I am currently using Eviews 7 for a single equation cointegration analysis (FMOLS, CCR, DOLS).
I include a dummy variable (as a structural break in the intercept) into the "Deterministic regressors". According to the User's Guide II (page. 227), the standard error, tstatistic, and pvalue for the constant are not strictly valid. So I was wondering:
(1) whether the standard error, tstatistic, and pvalue for the dummy variable is valid or not;
(2) whether the standard error, tstatistic, and pvalue for the time trend is valid or not;
(3) is above conclusion also suitable for CCR and DOLS estimators, or just for FMOLS?
Thanks so much.
dummy in cointegration regression
Moderators: EViews Gareth, EViews Moderator

 EViews Developer
 Posts: 2600
 Joined: Wed Oct 15, 2008 9:17 am
Re: dummy in cointegration regression
That's a misinterpretation of what the manual says. It cautions against using the inference results for the constant term C in the reported regression...
The standard errors and tstatistics in the cointegrating equation should in general be much more forgiving of the inclusion of deterministic regressors than are the actual tests for cointegration (which require simulated critical values). In addition, your dummy variable is I(0) and really should be irrelevant in the midst of all of the I(1) components you have running around. I may have some concerns about inference for its coefficient for reasons akin to the C term, but I would have to give that one some thought before commenting.
Hansen's 1992 JBES article "Test for Instability with I(1) Processes" discusses briefly the issues associated with the more interesting breaking trend functions. I am certain that there are followup discussions, but I don't have the references handy.
The standard errors and tstatistics in the cointegrating equation should in general be much more forgiving of the inclusion of deterministic regressors than are the actual tests for cointegration (which require simulated critical values). In addition, your dummy variable is I(0) and really should be irrelevant in the midst of all of the I(1) components you have running around. I may have some concerns about inference for its coefficient for reasons akin to the C term, but I would have to give that one some thought before commenting.
Hansen's 1992 JBES article "Test for Instability with I(1) Processes" discusses briefly the issues associated with the more interesting breaking trend functions. I am certain that there are followup discussions, but I don't have the references handy.
Re: dummy in cointegration regression
Thanks so much. I will read that paper. Cheers.
Re: dummy in cointegration regression
gman wrote:I include a dummy variable (as a structural break in the intercept) into the "Deterministic regressors".
I stumbled upon this post and suddenly got confused. The source of confusion in the "deterministic regressors" box for single equation DOLS estimation.
Say I am estimating a cointegration relathionship with regime shift:
Y = a + a1 Dum + b X + b1 Dum X + ut
I understand that "deterministic regressors" box is where I should enter shift dummy (in order to obtain coefficient a1).
But my question is:
Is this the place where I should also enter the interaction term between Dummy and X in order to get the coefficient b1?
Or this interaction term should in fact be included in cointegrating regressors box (following Y and X)?
Who is online
Users browsing this forum: No registered users and 7 guests