The RATS Software Forum

Hi there,

I estimated the Ar(3)-GARCH-IGARCH-FIGARCH models. But I'm not sure about those estimates.
I want to use GARCH-IGARCH-FIGARCH models in ARMA(1,1) model structure.
I couldn't understand the error in my code.

Whether any particular model will translate well from one data set to another is not clear. However, your data set has some very serious issues---the first 3000 data points look very little like the remainder of the data and there is no chance that a single GARCH model, even with some tweaks is going to fit both ranges.

TomDoan wrote:Whether any particular model will translate well from one data set to another is not clear. However, your data set has some very serious issues---the first 3000 data points look very little like the remainder of the data and there is no chance that a single GARCH model, even with some tweaks is going to fit both ranges.

Thank you for your reply.

I am sorry i couldn't understand what you mean by "the first 3000 data points look very little like the remainder of the data" . Those data are TRY/USD values and were taken from the central bank of Turkey. Can you be more specific about what you mean by "very little"?

Graph the data. The first 3000 look very little like the remainder---much lower period to period movement with a few enormous spikes. I assume there was some attempt to keep the exchange rate relatively stable over that period, with a few major corrections. If that's the case (this is your data, and you should know the history), it won't behave at all like a return series in a freely traded market.

TomDoan wrote:Graph the data. The first 3000 look very little like the remainder---much lower period to period movement with a few enormous spikes. I assume there was some attempt to keep the exchange rate relatively stable over that period, with a few major corrections. If that's the case (this is your data, and you should know the history), it won't behave at all like a return series in a freely traded market.

Thank you, i got what you mean now.

I have another problem though. When i try to apply AR(3-FIGARCH) model, i got an error about the model structure and the analysis wont work. Can you tell me whats wrong with my codes?

Again, there is nothing wrong with the code---the model doesn't work because the model appropriate for the last roughly 1/2 the data isn't for the first half. If I take your program and change the start date on the initial LINREG to

linreg r 3001 *
# constant r{1 2 3}

(which flows through the rest of the analysis), it works fine. You should examine the history of the series to see what the best entryis for defining the transition out of the (apparently) managed exchange rate ( (3001 is just a rough guess).

TomDoan wrote:Again, there is nothing wrong with the code---the model doesn't work because the model appropriate for the last roughly 1/2 the data isn't for the first half. If I take your program and change the start date on the initial LINREG to

linreg r 3001 *
# constant r{1 2 3}

(which flows through the rest of the analysis), it works fine. You should examine the history of the series to see what the best entryis for defining the transition out of the (apparently) managed exchange rate ( (3001 is just a rough guess).

I've got another problem, I want to apply AR2-GARCH(1,1) but my codes are AR3-GARCH(1,2).
I think the phi parameter is arch, beta parameter is garch. But I don't know how can i calculate AR(2)-GARCH(1,1) ?

The mean model is handled automatically when you change the LINREG here (which you've already done):

linreg rt
# constant rt{1 2}
*
frml(lastreg,vector=mu) meanf
compute omega=%seesq,delta=0.0
nonlin(parmset=meanparms) mu

The mean model parameters are in the MU vector; all the other parameters are for various parts of the GARCH variance model.

TomDoan wrote:The mean model is handled automatically when you change the LINREG here (which you've already done):

linreg rt
# constant rt{1 2}
*
frml(lastreg,vector=mu) meanf
compute omega=%seesq,delta=0.0
nonlin(parmset=meanparms) mu

The mean model parameters are in the MU vector; all the other parameters are for various parts of the GARCH variance model.

I don't understand.
What exactly are the phi and beta parameters in the formula?
What parameter should I look for the Arch effect and the garch effect?


nonlin(parmset=garchparms) omega delta beta1 phi1 phi2
compute beta1=.4,phi1=.7,phi2=0.3
*
compute uupresample=omega0
set uuadj = uupresample
set h = uupresample
*
frml varf = beta1*h{1}+omega+delta*(gap-beta1*gap{1})+$
(phi1+phi2-beta1)*uuadj{1}-phi1*phi2*uuadj{2}
frml logl = u=rt-meanf,uuadj=u^2-delta*gap,h=varf,%logdensity(h,u)
maximize(parmset=meanparms+garchparms) logl gstart gend