Aplicação de processos autorregressivos sazonais com valores inteiros (SINAR(1)) em dados de qualidade do ar: uma abordagem robusta

Abstract

This paper investigates robust inference for the Seasonal Integer-Valued Autoregressive model (SINAR(1)), addressing the limitations of the classical Conditional Maximum Likelihood (CML) estimator under data contamination. We apply an alternative estimation approach based on Huber M-regression. Monte Carlo simulations evaluate the robustness efficiency trade-off under clean and contaminated scenarios. The robust model is more stable under contaminated scenarios, with a modest loss of efficiency under ideal conditions. An empirical application to seasonal Air Quality Index data derived from NASA’s MERRA-2 satellite reanalysis illustrates the proposed methodology, with coherent forecasting results indicating superior performance of the robust estimator. These findings highlight that robust M-regression provides a reliable alternative to likelihood-based estimation when data quality cannot be guaranteed.