This paper highlights four shortcomings of recent quantitative and deepstructure models in corporate finance: (1) These models have omitted too many plausible forces not based on evidence but based on authors' priors. (2) The link between their unobserved structures and their reduced-form empirical evidence has been too weak (even orthogonal forces could have affected their inference), (3) The existing tests have largely ignored many important econometric issues, such as selection and survivorship biases. (4) The models have never been held to reasonable test standards, such as performance in quasi-experimental settings. Constructively, my paper offers two primary suggestions: The first is to search for more direct empirical proxies instead of relying on "assumed" first-order conditions. The second is to design quasi-experimental tests of structural models. It illustrates these points in the context of Hennessy and Whited (2005) and Strebulaev (2007).