marima.sim {marima}

marima.sim

Description

Simulation of multivariate arma model of type 'marima'.

Usage

marima.sim(kvar = 1, ar.model = NULL, ar.dif = NULL, ma.model = NULL,
  averages = rep(0, kvar), resid.cov = diag(kvar), seed = NULL,
  nstart = 0, nsim = 0)

Arguments

kvar

dimension of one observation (from kvar-variate time series).

ar.model

array holding the autoregressive part of model, organised as in the marima$ar.estimates. May be empty (default = NULL) when there is no autoregressive part.

ar.dif

array holding differencing polynomium of model, typically generated by applying the function define.dif. May be empty (default = NULL) when differencing is not included.

ma.model

array holding the moving average part of model, organised as in the marima$ma.estimates. May be empty (default = NULL) when there is no moving average part.

averages

vector holding the kvar averages of the variables in the simulated series.

resid.cov

(kvar x kvar) innovation covariance matrix.

seed

seed for random number generator (set.seed(seed)). If the seed is set by the user, the random number generator is initialised. If seed is not set no initialisation is done.

nstart

number of extra observations in the start of the simulated series to be left out before returning. If nstart=0 in calling marima.sim a suitable value is computed (see code).

nsim

length of (final) simulated series.

Value

Simulated kvar variate time series of length = nsim.

Examples

library(marima)
data(austr)
old.data <- t(austr)[, 1:83]
Model2   <- define.model(kvar=7, ar=c(1), ma=c(1),
                    rem.var=c(1, 6, 7), indep=NULL)
Marima2  <- marima(old.data, means=1, ar.pattern=Model2$ar.pattern, 
 ma.pattern=Model2$ma.pattern, Check=FALSE, Plot="none", penalty=4)

resid.cov  <- Marima2$resid.cov
averages   <- Marima2$averages
        ar <- Marima2$ar.estimates
        ma <- Marima2$ma.estimates

N    <- 1000
kvar <- 7

y.sim <- marima.sim(kvar = kvar, ar.model = ar, ma.model = ma, 
  seed = 4711, averages = averages, resid.cov = resid.cov, nsim = N)

# Now simulate from model identified by marima (model=Marima2).
# The relevant ar and ma patterns are saved in 
# Marima2$out.ar.pattern and Marima2$out.ma.pattern, respectively: 

Marima.sim <- marima( t(y.sim), means=1, 
     ar.pattern=Marima2$out.ar.pattern, 
     ma.pattern=Marima2$out.ma.pattern, 
     Check=FALSE, Plot="none", penalty=0) 

cat("Comparison of simulation model and estimates", 
" from simulated data. \n")
   round(Marima2$ar.estimates[, , 2], 4)
round(Marima.sim$ar.estimates[, , 2], 4)

   round(Marima2$ma.estimates[, , 2], 4)
round(Marima.sim$ma.estimates[, , 2], 4)