StatsForecast

StatsForecast

Lightning-fast forecasting library from Nixtla with statistical and econometric models. Apache-2.0 licensed.

StatsForecast provides lightning-fast statistical forecasting models including ARIMA, ETS, Theta, and seasonal methods. Nixtla reports its AutoARIMA is 20x faster than pmdarima and 1.5x faster than R in its own benchmarks.

Good for

• Baseline forecasts for financial time series
• Large-scale forecasting across thousands of series
• Benchmark comparisons for ML models

Verdict

Essential starting point for time series forecasting. In Nixtla's own benchmark on M4 competition data (daily, hourly and weekly series), its AutoARIMA had lower MASE than pmdarima on the daily and weekly sets (pmdarima failed on hourly data) and was about 20x faster on the daily set. These are Nixtla's results, not an independent test. Install with `pip install statsforecast`. Use these classical models as baselines before exploring neural approaches.