direct_GCNB_ARMA_likelihood: direct_GCNB_ARMA_likelihood

In hlennon/copulaIVTS: Dependence modelling of a digitised Gaussian ARMA(p,q) copula model for Integer-Valued Time Series

Description

Direct Likelihood evaluation of GCNB model with ARMA(p,q) correlation matrix Requires: mnormt, mvtnorm

Usage

direct_GCNB_ARMA_likelihood(covariances, nb.p, nb.s, data, method="mnormt")

Arguments

covariances	Theoretical ARMA(p,q) covariances corresponding to the dependence structure
nb.p	X_t ~ NB (p, s)
nb.s	X_t ~ NB (p, s)
data	observed data
method	a string of value "mnormt" or "mvtnorm" only

Details

Evaluated likelihood function directly

Value

A numeric value of the evaluated likelihood function given the observed data and dependence parameters, and marginal parameters nb.p and nb.s.

Note

ARMA(p,q) dependence structure is used.

Cannot be optimised using optim(numeric(n), fn=direct_GCNB_ARMA_likelihood, nb.p=0.3, nb.s=3, data=data) which i think is due to the sadvm() function it calls from Fortan. direct_GCNB_AR1_likelihood can be optimised over one parameter but is SLOW for n > 8.

Author(s)

Hannah Lennon <[e-mail]>

References

Shi and Valdez, (2012), Longitudinal Modelling of Insurance Claims using Jittering.

Examples

library(mnormt)
library(mvtnorm)


n <- 4
set.seed(n)
simdata    <- arima.sim(n, model=list(ar=0.7))
data       <- qnbinom(pnorm(simdata), prob=0.3, size=3)
cov        <- ARMAacf(ar=0.7, lag.max=n-1, pacf=FALSE)

direct_GCNB_ARMA_likelihood(cov, 0.3, 3, data, method="mnormt")
direct_GCNB_AR1_likelihood(0.7,  0.3, 3, data, method="mnormt")

direct_GCNB_ARMA_likelihood(cov,  0.3, 3, data, method="mvtnorm")
direct_GCNB_AR1_likelihood(0.7,  0.3, 3, data, method="mvtnorm")

optimise(f=direct_GCNB_AR1_likelihood, interval=c(-0.9, 0.9), nb.p=0.3, nb.s=3, data=data, method="mnormt",  maximum=TRUE)

hlennon/copulaIVTS documentation built on Dec. 20, 2021, 4:45 p.m.