Skip to content

Passing optim.control arima arguments to ugarchfit in rugarch

3 messages · Alexios Ghalanos, Andreas Keller Leth Laursen

#
Dear all

I am currently estimating a number of GARCH models using the excellent
rugarch package.

I am however having a small problem, as the mean equation has convergence
problems. It is a simple ARMA(6,5) model, that is being used as mean
equation.

When estimating the mean equation, I understand that ugarchfit() use the
standard arima() function. Is there a way to pass optim.control arguments
to the arima solver? A simple increase in the default maxit should solve
the problem, but I cannot figure out if there is a way to pass this
argument.

As a small hack I have tried to simply give the arma(6,5) estimates as
stating values to ugarchspec() using arima() with a higher maxit.
However, ugarchfit() still reports convergence problems in the mean
equation, when I attempt this.

Any help will be much appreciated.

Current code:

arima_start_val <- arima(input_series, order = c(6, 0, 5), optim.control =
list(maxit = 2000)) %>%
    coef %>%
    as.list

names(arima_start_val) <- c("ar1", "ar2", "ar3", "ar4", "ar5", "ar6",
"ma1", "ma2", "ma3", "ma4",     "ma5", "mu")

model <- ugarchspec(
    variance.model = list(model = "sGARCH",  garchOrder = c(1, 1)),
    mean.model = list(armaOrder = c(6, 5)),
    start.pars = arima_start_val,
    distribution = "norm")

ugarchfit(spec = model, data = input_series)

Best regards,
Andreas Keller

--
Andreas Keller
#
Hi,

arima is only used to obtain mean equation starting values for the joint 
recursion which is then ML based. Your best bet is to try using 
solver="gosolnp" which searches the parameter space for good starting 
values.
Also, try setting variance.targeting=TRUE (in variance.model), and 
fit.control=list(scale=1).

Ofcourse, this is all dependent on using a reasonable amount of data for 
the estimation (already discussed numerous times over the years).

Regards,

Alexios
On 21/01/2015 10:38, Andreas Keller Leth Laursen wrote:
#
Ah, a misunderstanding on my part. Of course it makes perfect sense that
they are jointly estimated.

Thanks for the suggestions and the swift response.

Best regards,

Andreas

--
Andreas Keller
M.Sc. student in Economics and Management
On Wed, Jan 21, 2015 at 12:19 PM, alexios <alexios at 4dscape.com> wrote: