GARCH Model doubts

[gastonpresente]

Im planning the typical exercise of returns (on exchange rates) and news coming into the market. Im planning to use a GARCH (1,1). And the news variable is news paper headlines with the phrases "crisis, etc". Im planning to use dummy variables. I have two questions:

1) alpha + beta is more than one. Thats a condition to have a mean reverting variance process. But im using exchange rates. Any help on this point?
2) Im using intraday data (5 minutes intervals). And i want to introduce the information of daily newspapers, and i read a couple of papers that information impacts in the first 15 to 30 minutes. So i dont know how to introduce the dummys, or if i have to use a dummy? Sound weird to put in the first minutes 1s and the last intervales of the day 0s. Again any help on this?

Thanks a lot

[JimForest]

I would consider stepping back from intraday to daily as you may pick up the ill effects of intraday noise. Perhaps intraday returns as based on change between the open and close.

You might want to check out the following paper:

Macroeconomic News and Bond Market Volatility by Jones, Lamont, and Lumsdaine in JFE 1998 as they show a technique for specifying a dummy to capture the volatility effect. I have been able to run the first GARCH model but i am having trouble getting eviews to handle the regime switching model to evaluate volatility persistence.

[gastonpresente]

Hey Jim,

Thanks for your comment.

You mean that the reason why im getting an alpha + beta bigger than 1 is caused by the intraday noise?
Thats probably why i get an R2,, negative? and R2 negative means that the curve to fit is worst than a horizontal straight line?
im going to try to get that paper in the univesitty.. thanks for the tip.

Gaston

[JimForest]

Hey Jim,

Thanks for your comment.

You mean that the reason why im getting an alpha + beta bigger than 1 is caused by the intraday noise?
Thats probably why i get an R2,, negative? and R2 negative means that the curve to fit is worst than a horizontal straight line?
im going to try to get that paper in the univesitty.. thanks for the tip.

Gaston

The R^2 being negative is meaningless in that it only applies to the mean equation. I believe the alpha + beta above 1 indicates the conditional variance process is explosive. What does the graph of its forecast look like? Does it converge to either the unconditional variance or infinity?