[Gretl-users] Heteroskedasticity / correction

Allin Cottrell cottrell at wfu.edu
Fri Jan 10 16:26:08 EST 2014


On Fri, 10 Jan 2014, Ana Amaro ISG wrote:

> The test of this white Regression model does not have a constant...
> Right?

Right, and that was a bug, and it's now fixed in CVS and the 
snapshots for Windows as OS X.

However, a puzzle remains: how are you arriving at the idea that the 
constant "disappears" from the original model when you divide the 
data by sqrt(x) to compensate for heteroskedasticity?

To get the adjusted constant you would do something like

series adjc = const / sqrt(x)

The result (adjc) will "disappear" (go to zero) only if the sqrt(x) 
values are infinite, and I suppose it's safe to assume they are 
actually finite.

Allin Cottrell


More information about the Gretl-users mailing list