Cointegrating Regression

[jmpjjp]

I am a bit confused and am hoping for a bit of help. I check for stationary variables. I get stationary in the first or second difference. However, when I do the regressions with the difference variables none of the models are significant. So I checked for cointegration. Cointegration exists. So I did a cointegration regression. My question is do I use the original variables x and y or do I use the difference variables dx and dy?
Thank you for the help!