A0776
Title: Vector generalized linear models for time series of counts
Authors: Victor Miranda - Auckland University of Technology (New Zealand) [presenting]
Thomas Yee - University of Auckland (New Zealand)
Abstract: Modeling time series of counts (TSCs) presents unique challenges due to the discrete, non-negative, and often over-dispersed nature of the data. In recent decades, substantial progress has been made in developing models in the context of generalized linear models (GLMs) to accommodate these features, such as Poisson and negative binomial autoregression. However, quite often, GLM-based models for TSCs can be restrictive when addressing multivariate responses or structural features such as zero inflation and parameter constraints. The class of vector generalized linear models (VGLMs) is shown to confer advantages for modeling count-valued data exhibiting time-varying trends and dynamic patterns. Estimated by maximum likelihood using Fisher scoring, we essentially adapt VGLMs to time-indexed count data, offering a range of alternatives by incorporating lagged responses and covariates, as well as the ability to handle multiple, potentially interrelated response variables, linear predictors, link functions, and distributional assumptions within a unified likelihood-based framework. Following previous successful endeavors of developing VGLMs for other data types, such as categorical data, extremes, and quantile regression, positioning VGLMs as a natural extension of the foundational prior study on likelihood-based modelling for counts, so broadening the class of tools for discrete time series analysis.
