EViews.com

Parameter EstimationEviews Matlab
omega -0.11960 -0.21089 -0.12456
g 0.11307 0.11265 0.11697
a -0.08916 -0.08991 -0.089768
b 0.98719 0.98698 0.98662
LT Vol 14.83% 0.48% 15.05%

The parameter estimates above were made for a daily FTSE100 (01/03/1995 to 08/29/2007) using Excel, Eviews and Matlab by the Carol Alexander in Practical Financial Econometrics Vol II. I have repeated the analysis using the EGARCH method getting the same results which as pointed out by Alexander seem inappropriate for the Eviews calculation. Please let me know if you have realised this and does it affect the calcs for EGARCH across the program.

Brett Ifill

I'm not sure I follow what you are saying, exactly.

You've estimated the same EGARCH model with three different software packages, using the same data (I presume) and achieved three different results, with the EViews result standing out a little more from the other two?

Do you know what the starting values used in the other packages were? EGARCH is highly nonlinear and your model is probably very sensitive to starting values.

EViews use an exponential smooth method to calculate the presample variance, while most other programs use the unconditional variance. My guess is this difference is the cause of the discrepancy between EViews and other GARCH programs. This hypothesis is easily testable—use the unconditional variance as the presample variance when estimating your EViews EGARCH model (setable under the options tab). (See the Backcasting section in the GARCH chapater for additional information.)

Thanks for your replies, they helped. Gareth, you pointed out that the starting values could be critical, so I am reviewing this. Can I set starting values from within the package or should I write a program?

Starting Values methods can be chosen on the options tag of the estimation dialog. You can also tell EViews to display the starting values options on the equation output.