I think that VAR and other methods can, in principle, also be used for differences. But before choosing a method and before testing one should develope a good, clear specification: What is the dependent variable? How is the influence of the explaining variables (diirection, linear, progressive, degressive)? Only with a satisfactory specification, you can derive what transformations to apply. What concerns the methods, I recommend to first apply the simplest possible like diagrams, correlations and OLS.