PARCH

[puf]

Hi,
when I try to estimate a PARCH model (GARCH model with option PARCH), I get the following error message: "Attempt to raise a negative number to a non integer power in "EQUATION PERCEQ.ARCH(1,1,PARCH) PERCR. The data file is addtached, the code itself is very simple:

Code: Select all

wfopen 6.csv genr logr = dlog(abs(prc)/cfacpr) genr percr = @pch(abs(prc)/cfacpr) equation logeq.ARCH(1,1,parch) logr equation perceq.ARCH(1,1,parch) percr

Surprisingly, I have this problem only with percentage returns, parch model with logaritmic returns works fine. This seems to be really strange, since percentage and relative returns are very similar.
Is there bug in estimation of PARCH model, or am I just misunderstanding something?

[EViews Gareth]

Grab the latest patch.

[puf]

Hi Gareth,
I have another problem related to GARCH models. I believe that there should be some restrictions regarding the coefficients, e.g. all the estimated coefficients in the standard GARCH(1,1) should be positive. This seemts to work. However, when I try to estimate GARCH(0,1) with exogenous variable (which is there instead of squared returns, I sometimes get negative constant omega. Is it possible to somehow specify that omega should be positive? If this is just a bug, could you please fix it? Actual code is very simple:

Code: Select all

import 22.csv genr logr = dlog(abs(prc)/cfacpr) genr park = (log(askhi/bidlo))^2/(4*log(2)) equation eq1.arch(1,1,tdist) logr equation eq2.arch(0,1,tdist) logr @ park(-1)

My concern is that eq2.c(1) is negative.
thank you

[EViews Gareth]

This isn't a bug.

[puf]

Hi Gareth,
in this case, is it possible to impose this restriction (omega>=0) in eviews?

thank
Peter

[EViews Gareth]

Unfortunately not.

[puf]

Dear Gareth,
I have problem estimating GARCH model with exogeneous variable in the variance equation. The code is again very simple:

Code: Select all

wfopen 2.csv series park = (log(askhi/bidlo))^2/(4*log(2)) series logr = dlog(abs(prc)/cfacpr) scalar length = @obsrange scalar estimation_window = 200 for !i=estimation_window to length-2 smpl @first+!i-estimation_window @first+!i equation kamo.ARCH(0,1,tdist) logr @ park(-1) next

I get the error message:
"Positive or nonnegative argument to function expected in "equation kamo.ARCH(0,1,tdist) logr @ park(-1)"
I believe that there is some problem with the optimization routine. Let me give you some comments which might be useful:
1. this is not a problem with normally distributed residuals.
2. estimation of the model with t-distributed residuals does not work only sometimes, typically it works.
However, since I must do calculations in such a way as code describes (to perform rolling forecast), it is a problem. Please, could you fix this problem?

[puf]

Hi,
in addition to previous unsolved problem, I have additional one: estimation of the PARCH model with GED distributed errors does not work. This simple code:

Code: Select all

wfopen 6.csv genr logr = dlog(abs(prc)/cfacpr) equation eq.ARCH(parch,ged) logr

yields this error message:
"Attempt to raise a negative number to a non integer power in "EQUATION EQ.ARCH(PARCH,GED) LOGR".
This model works with normally and student-t distributed errors, but not with GED distributed errors.

Could you please fix this problem? (preferably the previously mentioned problem too).
thank you

[EViews Gareth]

Sorry, must have missed your previous one. I've found and fixed both, and hopefully the next patch will fix it for you.