Modeling Integer-valued Time Series via Latent Gaussian Process
As the package is not yet on CRAN, it needs to be installed from GitHub:
library(devtools) # if you don't have devtools: install.packages("devtools")
install_github("jlivsey/LatentGaussCounts")
- Fit count time series model for most classic discrete distributions
- Currently we support the following discrete dstributions
- Poisson
- mixture of Poissons
- Allow autocorrelation to depend on an underlying Gaussian process.
- Currently we support the following underlying Gaussian processes
- AR(1)
- FARIMA(0,d,0)
- Estimate model parameters with Gaussian likelihood of a particle filtering MLE.
LGC(x, count.family = "Poisson",
gauss.series = "AR", p=1,
estim.method = "gaussianLik")
James Livsey (United States Census Bureau) and Vladas Pipiras(University of North Carolina, Chapel Hill)
TBD