On Mon, 21 Aug 2006, Arnaud Bervas wrote:
I wanted to know how you can get HAC standard errors once you
have sub-sampled the dataset, since sub-sampling transforms
the initial time-series dataset into a cross-sectional one ?
For example:
open data9-7
smpl QNC > 2200 --restrict
ols 1 0 2 3 --robust
setobs 4 1960:1
ols 1 0 2 3 --robust
The first ols does not use HAC, because the dataset is not
considered time series. The second ols does use HAC, by
default (although this is of questionable validity, since the
data are no longer contiguous time-series observations).
The action of the "setobs" line is the same as selecting
"Dataset structure" from the menus, and setting the data
as quarterly time series.
Allin Cottrell