arimasel

Cartesian product-based (seasonal) ARIMA model identification and selection for Python.

A Python port of the R package arimasel. Rather than relying on ACF/PACF plots or a stepwise search (as in auto_arima), arimasel exhaustively evaluates every candidate (p,d,q)(P,D,Q)[m] model in the Cartesian product of user-supplied index sets, ranks all converged models simultaneously by AIC, AICc, BIC, and HQIC, and quantifies model uncertainty with Akaike weights. Built on statsmodels.

Features

• Exhaustive, transparent search — every candidate model in the search space is fit; no greedy stepwise heuristics.
• Seasonal ARIMA — search over (p,d,q) combined with seasonal (P,D,Q)[period].
• Exogenous regressors — regression with ARIMA errors via exog / exog_future.
• Four criteria at once — AIC, AICc, BIC, HQIC, plus a criterion "vote table".
• Akaike weights — model uncertainty quantification and weighted ensemble forecasting.
• Rolling-origin cross-validation — genuine out-of-sample backtesting via arima_cv.
• Feature-based EDA — ts_features / ts_eda compute scale-free time series characteristics (trend/seasonal strength, spectral entropy, ACF(1), lumpiness, stability) in the spirit of Hyndman, Wang and Laptev (2015).
• Feature-guided automatic search — smart_arima narrows the search space using those features, transparently, before running the exhaustive search.
• Bundled datasets — three Nigerian macroeconomic series (gdp_ng, inflation_ng, exchange_ng) for examples and testing.

Installation

pip install arimasel

# with plotting support
pip install arimasel[plot]

Quick start

import arimasel as am

x = am.load_gdp_ng()

result = am.cart_arima(x, p_set=range(3), d_set=range(2), q_set=range(3))
print(result)

forecast = am.arima_forecast(result, h=5, ensemble=True, top_k=3)
print(forecast)
forecast.plot()  # requires arimasel[plot]

Seasonal search

x = am.load_inflation_ng()

result = am.cart_arima(
    x, p_set=range(2), d_set=[0, 1], q_set=range(2),
    seasonal={"P": [0, 1], "D": [0, 1], "Q": [0, 1], "period": 12},
)
print(result)

Exploratory data analysis and feature-guided search

eda = am.ts_eda(x, period=12)
print(eda)
eda.plot()

result = am.smart_arima(x, p_set=range(3), q_set=range(3), period=12)

Rolling-origin cross-validation

cv = am.arima_cv(result, h=3, initial=200)
print(cv)
cv.plot(kind="rmse")

API overview

Companion R package

An R version of this package, also called arimasel, is available at https://github.com/Olawaleawe/arimasel.

License

GPL-3.0-or-later

Author

Olushina Olawale Awe (olawaleawe@gmail.com) — ORCID: 0000-0002-0442-4519