Running an AR(p) simulation
Posted: Wed May 09, 2012 2:38 am
Dear and beloved Board members,
I am completely new to Eviews and am trying to run a simulation of an AR(1) model as an example. The idea is
to find the distribution of the AR(1) coefficient, which should be asymptotically normal for finite samples ( around 50 ) according to the theory, however
when I run the simulation I don't get this. I use a burn-in period of 100 samples and then store the coefficients estimated from the last
50 samples. Basically, simulation of 150 y-values ( for an y=0.5*y(-1)+nrnd model ) removing the first 100 samples and then estimating the AR coefficient based
on the last 50 samples. This should be simple.
The histogram of coef1 that I get is not what I expect. It should be much more nicely normally distributed. It isn't. What is wrong with the code ?
I am completely new to Eviews and am trying to run a simulation of an AR(1) model as an example. The idea is
to find the distribution of the AR(1) coefficient, which should be asymptotically normal for finite samples ( around 50 ) according to the theory, however
when I run the simulation I don't get this. I use a burn-in period of 100 samples and then store the coefficients estimated from the last
50 samples. Basically, simulation of 150 y-values ( for an y=0.5*y(-1)+nrnd model ) removing the first 100 samples and then estimating the AR coefficient based
on the last 50 samples. This should be simple.
Code: Select all
create examp24 u 1 1000
series coef1
series coef2
smpl 1 150
for !k = 1 to 1000
smpl 1 1
genr y=0
smpl 2 150
genr y = .5*y(-1)+nrnd
smpl 101 150
equation eq_a.ls y c AR(1)
coef1(!k) = c(1)
coef2(!k) = c(2)
next