DynCount: Bayesian Dynamic Models for Poisson and Binomial Time Series
Fits Bayesian state-space models for non-Gaussian time series using a latent log-rate (Poisson) or
latent logit (binomial) formulation. The latent trajectory follows a
first-order random walk or a stationary AR(1) process, sampled by
Metropolis-within-Gibbs using the implied Gaussian Markov random field (GMRF) full conditionals. Four innovation
structures are supported for the latent increments: constant-variance
Gaussian, Student-t, a finite scale mixture of normals, and stochastic
volatility. Both families support time-constant zero inflation. The
package provides simulation, fitting, forecasting, summary and plotting
tools. It implements and extends the methodology of Zens and Bijak (2026)
<doi:10.1214/26-AOAS2171>.
Documentation:
Downloads:
Linking:
Please use the canonical form
https://CRAN.R-project.org/package=DynCount
to link to this page.