Hi! I just wanted to ask if it is possible to include dummy variables in the first step of an and Engle-Granger cointegration test? A one-time event in the time series seems to dislove the cointegrating relationship otherwise. If not, can I come around the problem in any other way?
Thanks a lot Daniel Meierrieks! So, in the corrected time series I should use is the residual is the dependent variable in the new series I should use?
Assuming the effect affects only to the cointegration relatiohsip, then you should take it into account and include a dummy variable in the long-run relationship. It would be advisable to also include it later into the short-run model, but in fist differences. Thus, you should include a step variable (D1t, =1 if t>Tb, 0 otherwise) in the long-run and an outlier in the short-run (D2t = D1t - D1(t.-1) = 1 if t=TB, o otherwise). Then, you should test for cointegration by using either Gregory and Hansen or the most recent Perron and Kejriwal (2010), JBES. You can also check the use of the dummy variables in Montanes and Sanso (2002), Applied Economics... Hope this can help u. Do not hesitate contact me if you need more help.
Daniel Meierrieks, when I use the method you recommend, I do indeed get co-integration. But when I use the corrected time series of my dependent variable, the short term relations go away completely. That seems odd, considering very strong relationships to the dependent variables used with the old time series. I have 110 observations of the dependent variable, and the dummy variable I use for correcting it only have 7 cases of 1, vs 113 cases of 0. Is this correct, or should I use the old, uncorrected series as first differences is in the VECM? Should I include the dummy variable I used to correct the dependent variable in the VECM? And by the way, should I use a time dummy too?
I dont think there is any need to include a dummy variable in a cointegration test. It works with ant two time series that are integrated of different order (of integration).
Hi Lar, Daniel and Antonio have seriously attempted to explain your question with a good econometric approach but I have both question and suggestion. why do you want to attempt including dummy variable in cointegration test? However, if it is to identify structural break as mentioned in the discussion, why not introduce interaction variable of variable(s) of interest then apply the cointegration test.
- Badges
- Science topic
- Similar topics
- Instrumentation Engineering
- Instruments