GARCH(1,1) or (2,2)? - only look at logL values or...?
Posted: Mon Oct 08, 2012 8:43 am
by T27667
In order to choose between a GARCH(1,1) and a (2,2), is it enough just to compare the logL values from the two estimations? If the AIC values are also important, does anyone know how to print those in either the GARCH or the PcGive module in OxMetrics?
Re: GARCH(1,1) or (2,2)? - only look at logL values or...?
Posted: Fri Oct 19, 2012 12:33 am
by JimForest
I don't know of any meaningful use of higher order garch models than Garch(1,1). In my opinion, the more interesting variations of Garch models have nothing to do with order.
The specification of the mean and variance equations and the distributional assumptions are where I tend to focus.
You'll notice that EViews allows for a number of specifications. For example, Garch-M allows the volatility term to enter the mean equation. This allows you to examine whether volatility affects the dependent variable. One example of where this would be of interest is in the area of finance, where the dependent variable is the return on an asset. (Engle, Lilien and Robins, 1987)
EViews also allows you to enter exogenous variables in the variance equation.
In some cases you might want to assume the dependent variable follows a t-distribution.