AutoARIMA

AutoARIMA automatically finds the optimal ARIMA (AutoRegressive Integrated Moving Average) model according to an information criterion. It performs a search over the model orders within given constraints, and selects the set of parameters that optimizes the provided information criterion. The algorithm also supports seasonal components (S) for seasonal patterns and exogenous variables (X) for external influences on the forecast.

Stationary time series

Start Maximum

p is the order (number of time lags) of the auto-regressive (AR) model.

First-differencing order (d)

Degree of differencing. ARIMA models that include differencing (i.e. when d is positive) assume that the data becomes stationary after differencing.
If no value is provided, the value will automatically be selected based on the results of the unit root test; the runtime might be significantly longer.

Unit root test

Try Kwiatkowski–Phillips–Schmidt–Shin Try Augmented Dickey Fuller Try Phillips-Perron

Type of test to use to detect stationarity.

Maximum value of d

Start Maximum

q is the order (number of time lags) of the moving-average (MA) model.

High values are not recommended for season length with AutoARIMA

Start Maximum

P is the order of the auto-regressive (AR) part of the seasonal model.

Seasonal differencing order (D)

If no value is provided, the value will automatically be selected based on the results of the seasonal unit root test.

Seasonal unit root test

Try Osborn, Chui, Smith, and Birchenhall Try Canova-Hansen

Type of test to use to detect seasonality.

Maximum value of D

Start Maximum

Q is the order of the moving-average (MA) part of the seasonal model.

Upper bound for orders

Allows to limit the search space: ARIMA models where p + q + P + Q is greater than the given value will not be fitted.
Must be strictly greater than the sum of the starting values of each order.

Maximum iterations

Maximum number of function evaluations during the search of the best ARIMA model orders.

Information criterion

Try Akaike Try Corrected Akaike Try Bayes Try Hannan-Quinn Try "Out-of-bag" score

Information criterion minimized to find the best ARIMA model.

Solver method

Try Newton-Raphson Try Nelder-Mead Try Broyden-Fletcher-Goldfarb-Shanno Try Limited-memory BFGS Try Powell Try Conjugate gradient Try Newton conjugate gradient Try Global basin-hopping