[R-SIG-Finance] Loop For - ARMA+GARCH Model estimation and selection

[Andrea Bosio]

Hello,

I would like to put you the other question that I have about how to use 
loops for in R. The first is entitled “Loop For – ARMA model estimation 
and selection”. If you take a look and try to give me an answer to the 
first, I would be extremely thankful. I am an R beginner and I need R 
technical experts’ judgement to conduct my empirical analysis as 
efficiently as I can.

My second question is on how to loop over six variables (a,b,c,d,e) for 
an ARMA+GARCH model estimation.

Here is the function to loop.

USDlogreturns=diff(log(prices))

model<-ugarchspec(variance.model=list(model="fGARCH",submodel="c",garchOrder=c(d,e)),mean.model=list(armaOrder=c(a,b),include.mean=TRUE),distribution.model="f")

modelfit<-ugarchfit(spec=model, data=USDlogreturns, solver="hybrid")

Here is a description of the variables from a to f to be looped.

·ARMA orders (a,b). Both orders should be allowed to vary between 0 and 10.

·GARCH version (c). The versions should be allowed to switch from GARCH 
to AVGARCH, GJRGARCH, TGARCH, NGARCH, NAGARCH, APARCH and EGARCH. Note 
that the EGARCH version must be passed directly as “model” to the 
ugarchfit function.

·GARCH orders (d,e). The d order should be allowed to vary between 1 and 
2. The e order should be allowed to vary between 0 and 2.

·Error distribution (f). The error distribution should be allowed to 
switch from Normal Gaussian to Student-T and GED.

The objects to be returned from the loop are the model Akaike 
Information Criterion indicators, the p-values of the ADF test for 
standardized (stz) residual stationarity, the p-values of the Box-Ljung 
tests for stz residual serial uncorrelation and independence and the 
p-values of the Jarque-Bera test for stz residual normality. Here is the 
full list of the objects.

armagarchaic=infocriteria(modelfit)[1]

armagarchadftest=adf.test(modelfit at fit$z,k=trunc((length(modelfit at fit$z)-1)^(1/3)))$p.value

armagarchboxtest=Box.test(modelfit at fit$z,lag=30,fitdf=d+e,type=c("Ljung-Box"))$p.value

armagarchboxtest2=Box.test(modelfit at fit$z^2,lag=30,fitdf=d+e,type=c("Ljung-Box"))$p.value

armagarchgoftest=gof(modelfit,c(20,30,40,50))

P.S. How do you tell R to set (d+e) (i.e. the sum of ARCH and GARCH 
orders) as fitdf in the Box-Ljung test command? Is it right as I typed?

If you also believe that I should run the same model estimation in a 
simpler and efficient way, let me know on how to do that.

Thank you in advance.



---
Questa e-mail è stata controllata per individuare virus con Avast antivirus.
https://www.avast.com/antivirus

	[[alternative HTML version deleted]]