EViews.com

Posted: **Tue Jun 23, 2015 3:53 pm**

Hi Eviews Gareth

I am estimating a Garch rolling regression with a moving window of 126 observations and a step size of 20.6. I have used the following code in Eviews 8 and I got some decent results, however when I use the same code in Eviews 9, I get the same value for each coefficient for the whole sample problem. Do you have any idea of what could be the problem with my code?

Code: Select all

'run rolling regression
' set window size
!window = 126 
' set step size
!step = 20.6 
' get size of workfile
!length = @obsrange 
' declare equation for estimation
equation eq1

'calculate number of rolls
!nrolls = @round((!length-!window)/!step)

'matrix to store coefficient estimates
matrix(9,!nrolls) coefmat ' where 9 is the number of coefficients
matrix(9,!nrolls) tmat
'variable keeping track of how many rolls we've done
!j=0

' move sample !step obs at a time
for !i = 1  to  !length-!window+1-!step step !step
!j=!j+1

   ' set sample to estimation period
   ' estimate equation
sample s0 @first @first+1
sample s1 @first+2+!i-1  @first+!i-1+!window-2
smpl @all

' declare coef vectors to use in ARCH likelihood
coef(4) beta = 0
coef(3) alpha = 0
coef(1) dof = 0
coef(1) mu = 0

' get starting values
equation eq1.arch(1,1,ged,archm=sd,backcast=1) y c y(-1)
eq1.makegarch garchm
equation eq1.arch(1,1,ged,archm=sd,backcast=1) y c garchm*y(-1) y(-1)
beta(1) = eq1.c(1)
beta(2) = eq1.c(2)
beta(3) = eq1.c(3)
beta(4) = eq1.c(4)
alpha(1) = eq1.c(5)
alpha(2) = eq1.c(6)
alpha(3) = eq1.c(7)
dof(1) = eq1.c(8)
mu(1) = eq1.@se^2

' set presample values of expressions in logl
series sig2 = mu(1)
series resid = resid

' set up GARCH likelihood
logl ll1
ll1.append @logl logl
ll1.append @byeqn
ll1.append resid = y - beta(2) - beta(1)*sig2 - beta(3)*sig2*y(-1)- beta(4)*y(-1)
ll1.append sig2 = alpha(1)+ alpha(2)*resid(-1)^2 + alpha(3)*sig2(-1)
ll1.append z = resid/@sqrt(sig2)
ll1.append logl = log(@dged(z,dof(1))) - log(sig2)/2

' declare coef vectors to use in ARCH likelihood
coef(4) beta = 0
coef(1) thresh = 0
coef(3) alpha = 0
coef(1) dof = 0
coef(1) mu = 0

' get starting values
eq1.arch(1,1,thrsh=1,archm=sd,ged,backcast=1) y c y(-1)
eq1.makegarch garchm
eq1.arch(1,1,thrsh=1,archm=sd,ged,backcast=1) y c garchm*y(-1) y(-1)
beta(1) = eq1.c(1)
beta(2) = eq1.c(2)
beta(3) = eq1.c(3)
beta(4) = eq1.c(4)
alpha(1) = eq1.c(5)
alpha(2) = eq1.c(6)
thresh(1) = eq1.c(7)
alpha(3) = eq1.c(8)
dof(1) = eq1.c(9)
mu(1) = eq1.@se^2

' set presample values of expressions in logl
series sig2 = mu(1)
series resid = resid

' set up GARCH likelihood
logl ll1
ll1.append @logl logl
ll1.append @byeqn
ll1.append resid = y - beta(2) - beta(1)*sig2 - beta(3)*sig2*y(-1)- beta(4)*y(-1)
ll1.append sig2 = alpha(1)+ alpha(2)*resid(-1)^2 + alpha(3)*sig2(-1) + thresh(1)*resid(-1)^2*(resid(-1)<0)
ll1.append z = resid/@sqrt(sig2)
ll1.append logl = log(@dged(z,dof(1))) - log(sig2)/2
' estimate and display results
smpl s1
ll1.ml(showopts, m=1000, c=1e-5)
show ll1.output
delete s0 s1

'store coefficients
   colplace(coefmat,ll1.@coefs,!j)
   colplace(tmat,ll1.@tstats,!j)

next
show coefmat
show tmat

Thank you.

Posted: **Wed Jun 24, 2015 7:08 am**

How can you have a step size of 20.6?

Posted: **Wed Jun 24, 2015 1:47 pm**

Thank you Gareth for your response, the issue is I intend to have 115 rolls which translate into 115 monthly(rolling) coefficients over the sample period. So I used 20.6, which I now understand that it is mispecified, so that the number of rolls

Code: Select all

!nrolls = @round((!length-!window)/!step)

will equal to 115. I did not know if it is possible to simply state the actual number of rolls without specifying it as it is in the code. I have tried to change the step size to 21 and now I get 113 rolls, I have also tried to change the number of rolls to 115 but I get 113 filled columns and the remaining 2 are filled with zeros. The code still gives me the same value for each coefficient over the sample period. Please help me on how I can best solve this.

Posted: **Wed Jun 24, 2015 2:27 pm**

Two possible causes for the identical coefficients:

Your estimates are different in each iteration, but you're not recording them properly
Your estimates are the same in each iteration.

I would just stop the program randomly half way through and look at the estimation output and check what the estimates are. That should tell you which of the two cases it is. If the estimates are different in each iteration, you can take a look at the code that's recording them. If the estimates are the same in each iteration, either you're not changing the sample, or something funky is going on.

Posted: **Thu Jun 25, 2015 5:08 am**

Thank you for those suggestions. I have tried to stop the program halfway, several times and noted that the sample is changing but the coefficients alternate between these two values -0.052891 and -0.056721 over the sample period or it could be a case of a string of the same value (e.g -0.052891) and then alternates (e.g to -0.056721). I have also looked at this part of the code where the coefficients are recorded

Code: Select all

colplace(coefmat,ll1.@coefs,!j)
colplace(tmat,ll1.@tstats,!j)

and tried to change where the estimates are stored from i.e either eq1 or ll1. The coefficients from eq1 are negative and statistically significant whereas those from ll1 are still negative but insignificant and the coefficient values are similar in both cases. May you please check if my whole code is right. Thank you.

EViews.com

Rolling Garch code in Eviews 9

Rolling Garch code in Eviews 9

Re: Rolling Garch code in Eviews 9

Re: Rolling Garch code in Eviews 9

Re: Rolling Garch code in Eviews 9

Re: Rolling Garch code in Eviews 9