[R-SIG-Finance] rugarch: solnp vs nlminb default control parameters

[alexios]

Hi Bob,

Thanks for your feedback.

This is definitely related to the handling (and scaling) of external 
regressor in variance and their constraint bounds when using the solnp 
solver. The following works:

# 1. Avoid the external regressors:
s2.spec <- ugarchspec(variance.model = list(model = "sGARCH", garchOrder 
=c(1,1)), mean.model = list(armaOrder = c(0,0), include.mean = FALSE))
fit3 = ugarchfit(s2.spec, dat, fit.control = list(scale=1))

# 2. Keep External Regressor and increase the tolerance
fit2 <- ugarchfit(s.spec, data = dat, solver = "solnp", solver.control =
list(trace = TRUE, tol=1e-12, delta=1e-11))

In a future release I might allow user adjustment of the bounds since 
the external regressors in both mean and variance equations do sometimes 
cause some problems.

HTH.

Best,

Alexios

On 31/10/2011 18:33, Robert Harlow wrote:
> Hi,
>       First off, this is a fantastic package, so very well done Alexios.  I
> am going to use the external regressor option in the rugarch package.  In
> order to do double check that everything is working properly, I decided to
> fit a standard garch(0,1) where i use the lag of the residual as my
> external regressor (this is the same as fitting a garch(1,1) model) rather
> than explicitly expressing the order in ugarchspec.  As the following code
> shows, there are different results depending on whether you use solnp or
> nlminb as the solver.  I suspect this has to do with the default control
> parameters set for solnp, as it appears solnp gets stuck near the starting
> parameters.  I don't think I have enough expertise to recommend different
> default starting parameters for the solnp solver option, but it is an
> interesting issue nonetheless.
> Below is some sample code for the sp500ret dataset.
> Thanks,
> -Bob
>
> library(rugarch)
> data(sp500ret)
> dat<- as.numeric(sp500ret[,1])
> dat<- dat - mean(dat)
> vx<- c(NA, dat[-length(dat)]^2)
> dat<- dat[-1]
> vx<- vx[-1]
> s.spec<- ugarchspec(variance.model = list(model = "sGARCH", garchOrder =
> c(0,1), external.regressors = as.matrix(vx)), mean.model = list(armaOrder =
> c(0,0), include.mean = FALSE))
> fit1<- ugarchfit(s.spec, data = dat, solver = "nlminb", solver.control =
> list(trace = TRUE))
> fit2<- ugarchfit(s.spec, data = dat, solver = "solnp", solver.control =
> list(trace = TRUE))
> par(mfrow = c(2,1))
> plot(as.ts(sigma(fit1)))
> plot(as.ts(sigma(fit2)))
>
> 	[[alternative HTML version deleted]]
>
> _______________________________________________
> R-SIG-Finance at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-finance
> -- Subscriber-posting only. If you want to post, subscribe first.
> -- Also note that this is not the r-help list where general R questions should go.
>
>