Download
StatsForecast is a Python library for time-series forecasting that delivers a suite of classical statistical and econometric forecasting models optimized for high performance and scalability. It is designed not just for academic experiments but for production-level time-series forecasting, meaning it handles forecasting for many series at once, efficiently, reliably, and with minimal overhead. The library implements a broad set of models, including AutoARIMA, ETS, CES, Theta, plus a battery of benchmarking and baseline methods, giving users flexibility in selecting forecasting approaches depending on data characteristics (trend, seasonality, intermittent demand, etc.). Its internal implementation leverages numba to compile performance-critical code to optimized machine-level instructions, which makes the models much faster than many traditional Python counterparts.
Features
- Collection of widely used univariate time-series forecasting models (AutoARIMA, AutoETS, AutoCES, Theta, etc.)
- Scikit-learn–style .fit() and .predict() API for ease of use
- High performance via Numba JIT compilation — significantly faster than many standard Python/R implementations
- Support for exogenous variables and static covariates in forecasts for richer modeling
- Probabilistic forecasts with confidence intervals and built-in anomaly detection
- Scales to large workloads: can forecast thousands or millions of series using distributed backends (Ray, Spark, Dask)
Project Samples
