Title: Estimating a Poisson autoregressive model with the backfitting algorithm
Authors: Paolo Victor Redondo - University of the Philippines Diliman (Philippines) [presenting]
Erniel Barrios - University of the Philippines (Philippines)
Joseph Ryan Lansangan - University of the Philippines (Philippines)
Abstract: A Poisson autoregressive model (PAR) that accounts for discreteness and autocorrelation of count time series data is typically estimated within the context of state-space modelling with maximum likelihood estimation (MLE). The complexity of dependencies exhibited by count time series data however, complicates MLE. PAR is viewed as an additive model and is estimated using a hybrid of conditional least squares and MLE in the backfitting framework. Simulation studies show that estimation of PAR model viewed as an additive model is always better than PAR model in the state-space context whenever the non-normality of covariates for the latter is evident. In cases where the MLE of the PAR model in the state-space context exists, the estimates are comparable with the proposed method. The proposed method is then used in modelling incidence of tuberculosis, elucidating the role of various stakeholders in curbing the prevalence rate of the disease.