Abstract

This dissertation is composed of three chapters on estimation of vehicle choice and utilization models, simulated likelihood estimation, and Bayesian non-parametric additive methods for neighborhood effect models. The first chapter exploits differences in fuel efficiency between hybrid vehicles and their gasoline counterparts to investigate two behavioral questions relating to fuel economy standards: how car buyers value fuel economy (the energy paradox) and whether improved fuel efficiency increases travel (the rebound effect). Emphasis is placed on handling methodological and data issues that are typically ignored in prior studies, such as partially observed choice, endogeneity, and measurement error. Estimates of the rebound effect and consumer valuation of fuel economy remain imprecise despite the use of the most detailed household level data available and sound methodology to handle limitations with these data. The inability to precisely estimate these important policy questions suggests it is a worthwhile endeavor to obtain reliable, detailed data on household vehicles. The following chapter (joint with Ivan Jeliazkov) presents techniques, based on Markov chain Monte Carlo (MCMC) theory, for construction of the likelihood function in a broad class of hierarchical models where direct evaluation of the likelihood function is not possible. We review existing estimators, introduce new MCMC estimators, and examine their performance in applications to the Poisson-log normal and mixed logit models. The MCMC techniques outperform existing methods in both settings, with the existing methods performing especially poorly in the Poisson-log normal case. The final chapter applies Bayesian semiparametric additive methods to a neighborhood effects model. The baseline model assumes all covariates enter linearly, whereas the approach in this paper allows for flexible functional forms. An efficient Markov chain Monte Carlo (MCMC) algorithm that exploits the properties of banded matrices is proposed for estimation. The efficiency gains offered by the banded matrix algorithm are critical, as they permit the estimation of applications with large sample sizes. The model and estimation methodology are used to examine foreclosure contagion in California. The results reveal the impact of neighborhood effects on foreclosure rates as nonlinear, where the relationship resembles a tipping point phenomenon.