Em 25 de jul de 2018, ?(s) 22:16, Alexios Galanos <alexios at 4dscape.com> escreveu:
1. rmgarch does not support varma, only VAR
2. The model you estimated is univariate ARMA(1,1)
3. The NAs are because you set fit.control = list(eval.se <http://eval.se/> = FALSE)
i.e. you are telling the routine to not evaluate the standard errors.
Alexios
On Jul 25, 2018, at 16:08, Marcio Bernardo <marciobernardo1 at gmail.com <mailto:marciobernardo1 at gmail.com>> wrote:
Hi,
I was wondering if the current rmgarch version allows for a VARMA-GARCH modeling.
I tried forcing the issue, changing the rmgarch example:
uspec.n = multispec(replicate(30, ugarchspec(mean.model = list(armaOrder = c(1,1)))))
spec.dccn = dccspec(uspec.n, dccOrder = c(1, 1), distribution = 'mvnorm')
fit.1 = dccfit(spec.dccn, data = X, solver = 'solnp', cluster = cl, fit.control = list(eval.se <http://eval.se/> = FALSE))
but the results were a bit off:
*---------------------------------*
* DCC GARCH Fit *
*---------------------------------*
Distribution : mvnorm
Model : DCC(1,1)
No. Parameters : 617
[VAR GARCH DCC UncQ] : [0+180+2+435]
No. Series : 30
No. Obs. : 1141
Log-Likelihood : 70882.93
Av.Log-Likelihood : 62.12
Optimal Parameters
-----------------------------------
Estimate Std. Error t value Pr(>|t|)
[AA].mu 0.002643 NA NA NA
[AA].ar1 -0.693738 NA NA NA
[AA].ma1 0.664589 NA NA NA
[AA].omega 0.000065 NA NA NA
[AA].alpha1 0.115044 NA NA NA
[AA].beta1 0.869706 NA NA NA
[AXP].mu 0.002737 NA NA NA
[AXP].ar1 0.072418 NA NA NA
[AXP].ma1 -0.150777 NA NA NA
[AXP].omega 0.000011 NA NA NA
[AXP].alpha1 0.064777 NA NA NA
[AXP].beta1 0.934223 NA NA NA
.
.
.
[Joint]dcca1 0.004960 NA NA NA
[Joint]dccb1 0.942361 NA NA NA
Information Criteria
---------------------
Akaike -123.17
Bayes -120.44
Shibata -123.51
Hannan-Quinn -122.14
This seems to be a ARMA-DCC fit, instead of VARMA-DCC and the NA is troubling me.
I understand the package support VAR-Garch. Is there any package currently available in R that have VARMA-Garch model (I don?t have access to RATS)?
Any help would be much appreciated,
M?rcio R. Bernardo