Re: st: Package -ghansen- now available in SSC

[Jorge Eduardo Pérez Pérez <perez.jorge@ur.edu.co>]

What you are rejecting is the null of no cointegration against the
alternative of cointegration with a shift in the mean in 1979. As
Gregory and Hansen point out in their Journal of Econometrics paper
(different from the one you are reffering, see the help file of
-ghansen-) this is not a test of whether there is a break or not. You
should test for cointegration without breaks first (i.e, using
-vecrank-) and if you do not reject the null of cointegration, then
you should use the -ghansen- test.

The second output you provided is the same as the first one, are you
sure you are using differenced variables in that case?

If you find cointegration and want to estimate a structural model, you
should estimate a SVEC model.

Regards,
_______________________
Jorge Eduardo Pérez Pérez




On Sat, Sep 3, 2011 at 12:11 AM, Muhammad Anees <anees@aneconomist.com> wrote:
> Thanks to Nick for his suggestions and Special thanks to Jorge for
> providing his updated routine and offline support, I have updated the
> -ghansen- on my system and it successfuly did its job. I have
> estimated the Gregory & Hansesn (1996) test and obtained the following
> results.
>
> Now that I have estimated my time series for possible structural
> break, I wanted to compare my results with the results of their paper
> "Gregory, A.W., Nason, J.M., and Watt, D.G. (1996), “Testing for
> structural breaks in cointegrated relationships,” Journal of
> Econometrics, 71, 321–341.". Only a slight direction is needed please
> to confirm my conclusion of rejecting the null of no structural break
> has been rejected. It would be wise to estimate a structural VAR for
> the time series I have?
>
> My results using the level variable are:
>
> . ghansen  c y z e t, break(level) lagmethod(aic) maxlags(5)
>
> Gregory-Hansen Test for Cointegration with Regime Shifts
> Model: Change in Level                             Number of obs   =    36
> Lags  =  0  chosen by Akaike criterion             Maximum Lags    =    5
>
> Test       Breakpoint   Date        Asymptotic Critical Values
> Statistic                            1%           5%    10%
>
> ADF       -8.06          8        1979      -6.05        -5.56  -5.31
> Zt        -8.17          8        1979      -6.05        -5.56  -5.31
> Za       -47.01          8        1979     -70.18       -59.40  -54.38
>
> and differenced variable as:
>
> . ghansen  c y z e t, break(level) lagmethod(aic) maxlags(5)
>
> Gregory-Hansen Test for Cointegration with Regime Shifts
> Model: Change in Level                             Number of obs   =    36
> Lags  =  0  chosen by Akaike criterion             Maximum Lags    =    5
>
> Test       Breakpoint   Date        Asymptotic Critical Values
> Statistic                            1%           5%    10%
>
> ADF       -8.06          8        1979      -6.05        -5.56  -5.31
> Zt        -8.17          8        1979      -6.05        -5.56  -5.31
> Za       -47.01          8        1979     -70.18       -59.40  -54.38
>
>
>
> On Fri, Sep 2, 2011 at 10:13 PM, Nick Cox <njcoxstata@gmail.com> wrote:
>> In addition, reinstalling the same package from SSC will predictably
>> produce the same result. As I write, the version of -ghansen- on SS
>> is precisely the one that caused you problems. The author, Jorge, has
>> yet to fix it.
>>
>> Muhammad: You either wait for it to be fixed or fix it by hand on your
>> own machine, as I did. Sorry if you thought that I had fixed the
>> package on SSC, but that's not for me to do, as I am not the author.
>>
>> . mata mata clear
>>
>> flushes any Mata code in memory.
>>
>> Nick
>
> *
> *   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/
>
>
>


*
*   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/