| I propose a new method, utilizing vine-copulas, to perform regression analysis on marketing or econometric data in the presence of endogeneity. This method entails the researcher placing a vine-copula tree structure between the various endogenous regressors and the error term to model the joint distribution. Using this model, the researcher can perform inference using the likelihood derived from the joint distribution. This technique can be used in situations where there is no adequate instrumental variable available. Additionally, this technique can eliminate endogeneity bias in situations where the dependence structure is not symmetrical, a case where other instrument-free approaches do not perform adequately. In this dissertation I implement Bayesian methods to perform this inference on a variety of synthetic and real-world data, culminating with a demonstration which uses the method to estimate brand value using detergent purchases in the Nielsen data set. |
