ABSTRACT
Generalized linear autoregressive moving average (GLARMA) models are a class of observation-driven non-Gaussian nonlinear state space models in which the state process depends linearly on covariates and nonlinearly on past values of the observed process. Conditional on the state process, the observations are independent and have a distribution 52from the exponential family. This could include continuous responses but, in this chapter, we focus entirely on discrete responses such as binary, binomial, Poisson, or negative binomial distributions.
