PRELIMINARY ANALYSIS OF RETURNS DATA

[Osabuohien247]

Hi Tom,
Thanks for your comment. I can see that the data is flat. Does non Convergence of the GMM denote Lack of Arch Effect as suggested by the MCLEODI VALUES?
Most Importantly, How does one determine the lags value as used in page 20 - 22 of your e-course "ARCH/GARCH VOLATILITY MODELS", as shown below;

nlsystem(robust,lags=4,lwindow=newey,parmset=meanparms,inst) 2 * $
f1 f2 f3 f4

nlsystem(robust,lags=4,lwindow=newey,parmset=meanparms,inst) 2 * $
f1 f2

Thanks brother.

[TomDoan]

The GMM estimates don't converge. Did you check your data? The raw data are dead flat for almost half the the range. Are those already in logs? (If so, they're really scary).

[Osabuohien247]

Hi Tom,
Thanks for your comment. I can see that the data is flat. Does non Convergence of the GMM denote Lack of Arch Effect as suggested by the MCLEODI VALUES?
Most Importantly, How does one determine the lags value as used in page 20 - 22 of your e-course "ARCH/GARCH VOLATILITY MODELS", as shown below;

nlsystem(robust,lags=4,lwindow=newey,parmset=meanparms,inst) 2 * $
f1 f2 f3 f4

nlsystem(robust,lags=4,lwindow=newey,parmset=meanparms,inst) 2 * $
f1 f2

Thanks brother.

[TomDoan]

You edited away your original question, so I'll have to remember what was up with that.

Before you do any empirical work, you need to understand your data and what it means for what you're doing. I'm guessing that your data are exchange rates or something like it, and they were pegged for part of the sample. Pegged exchange rates don't follow a GARCH process. You need to take the part of data set which is pegged out of your analysis.

I also recall that your data transformation was a simple difference of your data. That's why I asked about whether they were in logs already. I'm guessing they aren't since otherwise they would indicate roughly a 10^70 increase over the course of the sample. You need to transform into log first differences to create returns. The inability of the NLSYSTEM to converge was because you were, in effect, using only a tiny part of the sample to estimate the moments since they were dominated by the last few observations. It says nothing about GARCH or not; it says that your data aren't prepared correctly.

[Osabuohien247]

Thanks my brother. It is clear now. I have done it. There is convergence in the GMM Estimate and the mean value tally. You are really the best. Please i want to know whether the lags value in the nlsystem is always 4 in this code. Or how can it be determine. Thanks