Hello, I run the Johansen cointegration test in Eviews 9 for three real prices. All three variables are are I(1) and display almost no trend. To run the test in Eviews one has to specify "lag intervals". The "lag intervals" should be specified as a pair. You could see the group dialog box of the test at http://www.eviews.com/help/helpintro.html#page/content/coint-Johansen_Cointegration_Test.html. Particularly it says that if you type “1 2” in the lag intervals edit field, the test VAR regresses deltaY(t) on deltaY(t-1), deltaY(t-2), and any other exogenous variables that you have specified. I understood that lag intervals of first differences (I use word delta for them) are included in the VECM equation (equation 47.2 on the web page I referred above) My question is what would be the meaning of the pair "0 1"? I get different results if I use "1 1" (means regresses deltaY(t) on deltaY(t-1)) and "0 1". Thank you.
0 indicates level and 1 indicates one year lag, however, you can't simply put the lag arbitrary, there is a process of selecting optimum lag
Kazi thank you for your response. Regarding the lag order I selected lag based on AIC criteria. It is still unclear for me what exactly "0 1" lag intervals selection mean.
As I understand the lag intervals selection "0 0" means the following form of the error-correction model for the test (only lag level is included):
delta_y(t)=Pi*y(t-1)+epsilon(t)
Then "1 1" means the following form (regression on one lag of delta_y):
delta_y(t)=Gamma*delta_y(t-1)+epsilon(t)
Does "0 1" means the following error-correction model (one lag of level and one lag of first difference)?
delta_y(t)=Pi*y(t-1)+Gamma*delta_y(t-1)+epsilon(t)
Thank you.
EViews assumes that you specify the lag of the test VAR in terms of pairs. So this is an EViews specific issue. The specification is for the test VAR is in first differences, i.e. "1 3" means that your test VAR has 3 lags in first differences, but 4 in levels. If the level VAR has only 1 lag, the specification "0 0" is appropriate, i.e. the differenced VAR does not have a lag. You can check the respective specification of the ECM in EViews, if you estimate the VAR as a VECM
(0 1) means use of lag from 0(level) to first lag . Since you cannot regress a variable on itself(level on level) , 0 1 is inappropriate too.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics