^ well said.
Structural is dead

Let's just say I know very well what I'm talking about. Not everyone sees the world as Darrell does.
Huh? Darrell is a theorist and wouldn't know data if it hugged him.
How many structural AP empiricists has Ken hired at Stanford? Zero.
How many reducedform AP empiricists? I see Nagel and van Binsburgen.
I think you are confused. Ken is extremely fond of models and modeldriven empirical work. He hates 99.9% of all structural work out there, which amounts to matching a few carefully selected moments and ignoring all the other variation in the data that rejects the model.

Lots of misinformation on this board. It appears some people can't get over the idea that some estimation methods are after different objects, like parameter magnitudes, and do not rely on the same identifying restrictions as OLS. Read more and post less please.

If you're not writing down an agent's optimization problem subject to constraints, and then estimate the parameters of the equilibrium conditions, you're not doing economics. You're making a contribution to society, you're learning facts about the world, but you're a quantitative social scientist and not an economist.

Huh? Darrell is a theorist and wouldn't know data if it hugged him.
...
I think you are confused. Ken is extremely fond of models and modeldriven empirical work. He hates 99.9% of all structural work out there, which amounts to matching a few carefully selected moments and ignoring all the other variation in the data that rejects the model.First, reduced form theory exists and that's a lot of what Darrell does. Confused? Look at the difference between his models for default and Merton's.
Second, Ken whacks students over the head with structural. He hates a lot of it because he wants it to be better, not because he thinks it is irrelevant.
Third, structural models done well can do a lot more than momentmatching. Reducedform models can often be fit by matching moments; that *structural* aspect comes from what the parameters mean and why you think they take the form they do.

I'm a PHD student learning structural now, after the "learn structural" mantra was repeated over and over again. In the end, great, I know some more obscure stuff....
...but i) I feel it is applied coherently to a narrow range of questions, ii) the literature seems so selfreferential, iii) only two people in my research community do the stuff. Thus, I'm left with little guidance, answering banal questions that only structural people appreciate, with less time spent on more substantive work that is publishable.
I'm not sure why everyone was so sold on this.

Data are just data. You can't learn anything from data directly. You learn and you gain understanding from building narratives. If you're not going to be clear and rigorous about the narrative your constructing, if you're not going to build and test a structural model, then the only difference between you and an anthropologist is that he waves his arms in the air and you wave them in the direction of a table of regression results.

Data are just data. You can't learn anything from data directly. You learn and you gain understanding from building narratives. If you're not going to be clear and rigorous about the narrative your constructing, if you're not going to build and test a structural model, then the only difference between you and an anthropologist is that he waves his arms in the air and you wave them in the direction of a table of regression results.
haha, you made me laugh...

If you're contemplating taking up structural estimation, I would advise that you think first about where you will be able to publish your results. The journals are there, but the set of them is much smaller than for reduced form.
There is definitely a club aspect, too, in that you mostly see the same people over and over.

What are referring to when we're talking about structural estimation?
Usually BLP demand estimation of the price elasticity of demand for carbonated soft drinks (CSD). The most innovative stuff I've seen tries to get at the effect of inventory behavior on the elasticity of demand (it has been a while, but I think that's the Ketchup paper).
Lately, structural models have ventured into dynamic models that are estimated either with BLP, nested logit, or some kind of MPEC method.
The problem with structural stuff is that so much time is spent on getting a model that can be estimated structurally (read: layer upon layer of modeling assumptions that are for tractability as much as sensible economic theory) that not much attention is usually paid to the economic intuition of the estimated parameters. Yes, there's usually a section entitled "identification" in these papers, but very rarely to people present reducedform regressions (estimation of the intermediate equations) on the path to building the fullblown structural model.
The worst of these papers has 35 equations (model > estimating equations > specifications) and two tables (parameters and counterfactuals). You could provide lots of evidence (from the data) on whether the assumptions that are built into the 35 equations make sense, but too rarely are checks of these assumptions presented. There are exceptions to this rule, but the 352 model is often used to produce a structural paper.
There's a place for structural estimation b/c counterfactuals are important to think about, but too many structural papers are out of touch with the theories they purport to test.