GARCH-in-mean

[CarlJohnsson]

Hi all,

I'm using Eviews 7 to estimate GARCH-in-mean volatility over a time series of log returns on Dutch markets. This is my model:

return = c1 - 0.5*variance(t) + error(t)
variance = c2 + c3*variance(t-1) + c4*error^2(t-1)
with error~N(0,sigma^2)

My question: what do I enter in the mean equation box? I don't know what the code is for variance(t). I know the equation should start with: return c and then some for the variance, but what?

I've searched the forum and the internet. Can anyone help? Tnx in advance!

Greetings CJ

[trubador]

First of all, I think you should read the users guide before going any further.

Since this is a simple GARCH(1,1)-M model, all you have to do is write "return c" into mean equation as you say and select the "Variance" option from the ARCH-M dropdown menu...

[CarlJohnsson]

First of all, I think you should read the users guide before going any further.

Since this is a simple GARCH(1,1)-M model, all you have to do is write "return c" into mean equation as you say and select the "Variance" option from the ARCH-M dropdown menu...

Hi Trubador,

Tnx for your quick reply.

I am new at Eviews and have been reading the manual for 2 days (seriously) but couldn't find the answer. Your argument to select the "Variance" option from the ARCH-M dropdown menu is correct but not the whole story - I think. This is because the results do not specify the variance in the return - as used in my return model.

return = c1 - 0.5*variance(t) + error(t)

I think I need to specify the 0.5*variance(t) in the mean equation box. Suggestions are welcome.

Tnx.

[CarlJohnsson]

Hi Trubador,

Could it be that ar(1) is the correct 'code' for variance(t-1) to enter in the mean equation box? But I can't get Eviews to recognize 0,5*AR(1) as my model would require.

How would you deal with that? Anyone?

Tnx.

[trubador]

Please note that, variance(t) is an unobserved variable and is estimated along the process. Therefore, I think it makes more sense not to predetermine its coefficent. You can, however, test the coefficient against 0.5 and see if they are significantly different after you have estimated the model. If you are sure about the formulation, then I am afraid you'll have model it through LogL object. You can search the forum and refer to users guide for more information on the specification and use of LogL object.

[CarlJohnsson]

Hi Trubador,

Tnx. Got the GARCH-M so far. That's the good news.

The bad news is I do have a far more complex problem involving component volatility - for which I probably need the Logl object. Do you know anyone who would be willing to help (for reward if asked)? It's for my master thesis. I get the whole theory and analysis and stuff, but the programming part is definitely not something which is learned in a few days (I think).

Perhaps you're in for a real challenge, or someone you know? I'm willing to pay for it - but I'm a student so it's not going to be very much ;)

Hope hearing from anyone.

Carl