I don't think it is quite right to see Sim's approach as "fully
atheoretical." He has modified it and agrees that structural issues must
come in. The question is how. The approach I like is that of Soren
Johansen's cointegrated VAR analysis, in which data analysis is informed
by theory, but not driven by theory. I don't think we can teach
principles students that, but we can teach them about the difficulties
that the data present.

Dave Colander