statsforecast

Why we included this project

Forecasting univariate series in bulk is where StatsForecast earns its keep. It puts production-grade ARIMA, ETS, CES, Theta, and MSTL implementations behind a plain fit/predict interface, and the C++-backed kernels are fast enough that large collections of series finish in a fraction of the time pmdarima or statsmodels would need. It also covers the details real pipelines ask for: exogenous regressors, prediction intervals, multiple seasonalities, and anomaly detection, with distributed execution on Spark, Dask, or Ray that slots into your existing workflow. The bundled benchmark models give you honest baselines when you compare a statistical approach against a neural or foundation-model forecaster. If you want a dependable statistical forecast or a faster alternative to Prophet-style tooling, this is a strong starting point.