Testing for PPP long run effect

[Konstantin]

Dear collegues
I am currently stuck on testing PPP for long run effect between France and Germany during post-Bretton Woods period (April 1973-December 1990). I need to check for cointegration (Johansen test) among Fr/Ger exchange rate, Fr CPI and Ger CPI. Taking logarithms, basically, the equation looks like ln(Ger/Fr)=ln(Ger CPI)-ln(Fr CPI). Preliminary, we should test for unit root, and there is the most interesting thing: utilizing ADF(with constant and trend) and Perron's Procedure I got Ger/Fr = I(1), CPIs are both I(2). The question is how can I perform cointegration test if ranks of integration are different? Is it eligible to take the difference concerning CPIs to derive I(1) process and proceed with Johansen test?
Regards