and building models around how your business actually runs. We also run practitioner training in forecasting, inventory management, and analytics.

Website: openforecast.org · Email: mail@openforecast.org · LinkedIn: OpenForecast

_{Methodology: ADAM (Svetunkov, 2023). Built and maintained by the OpenForecast team.}