Cointegration test

[bobsun]

Hi everybody,

I am trying to build a normal linear regression model and my time-series are non-stationary. On the 1st difference level are the correlation-coefficients too low. I have read that I can use the Johanson cointegration test to turn the level datas into stationary and use them for the further regression. But I can't understand how to interprete the results from the test and which coefficients do I have to use for the regression?

Thanks a lot and I hope of responds!

:o)

[samijo]

The fact that variables in level are non-stationary doesn’t mean that they are not cointegrated, so you should test for cointegration in the levels first. If they are cointegrated then you can use normal regression, because the combination of variables creates a stationary error term. If not stationary in levels you might move to differences.
Interpretation, levels as normal; differences refer to the change in the variables. You might try differences in the log which can be interpreted as the growth rate in the variables (or the acceleration in the variables if they are in growth rates in levels).