Dear Davron

1. If you want to run the VAR, all variables must be stationary even at level or at 1st difference that is not matter. So you just need to make sure that your variables are stationary.

2. If you data at not stationary all at level, you can test the contegration test that you have to do with all data at level.

The answer 1 has provides something for 3.

I hope that help

Canh

Dear Canh,

Thank you very much for your response. It really helped a lot.

I will do cointegration test at level data though it is nonstationary. I will select lag length based on the VAR model using the differenced data.

I would highly appreciate your expertise in the following question as well.

I have monthly data for two variables, bank credit and monetary base. I want to study their causal relationship. When I used full period data I found cointegration based on Johansen test. But when I divide the data into two periods there was no cointegration. Separate periods have around 70-month data. So I can run VECM for full data and VAR for the separate periods. Is it because data length for separate periods is not enough. What would you suggest? Would the results of separate periods analysis be robust? 

Thanks 

Davron

Dear Davron

through your word, i guess that your sample has 12 year of monthly data.

so, with vecm is a cointegration model for long term conintegration, the results of full sample have contegration is ok. And the results for two sub sample (aroud 6 years) do not have are make sense since 6 years are a little bit short run

in addition, if you want to test the causality of two variables you can also use granger causality test. 

Moreover, i may guess you full sample has structure break, that can be estimated by threshold VAR as well

but, the more important thing I care is about your variable: the monetary base (i guess m1) and bank credit that means you investigate the credit channel of monetary policy transmission where there are some other factors on it, you should find more to avoid the un identification of model for making sure of results consistent

best

Canh

For selecting lag length: Use information criteria like AIC or BIC to determine the optimal lag length in your VAR/VECM model.

For testing co-integration: Employ tests such as the Johansen test or Engle-Granger test to assess co-integration between variables.

For VAR/VECM with non-stationary I(1) data: Implement the VAR or VECM model directly on the non-stationary data, as these models can handle I(1) variables without differencing.