On Sun, 29 Sep 2019, Sven Schreiber wrote:
Am 29.09.2019 um 20:08 schrieb Allin Cottrell:
> On Sun, 29 Sep 2019, Sven Schreiber wrote:
>> omit const # error
>> This should work, no?
> It's not specifically "omit const" that's not working, it's
> "omit" on the sole regressor (whatever it may be) in the prior model,
> thereby leaving no model at all.
> Why would one want to do that?
Well, there are several use cases:
- In this case it's a regression-based test for a non-zero mean of the
LHS series, which is a standard testing application. I know there are
other ways to perform such a test, e.g. the explicit t-stat that I also
used in the example.
Actually, I stumbled over this thing while implementing a forecasting
test for FEP, where a certain constructed series has zero mean under the
null. The nice thing about using a regression is that you also get
robustified inference basically for free.
- Or think about a bivariate regression model for standardized series
without a constant term a priori. You still might want to test for the
influence of the only RHS variable.
OK, but these are just use cases for determining the statistical
significance of a single (unique) regressor. I agree that task is
not that uncommon, but it's easily accomplished without use of
If you take a look at the help for omit it's pretty clear (I think)
that it's not intended for that purpose: "It calculates a Wald test
for the joint significance of the variables in varlist, which should
be a subset of the independent variables in the model last
estimated." The subset can have just one member, but it's expected
to be a proper subset. And "By default the restricted model is
estimated and it replaces the original as the 'current model'".