Re: st: RE: rolling regression and hypothesis testing

[Nick Cox <njcoxstata@gmail.com>]

This is a long way from a program that will work with -rolling-. As
the help for -rolling- explains, it picks up results from a called
program. Those results can take various forms, but results put in
variables are not among them. More positively, Kit Baum replied to
this thread earlier indicating a likely program that he has written.
Why don't you start with that, at least as a template?

Nick

On Tue, Apr 5, 2011 at 9:01 PM, Nat Tharnpanich <nt289@cam.ac.uk> wrote:

> I have been trying to write the program that -rolling- can call in order to
> do regression and hypothesis testing in one go. However, I am having a
> problem in the hypothesis testing part as I want to test the significance of
> the difference between an estimated coefficient and a variable, say, Z which
> takes on a value that differs by year. So I have rolling windows(20) clear,
> progname and the program is as follows, program progname
>  reg Yt X1t X2t
>  test X1t = Zt
>  gen p = r(p) // to generate a variable for p values
>  gen f = r(F) // to generate a variable for F statistics
> end and t is year. If Zt is a constant, then presumably my program would
> work fine. However, since the value of Zt varies from from year to year,
> Stata interprets Zt as one of the regressors which is always not found
> because -rolling- command will not keep Zt when it is executed. I have spent
> all day and night to figure out how to write the correct code, but still
> completely clueless. I would greatly appreciate any advice on this. Thank
> you very much. Nat
>
> On Apr 4 2011, Nick Cox wrote:
>
>> There is no contradiction here, as you can write a program that does your
>> regression and testing and then that can be the single command that
>> -rolling- calls.
>>
>> However, a big, big, big caveat is that these tests are manifestly not
>> independent, so goodness know how you interpret them.
>>
>> Nick n.j.cox@durham.ac.uk
>> Nat Tharnpanich
>>
>>   I would like to run rolling OLS regression of, say, overlapping 20-year
>> periods and test a hypothesis whether the coefficient is statistically
>> different from 1. As far as I know, the rolling command in Stata allows you
>> to execute only one command at a time. But in my case, I want to run a
>> regression of each subperiod and then test the hypothesis for that subperiod
>> before Stata goes on to subsequent subperiods and this involves two
>> commands, reg and test.

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/