Dear researchers,
I‘m doing a panel VECM with four variables, x1, x2, x3, and x4. The first three variables are I(1) and the last one is I(0). I want to use x1 and x2 as endogenous variables, and x3 and x4 as exogenous variables.
I've done some readings and search to try to find answers to my questions, but they are still not solved. So I would appreciate it if anyone could enlighten me on any of the questions below:
1. Should the endogenous variables and exogenous variables be all cointegrated to run VECM? Or it's OK if only the two endogenous variables are cointegrated?
2. Should I test endogeneity before deciding which variables are exogenous?
3. If I can run the model with my data, how could I do it? e.g. calculate the residual between x1 and x2 in VAR, and put it into the VECM with x3 and x4? Could I use Stata, or Eviews or Gretl to do this?
4. If I cannot do VECM in with my data, what models would you recommend? For example, could I run a VAR model and do irf using d.x1 d.x2 d.x3 and x4 (d. means first difference)?
5. Could I take first difference of the I(0) variable to run the model? The variables will all mean percentage change with first difference.
Thanks in advance!
2. Should I test endogeneity before deciding which variables are exogenous?
YES, OF COURSE!
BUT THERE ARE EXOGENEITY TESTS, NOT ENDOGENEITY. TESTS. USE GRANGER TESTS.
1. Should the endogenous variables and exogenous variables be all cointegrated to run VECM? Or it's OK if only the two endogenous variables are cointegrated?
FOR AN VECM ALL ENDOGENOUS VARIABLES SHOULD BE INTEGRATED AT THE SAME ORDER, E.G. I(1). WHETHER THEY ARE ALSO COINTEGRATED IS DETERMINED BY THE RANK TEST.
THE EXOGENOUS VARIABLES SHOULD BE STATIONARY, I.E., I(0). IF YOU HAVE AN I(1) VARIABLE YOU CAN MAKE IT STATIONARY BY USING FIRST DIFFERENCES.
5. Could I take first difference of the I(0) variable to run the model? The variables will all mean percentage change with first difference.
YOU USE FIRST DIFFERENCES ONLY FOR INTEGRATED VARIABLES. THERE IS NO STATISTICAL REASON TO DIFFERENTIATE AN I(0) VARIABLE.
4. If I cannot do VECM in with my data, what models would you recommend? For example, could I run a VAR model and do irf using d.x1 d.x2 d.x3 and x4 (d. means first difference)?
VAR IS ONLY SUITABLE WHEN ALL VARIABLES ARE STATIONARY. IF NOT THEN VECM IS APPROPRIATE.
3. If I can run the model with my data, how could I do it? e.g. calculate the residual between x1 and x2 in VAR, and put it into the VECM with x3 and x4? Could I use Stata, or Eviews or Gretl to do this?
EVIEWS AND STATA (AT LEAST) ESTIMATE ALL VAR, VECM, ETC MODELS AUTOMATICALLY, THAT MEANS IT DOES ALL THE JOB AND YOU DON'T HAVE TO PERFORM THE MANUAL PROCEDURE YOU DESCRIBE.
GOOD LUCK!
- Badges
- Science topic