EViews.com

Hey EV gurus,

Relatively new to EViews, so I appreciate your time and patience. I am using EViews 7.1.

I have a workfile of 13 countries' bond yields for which I am attempting to model volatility dynamics using GARCH-M. The data ranges from 1/1/1990 to 2/29/2012 for a total of about 5800 obs.

Since ARCH estimation requires a continuous sample and complete observations don't begin until 1996, I reduced the data range to begin then, yielding a total of ~4200 obs. There are a couple hundred observations which contain NA's for one or more of the countries after this period, so I set the sample to omit these using "smpl if austria<>na and belgium<>na and..." for all countries. I opened all 13 countries as a group to verify there are no NA's, which there are not.

Then, I open Estimate Equation, select ARCH, specify my conditional mean equation "greece c germany gc gc*germany gdg gdg*germany", where gc and gdg are dummy variables, specify the log(var) option, and run.

The error I get back is "ARCH estimation requires a continuous sample." I attempted to coerce the data into irregular form using "Dated-specify by Date series", selecting "date" (object containing the date information), click OK, to which it gives the message "Date series has no duplicates or gaps. Converting to a regular frequency."

The data series contains no missing observations, occurs at an irregular frequency, yet attempts to coerce it as such fail.

Any thoughts?

What's the specification of your equation? (i.e. what are the variables etc..)

Base equation:
Rd,t = ß0 + ß1Rg,t + ε,t

,where Rd is the domestic yield @ time t and Rg is the German yield @ time t

Taking the first difference of yield levels to satisfy unit-root stationarity and adding exogenous dummy variables for the global crisis and Greek/Spanish government debt downgrades, and including interaction variables to gauge the sensitivity of a change in the average daily yield in the domestic country d to the global crisis/debt downgrades, the conditional mean equation becomes:

rd,t = ß0 + ß1rg,t + ß2GCt + ß3(rg,t*GC,t) + ß4GDG,t + ß5(rg,t*GDG,t) + log(σ2,t-1) + ε,t

where the in-mean GARCH term is included to measure risk premium of domestic bond yield.

The conditional variance equation is then:

σ2,t = ω + α1 ε2,t-1 +…+ αp ε2,t-p + ϒ1σ2,t-1 + ϒqσ2,t-q

where p and q are the respective ARCH and GARCH orders.

I would like to use a GED parameter to account for the leptokurtosis.


Cheers

So the problem is the NAs in the Greece and Germany series. You'll have to remove them from the workfile. The best way to do this is to:

• Click on Proc->Copy/Extract from Current Page... -> By Value to New Page or Workfile
• In the sample box enter:

  Code: Select all

  if greece<>na and germany<>na
• Hit OK.

Then on this new page you can estimate ARCH properly. Doing ARCH-M appears to be problematic with your data, but I was able to estimate an ARCH properly.

Thank you, Garth, for coming through in a big way.

I'm having some convergence issues myself. Aside from the strange things going on in the Greek data, any ideas as to what it could be causing issues with implementing ARCH-M? I'll try using data from a "normal" country and see if that changes anything.


Thanks again.