Robust Identification of Large Subset ARX Systems

Robust Subset stepwise regression for large-scale single-output, multiple-inputs ARX systems with irregular polynomial structure.

The archive contains 2 Matlab functions for the estimation and identification of single-output ARX systems (namenly, Auto-Regressive with eXogenous variables), with many inputs and many sparse lagged terms. The subset identification/estimation strategy consist o selecting significant exogenous lagged terms Xj(t-i) by estimating m-bivariate ARX(p,q) models, putting them in the multiple model and estimating it with backward regression. Estimation methods are ordinary least squares (OLS) with heteoskedastic consistent (HC) estimates of standard errors, and robust M-type estimators with bisquare loss functions. Two demo scripts illustrate the functions.