Dear all,
I have posted a query on simulating a VAR (Vector Auto-regression) model on
R-help, however still unable to get some satisfactory reply. That post can
be found here
"http://n4.nabble.com/Simulation-of-VAR-td1693295.html#a1693295". Therefore
I am re-posting the same here as well. Hopefully this query will be able to
fetch some educated attention from the experts here, who are presumably more
into time series analysis/finance.
This is my post :
I want to simulate a VAR model on some estimated coefficients. And employed
two approaches, one is my native approach completely based on the my
understanding on VAR and second based "simulate" function from "dse"
package. Surprisingly those two are giving quite different results, although
they are based on same set of random numbers. Here is that :
library(dse)
A1 <- matrix(rnorm(16),4)
A2 <- matrix(rnorm(16),4)
mu <- rnorm(4)
sigma <- matrix(c(0.006594712,
0.006467731,
-0.000254914,
0.005939934,
0.006467731,
0.006654184,
-0.000384097,
0.005658247,
-0.000254914,
-0.000384097,
0.000310294,
4.34141E-05,
0.005939934,
0.005658247,
4.34141E-05,
0.00574024), 4)
initial.val <- matrix(c(-0.2347096,
-0.1803612,
-0.2780356,
-0.2154427 ,
3.740364,
3.757908,
3.50216 ,
3.57783), 2)
##### My approach
res <- matrix(NA, 4,4); res[c(1,2),] <- initial.val
library(mnormt); shocks <- rmnorm(2, rep(0,4), sigma)
for (i in 1:2) {
res[i+2,] <- mu + A1%*%res[i+2-1,] + A2%*%res[i+2-2,] + shocks[i,] }
res
##### dse approach
temp1 <- matrix(t(cbind(diag(4), A1, A2)), ncol = 4, byrow = TRUE)
model <- ARMA(A=array(temp1, c(3,4,4)), B=diag(4), TREND=mu)
simulate(model, y0=initial.val, sampleT=2, noise=shocks)
So I guess something terribly going wrong in any of the approaches in terms
of my understanding. Can someone here point out what is that, if any? Your
help will be highly appreciated.
Thanks,