Models for Categorical Time Series
Christian H. Weiß · 2018
Also the modeling of categorical processes requires tailor-made approaches, see Chapter 7. Markov models are generally attractive for categorical processes, but without further restrictions concerning the conditional distributions, the number of model parameters becomes quite large. Therefore, this chapter first presents approaches for defining parsimoniously parametrized Markov models. Then we turn to the NDARMA models known from Chapter 5, which are easily adapted to the categorical case. These models constitute some kind of counterpart to the conventional ARMA models, with a serial dependence structure being analogous to those of the ARMA models. Also Hidden-Markov and regression models models are easily applied to categorical processes. The latter commonly use a logit link in the categorical case, and we present regression models for both nominal or ordinal time series.