We propose a maximum likelihood framework for estimating finite mixtures of multivariate regression and simultaneous equation models with multiple endogenous variables. The proposed “semi‐parametric” approach posits that the sample of endogenous observations arises from a finite mixture of...