Nichtparametrische Bayesianische Inferenz für Copula-basierte Endogenitätskorrekturen

• Applicant: Professor Dr. Rouven Haschka
• Subject Area: Statistics and Econometrics ; Management and Marketing
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 568088747

Causal inference is a central topic in empirical research, especially when researchers need to control for the endogeneity of regressors. A traditional approach to correcting for endogeneity is to use instrumental variables (IV), but this requires valid and strong instruments, which are difficult to find. Therefore, more flexible methods for correcting for endogeneity biases are sought in the more recent (marketing) literature. In the context of this project, an IV-free approach to correcting for endogenous biases is proposed, which nonparametrically-bayesically models the joint distribution of the structural error and the endogenous regressors by a copula function. So far, mainly parametric models such as the frequentist copula approach (JCM) by Park and Gupta (2012) have been used. Our approach bypasses these parametric assumptions and uses Bayesian inference. The approach models the distribution of the structural error as a Dirichlet process mixture prior and the copula function as an infinite mixture, depending on the regressors. This eliminates the need for restrictive parametric assumptions. The approach offers a flexible and general methodology for IV-free endogeneity correction and better captures estimation uncertainty. Advances in Markov Chain Monte Carlo methods enable efficient implementation. Furthermore, there is no need to investigate asymptotic properties, since Bayesian inference allows for full posteriori estimates. Three methodological manuscripts are planned for submission to leading marketing journals, as these are very open to new identification methods (Journal of Marketing Research, Marketing Science, Journal of the Academy of Marketing Science). In addition, software implementations will be developed to make copula-based endogeneity corrections available in a user-friendly way in R and Stata. This supports the open science community and makes it easier for researchers to access new IV-free methods.

• DFG Programme: Research Grants