GARCH (1,1) Results

[rtlee100]

Hi,

I have a question relating to some GARCH analysis that I am undertaking and am using weekly and monthly data. When I model the weekly data the results conform to as I would expect and align closely to results that Matlab will give, however, when I model the monthly data the parameters estimted are spurious and I can't seem to figure out why the resulting parameters are so incorrect. When I use the same data in Matlab I get results that would be expected. I was wondering if anybody has come across the same issuue?

As I am trying to model a SWARCH model for weekly and monthly data the above results would lead me to wonder if the monthly results for the SWARCH model are equally spurious.

Results for Weekly data...

GARCH Model - Estimation by BFGS
Convergence in 20 Iterations. Final criterion was 0.0000027 < 0.0000100
Weekly Data From 1983:05:27 To 2009:07:17
Usable Observations 1365
Log Likelihood 2230.95072179

Variable Coeff Std Error T-Stat Signif
*******************************************************************************
1. Mean 0.0003798312 0.0010405130 0.36504 0.71507989
2. C 0.0000657066 0.0000269600 2.43719 0.01480184
3. A 0.1995133158 0.0260659727 7.65417 0.00000000
4. B 0.8011515432 0.0251262394 31.88506 0.00000000

Curriously the results of A+B add to greater than 1 (i.e. 1.000664859)?

Results for monthly data...

GARCH Model - Estimation by BFGS
Convergence in 14 Iterations. Final criterion was 0.0000012 < 0.0000100
Monthly Data From 1983:06 To 2009:06
Usable Observations 313
Log Likelihood 310.14575867

Variable Coeff Std Error T-Stat Signif
*******************************************************************************
1. Mean 0.009627324 0.004526986 2.12665 0.03344903
2. C 0.005572297 0.001153332 4.83148 0.00000136
3. A 0.504862518 0.116844941 4.32079 0.00001555
4. B -0.056441103 0.112846948 -0.50016 0.61696508

I am running version 6.10.

Thanks,

Rob.
Thanks.

[TomDoan]

Try giving it the initial guess values from the other GARCH model: add the option INITIAL=||.0004,.00007,.2,.8|| to your quarterly data GARCH. We changed our guess values with a later version of RATS to avoid some of these problems with local modes.

Regarding the sum of coefficients being just barely above one - that would seem to indicate that an IGARCH model is appropriate.