The scene at SoFiE 2013 leads me to reflect on structural vs. reduced-form modeling approaches in econometrics. Much financial-econometric work is reduced-form, whereas structural modeling has recently become fashionable in certain other areas of econometrics. The structure police, especially new recruits, are often fanatical. But the reality is that reduced-form "statistical" models are every bit as scientific as structural models. Structural models are simply restricted reduced-form models, and it's a delicate and situation-specific matter as to whether imposing structural restrictions on reduced forms is necessary or desirable.
Many central activities in finance involve descriptive and predictive tasks, which are often most effectively executed in reduced-form mode. That is, we don't necessarily need deep structural understanding to succeed, for example, at prediction, which is wonderful, because we often don't have deep structural understanding. (Admit it.)
One key example, on my mind because it features prominently at SoFiE, is financial market volatility modeling. GARCH, stochastic volatility, realized volatility, whatever -- all such approaches are reduced-form, essentially autoregressive. Yet financial econometric volatility modeling has been hugely successful in both academic and industrial finance. It is now used routinely and productively in risk management, portfolio management, spot and derivative asset pricing, and more.
And financial econometric volatility modeling is just one example. All told, for descriptive and predictive tasks in a variety of sub-areas of econometrics, reduced-form modeling often provides the best of all worlds, delivering major advances while avoiding structural modeling pitfalls.