arimasel: Cartesian Product-Based ARIMA Model Identification and Selection
Provides an alternative algorithm for ARIMA and seasonal ARIMA
model identification based on Cartesian products of user-supplied
parameter sets. Rather than relying on ACF/PACF plots or stepwise search
(as in auto.arima()), the package exhaustively evaluates every candidate
(p,d,q)(P,D,Q)[m] combination in the requested index sets, ranks all
converged models by AIC, AICc, BIC, and HQIC simultaneously, computes
Akaike weights for model uncertainty quantification, supports exogenous
regressors, produces ensemble forecasts, evaluates candidate models by
rolling-origin (expanding window) cross-validation, and provides
publication-quality diagnostic and comparison plots. A feature-based
exploratory data analysis suite computes scale-free time series
characteristics (trend and seasonal strength, spectral entropy,
autocorrelation, lumpiness, stability) in the spirit of Hyndman, Wang
and Laptev (2015), and a feature-guided automatic search narrows the
Cartesian product model space before the exhaustive search runs. The
algorithm is flexible, transparent, and widely applicable for quick,
reproducible ARIMA model selection in both academic research and
industry forecasting pipelines. Applications are demonstrated with
Nigerian macroeconomic time series data.
Documentation:
Downloads:
Linking:
Please use the canonical form
https://CRAN.R-project.org/package=arimasel
to link to this page.
[aut, cre]