Estudo de previsão e estimação do processo Poisson INAR(1)

Abstract

This work aims to study parameter estimation and forecasting in the Poisson first-order Integer-Valued Autoregressive (INAR(1)) process through Monte Carlo simulation. We consider Yule-Walker, Conditional Least Squares and Conditional Maximum Likelihood estimation methods. We compare the performance estimators using bias and Mean Square Error (MSE). Given that we know the series up to time t, we propose the nearest integer of the conditional expectation two steps ahead and the median of the conditional distribution two steps ahead as a prediction of time series value at time t + 2. We evaluate the performance of predictors using the mean squared prediction error and mean absolute prediction error considering the different estimation methods.