Hidden Markov Models for Discrete-Valued Time Series

Authored by: Iain L. MacDonald , Walter Zucchini

Handbook of Discrete-Valued Time Series

Print publication date:  December  2015
Online publication date:  January  2016

Print ISBN: 9781466577732
eBook ISBN: 9781466577749
Adobe ISBN:


 Download Chapter



In the search for useful models for discrete-valued time series, one possible approach is to take a standard model for continuous-valued series, for example, a Gaussian autoregressive moving-average process, and to modify or adapt it in order to allow for the discrete nature of the data. Another approach, the one followed here, is to start from a model for discrete data which assumes independence and then to relax the independence assumption by allowing the distribution of the observations to switch among several possibilities according to a latent Markov chain. To take a simple example, a sequence of independent Poisson random variables with common mean would be inappropriate for a series of unbounded counts displaying significant autocorrelation. But a model that allowed the observations to be Poisson with mean either λ1 or λ2, the choice being made by a discrete-time Markov chain, might well be adequate; a model of this kind, it will be seen, allows for both serial dependence and overdispersion relative to the Poisson.

Search for more...
Back to top

Use of cookies on this website

We are using cookies to provide statistics that help us give you the best experience of our site. You can find out more in our Privacy Policy. By continuing to use the site you are agreeing to our use of cookies.