cenarma: Censored arma estimation

In cents: Censored time series

Description

A quasi-EM algorithm is implemented. The R function arima() is used in the maximization step. The Durbin-Levinson recursions are used to compute conditional expectations.

Usage

cenarma(y, iy, p=0, q=0, include.mean = TRUE, verbose = FALSE, 
  MaxIter = 100, ETOL = 1e-05, algorithm = c("exact", "approx"), ...)

Arguments

y	time series as a vector of length n
iy	indicator with entries: "o","L","R","na". If missing, it is assumed there is no censoring and iy entries are set to "o" or "na" according to whether the corresponding value in y is numeric or NA.
p	ar order
q	ma order
include.mean	Default is to estimate the mean. FALSE means we assume the mean is zero.
verbose	If true, show successive log-likelihoods
MaxIter	maximum number of iterations
ETOL	error tolerance
algorithm	"exact" uses our tacvfARMA() and approximate uses acfARMA()
...	options passed to arima

Value

fitted model out is a list:

Arima	the output for the function arima()
v0	covariance matrix of the parameters
dataSummary	number of data values in each class
exitStatus	"converged" or "Maxit iterations reached"

Examples

#Default example
#Example. Left-censoring, 10%
## Not run: 
set.seed(313177)
n <- 500
out <- rcarma(n, ar=0.8, ma=-0.6, mu=100, siga=15, rates=c(0.1, NA))
y <- out$y
iy <- out$iy
ans <- cenarma(y, iy, p=1, q=1)
ans[[1]]
#
#Example ARMA(1,1) with missing values.
#Fit using arima() and cenarma()
#compare final relative likelihood and difference log-likelihoods
set.seed(313177)
n <- 500
out <- rcarma(n, ar=0.8, ma=-0.6, mu=100, siga=15, rates=c(NA, NA), Mrate=0.25)
y <- out$y
iy <- out$iy
(ans0 <- arima(y, order=c(1,0,1)))
(ans1 <- cenarma(y, iy, p=1, q=1))[[1]]
logL0 <- ans0$loglik
betaHat <- coef(ans1[[1]])
arHat <- betaHat[1]
maHat <- betaHat[2]
muHat <- betaHat[3]
ans1B <- arima(y, order=c(1,0,1), fixed=c(arHat,maHat,muHat),transform.pars=FALSE)
logL1 <- ans1B$loglik
RL <- exp(logL1-logL0)
RL
logL1-logL0

## End(Not run)

cents documentation built on May 1, 2019, 8:19 p.m.