[R-SIG-Finance] GARCH convergence error in for-loop

[Hannah Linder]

Hello,

I am using the rugarch package to fit to simulated data. I am fitting the
same garch model to 1000 simulated data sets (all very similar with
slightly different error). Within each data set I am using 10-fold CV, so I
am fitting the model to 10 training sets. When I run the code (shown below)
I occasionally receive the error:

Solver Message: Error in is.nloptr(ret) : at least one element in x0 < lb

When I re-run the exact same code starting from the iteration that caused
and error, the error goes away. Does anyone know if this may be an error
with R or if I am missing a problem in the model?

Thank-you very much for your time!

Code:


library(rugarch)

fit.spec1=array(list(),c(1000,10))
fit1=array(list(),c(1000,10))


for (j in 1:1000){
  for (i in 1:10){
    fit.spec1[[j,i]]=ugarchspec(variance.model = list(model = "sGARCH",
                                       garchOrder = c(1,
1),external.regressors=as.matrix(train.all[[j]][[i]][,c(5,6)])),
                                mean.model= list(armaOrder = c(1,1),
                                                 include.mean = T,

 external.regressors=as.matrix(train.all[[j]][[i]][,c(2,4,5,6)])),
                                distribution.model = "sstd")
    fit1[[j,i]] <- ugarchfit(data=train.all[[j]][[i]][,c(1)],spec =
fit.spec1[[j,i]],solver="hybrid")
 }}

I am actually using the model on biological data, rather than finance. The
regressors in the model are  Julian day count (2), tidal range (4), and a
sin and cos (5,6) transform for time-of-day (24-hr period).

	[[alternative HTML version deleted]]