EViews.com

Dear Friends, Colleagues, Teachers and Professionals

I'm currently working on my dissertation, which is about the effects of macroeconomic indicators on stock returns.
I'll be delighted if you could help me with my problem...
I'm analysing 2 economies, with 1 dependent and 4 indenendept variables.
I've converted all of them into logs.
2 variables have unit root, so I've converted them (as a new series by taking 1st difference)
So: every variable is stationary.

1- I've tried to do VAR with already stationary and stationary-converted variables, and AIC indicates I should choose 5 lags.
So I went to Johansen Cointegration test and put only non-stationary variables. The test says I do not have cointegration.
With 2 lags, as Hannah and SC criterion suggests, I get cointegration.
Is my method correct, should I leave it and say there is no cointegration, or, should I choose Hanna Criterion's suggested lags in Johansen?

2-I've run a regression without converting them into stationary form and saved residuals. Then, I've taken first diff. of non-stationary variables and run the same regression again buy including the lagged residual. Residual does not have unit root, it is stationary.
No spurious regression. Residual is normally distributed and there is no serial correlation.
Lagged residual is significant, with 8% adjustment each term. 3 variables out of 4 are not significant, but the 4th one is significant.
Should I interpret this as a "short run" or "long run" relationship?
Secondly, Is this the "vector error correction model" ?
If thats not VECM, then should I do it by selecting "Vector Error Correction" in VAR equation screen and add "1st difference taken variables"I I still dont know whether VECM or VAR captures long run relationship.

3-Is there a rule like "use VAR if there is no cointegration and use VECM if there is cointegration" ?

4-To see the long run relationship between the variables, can I simply put 1st difference taken forms of them into simple regression model?


Best Regards,
Noyan

Bump. I would love an answer to this. Where on earth is the long run relationship? Why doesn't EViews seem to provide it and should you just estimate it yourself via OLS?

Thank you.

1.) Your method is correct. It depends what IC you want to put more faith in. Ultimately these are both just results to guide whatever theory you have about the relationship between the variables. There are a large, large number of extensions from this basic framework to analyse whether there exists cointegration between these variables in specific samples.

2.) You are describing the dickey-fuller two-step procedure.

3.) That is pretty much the rule, yes.

4.) No.

The long run relationship is shown by the beta vector in the VECM output. EViews does provide it.

Charlie

Hi,

Could you maybe describe where in the output we find the alpha vectors? I am very confused if the output under the error correction part for CointEq is the alpha vectors or the alpha*beta´part of the regression.

Thank you