Robust Akaike Information Criterion for ARMA models

A robust version of the Akaike Information Criterion (AIC) [5] is defined to the aim of selecting the order of an ARMA model. It is an extended version of the classical criterion based on weighted likelihood methodology [12]. To achieve robustness a weight is associated to each component of the conditional log–likelihood [3]. This criterion is asymptotically equivalent to the classical one when no outliers are present; its robustness are studied in presence of additive and innovative outliers [7] with symmetric and asymmetric contamination by Monte Carlo simulations