Skip to Main Content
 

Global Search Box

 
 
 
 

Files

ETD Abstract Container

Abstract Header

Three Essays on Bayesian Econometric Methods

Abstract Details

2017, PhD, University of Cincinnati, Business: Business Administration.
This dissertation contains three essays examining new Bayesian econometric methodologies. The first develops a heterogeneous Spatial Autoregressive Model by integrating a finite mixture model structure into the traditional homogeneous specification. The second essay builds upon the first by extending this Spatial Mixture Model structure to the more general Spatial Durbin and Spatial Durbin Error specifications. Additionally, this essay covers the interpretation of these new model specifications. Finally, the third essay develops a predictive based model selection process by integrating cross-validation algorithms into standard Bayesian sampling methods with a focus on explicit out-of-sample prediction. 1.0.1 Embracing Heterogeneity: The Spatial Autoregressive Mixture Model In this essay, a mixture distribution model is extended to include spatial dependence of the autoregressive type. The resulting model nests both spatial heterogeneity and spatial dependence as special cases. A data generation process is outlined that incorporates both a finite mixture of normal distributions and spatial dependence. Whether group assignment is completely random by nature or displays some locational "pattern", the proposed spatial-mix estimation procedure is always able to recover the true parameters. As an illustration, a basic hedonic price model is investigated that includes sub-groups of data with heterogeneous coefficients in addition to spatially clustered elements. 1.0.2 Spatial Durbin Mixture Models This essay extends the finite mixture model structure to include Spatial Durbin and Spatial Durbin Error model specifications. The partial derivatives of this heterogeneous spatial model structure are shown to differ between border and interior agents; the designation of which is based on group assignment and first order neighbor designation. As an illustration, individual income based on data from the Panel Study of Income Dynamics (PSID) is examined using the Spatial Durbin Mixture Model specification. Results from the model indicate that returns to income from education are heterogeneous with some agents receiving negative returns on each additional year of human capital development. 1.0.3 Bayesian Predictive Model Selection using Cross-Validation In this essay, leave-one-out cross-validation is combined with Bayesian sampling and inference to develop a new model selection process. This process relies both upon standard hypothesis testing, and, in the absence of sufficient evidence, Ockham's Razor to selection a model from a competing set. It is shown that this process outperforms many of the criteria used in empirical work both by selected the true process more often, and selecting a smaller set of models overall.
Olivier Parent, Ph.D. (Committee Chair)
David Curry, Ph.D. (Committee Member)
James LeSage, Ph.D. (Committee Member)
Jeffrey Mills, Ph.D. (Committee Member)
119 p.

Recommended Citations

Citations

  • Cornwall, G. J. (2017). Three Essays on Bayesian Econometric Methods [Doctoral dissertation, University of Cincinnati]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1504801632767553

    APA Style (7th edition)

  • Cornwall, Gary. Three Essays on Bayesian Econometric Methods. 2017. University of Cincinnati, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=ucin1504801632767553.

    MLA Style (8th edition)

  • Cornwall, Gary. "Three Essays on Bayesian Econometric Methods." Doctoral dissertation, University of Cincinnati, 2017. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1504801632767553

    Chicago Manual of Style (17th edition)