Starting values
The estimation of a GARCH model requires to optimize the likelihood. This optimization may fail in case of bad starting values. Fortunately, Alexios Ghalanos, the author of the R package rugarch
, did a great job in setting the optimization defaults such that the optimization is accurate in most of the cases. If you have a doubt, you can use the method setstart()
to try your own starting values and verify that it leads to a similar result in terms of estimated parameters and likelihood.
Here you can test this in case of the daily EUR/USD returns in EURUSDret
and assuming a constant mean standard GARCH model with Student t distribution. The model specification is available in the console as garchspec
.
This exercise is part of the course
GARCH Models in R
Exercise instructions
- Estimate the model
garchspec
using default starting values. - Print the estimated parameters and the likelihood.
- Set other starting values and re-estimate.
- Print the estimated parameters and the likelihood.
Hands-on interactive exercise
Have a go at this exercise by completing this sample code.
# Estimate using default starting values
garchfit <- ___
# Print the estimated parameters and the likelihood
___
___
# Set other starting values and re-estimate
___(garchspec) <- ___(alpha1 = 0.05, beta1 = 0.9, shape = 6)
garchfit <- ___
# Print the estimated parameters and the likelihood
___
___