Classical time series

Back to Models and Features

Classical time series models work from the target’s own history, including autoregressions, ARIMA, and exponential smoothing, and serve as the standard benchmarks.

Pass any model string below as Arm(model=...). Extra names an optional dependency, Scaling flags whether predictors should be standardized, and Tunable counts the hyperparameters the search space exposes.

Model string	Description	Input	Extra	Scaling	Recommended preprocessing	Tunable
ar	Univariate autoregression.	supervised	none	no	default	1
arima	(Seasonal) ARIMA model.	target	none	no	default	1
auto_arima	Automatic (seasonal) ARIMA order selection (forecast::auto.arima).	target	none	no	default	0
bvar_minnesota	FAVAR::BVAR / bvartools Minnesota-prior Bayesian VAR posterior sampler.	panel	none	no	default	3
bvar_normal_inverse_wishart	FAVAR::BVAR-aligned Bayesian VAR with normal/inverse-Wishart prior controls.	panel	none	no	default	1
ets	Statsmodels ETS target-only forecasting model.	target	none	no	default	0
hist_mean	Historical (prevailing) mean benchmark of the transformed target.	target	none	no	default	0
holt_winters	Holt-Winters exponential smoothing target-only forecasting model.	target	none	no	default	0
naive	Random-walk (naive) baseline: carry the last value forward (forecast::naive).	target	none	no	default	0
random_walk_drift	Random-walk-with-drift baseline (forecast::rwf(drift=TRUE)).	target	none	no	default	0
seasonal_naive	Seasonal-naive baseline: repeat the last seasonal cycle (forecast::snaive).	target	none	no	default	0
stlf	STL decomposition + forecast of the seasonally-adjusted series (forecast::stlf).	target	none	no	default	0
theta_method	Theta method target-only forecasting model.	target	none	no	default	0
var	R vars::VAR-aligned vector autoregression point forecast.	panel	none	no	default	1

Reference

• Models reference page for ModelSpec, ModelFit, and fit conventions.