Title: Estimation of a Poisson autoregressive hidden Markov process with Poisson regression-type measurement errors
Authors: Ruzzel Ragas - University of the Philippines-Diliman (Philippines) [presenting]
Abstract: A method of estimation for Poisson autoregressive model when data is contaminated with error is proposed in the context of hidden Markov modeling paradigm. Model parameters are then estimated by its maximum likelihood estimator computed using data cloning method. Subsequently, particle filter is used to estimate the hidden process. Simulation studies indicate that bias and standard error of the parameter estimates decreases as length of the time series increases. However, misspecification of the covariate parameter leads to poor predictive ability of the model, while in most cases, predictive ability is not affected by the length of the time series.