Johansen test (VECM) is superior compared to EG test. EG test better for 2 variables. Johansen test better for more than 2 variables.J test allows for more than one cointegrating vector, but all the variables should be integrated of the same order. Normally all types of residual test, autocorrelation LM test, heteroskedasticity test, normality test, even impulse response and variance decomposition can be done after that.

Thank you professor. My series are I(1). I implemented Johansen test and found no cointegration. But EG results support cointegration relationship. I have two models, namely long-run and  short-run (ECM) models. I wonder which one should be checked for the assumptions? LR or ECM or both?

Let me add something to what Mousumi said:

She states that Johansen requires all variables to be the same order. Well, that is not correct. because Johansen requires variables: I(1) and I(0). Concerning checking the  assumption, you do that in the VAR as well as VECM. For the VAR, you need to check for the lag order as well as residual correction etc. and in the VECM, you should also check for whether the Ecm term enters the normalized equation. Also,check for serial correlation,  normality,  heteroskedasticity and so ........ 

Do you say that Johansen test can be implemented for variables with different integration orders (max I(1))? I think ARDL Bounds test should be used in this situation.

Stationary variables can enter unrestrictedly into the cointegration space (see Rahbek and Mosconi (1999) and can enter outside (exogenously) the cointegration space (see Johansen (1995). Therefore, it is not only through ARDL okay!

B. Veil: Sir, you may want to see also Harbo, Johansen, Nielsen and Rahbek (1998). Thake care, Prof. Arize

thank you proffesor, i will check them

Thank you, my good friend but don't work too hard!

Regards,

Prof. Arize