The comparison should be between “structural” theories (e.g., from observables, derive/estimate preferences, best-response functions, then compute equilibrium bids) versus “reduced form” theories (e.g., mapping directly observables to bid predictions).

The advantage of structural theories including unobservable (theoretical) entities would be that:

1) they provide a possible explanation of the mechanism through which the world implements the reduced form; so a) we know that there is at least one conceivable such mechanism and b) we can discuss whether such mechanism passes some “smell” test or not (that is where the intuition stuff matters);

2) we may found out that sometimes the unobservable variables (or something sufficiently correlated with them) is actually observable, thus providing more information for the estimation and/or more opportunities to test the theory.

Anyway, hasn’t the debate about behaviorism been settled already?

]]>While that may be true for some models, I don’t think that it’s true in general.

What if competing theories generate the same empirical predictions regarding a set of observed variables, but have different policy implications? If this were the case, it would matter which underlying hypotheses are rejected.

An example in which this situation can arise is provided by a recent blog post by David Andolfatto (http://andolfatto.blogspot.com/2011/09/interest-rates-and-slumps-competing.html).

]]>I think it matters a great deal if you match data when the underlying hypothesis is rejected, because that means the underlying hypothesis does not apply to that situation. You only reject when you cannot match.

Maybe you can’t measure a parameter exactly, but you can – by plowing through the model – understand how to think of your particular empirical situation because of the theory. So is that the value you’re talking about?

]]>