Date of Graduation
Chambers College of Business and Economics
The first chapter proposes a nonparametric test of significant variables in the partial derivative of a regression mean function. The test is constructed through a variation based measure of the derivative in the directions of the significant variables, with the derivative estimation through a local polynomial estimator. The test is shown to have the asymptotic null distribution and demonstrated to be consistent. The chapter further proposes a wild-bootstrap test, which exhibits the same null distribution regardless of whether the null is valid or not. Through Monte Carlo studies, the test shows encouraging finite sample performances. Through an empirical application, the test is applied to infer certain aspects of regression structures on labor's earning function. The second chapter investigates the role of debt in the firm's production frontier and technical efficiency by employing a firm-level dataset over 1998-2007 and 1998-2013. The impact of debt on frontier is decomposed into a stand-alone neutral effect and indirect non-neutral effects, which alter the output elasticity of production inputs. The effects are estimated through a semiparametric smooth coefficient stochastic frontier model. A nonzero probability for the firms to be fully efficient is allowed, modeled as a function of debt and technical progress. The study shows that an increase in debt significantly shifts firms' frontier downward across different ownerships, regions, and industries. Foreign and private firms are more efficient, with their full efficiency probability increased by debt and technical progress. By contrast, state-owned enterprises (SOEs) and collective firms are much less efficient and their probability of being fully efficient does not increase with more debt. Furthermore, lower efficiency levels are concentrated in the central and western regions and in the mining and public utility industries. The third chapter proposes a semiparametric additive stochastic frontier model for panel data, where inputs and environment variables can enter the frontier individually and interactively through unknown smooth functions. The inefficiency has its mean function known up to certain parameters, and influenced by its determinants that may or may not appear on the frontier. The model disentangles time invariant unobserved heterogeneities from inefficiency, which can be helpful to avoid overestimating the inefficiency level. Different from conventional stochastic frontier models, the proposed model can be identified without the distribution assumption on the composite error, and consistently estimated without suffering from the curse of dimensionality. Thus, a large number of interested variables for frontier or inefficiency determinants can be included, a potentially attractive feature for empirical studies. The study demonstrates the appealing finite-sample performance of the proposed estimator and two related hypotheses tests through the Monte Carlo study, and performs a world production frontier analysis with 116 countries during 2001-2013.
Wang, Taining, "Three Essays on Nonparametric Hypothesis Testing and Stochastic Frontier Analysis" (2019). Graduate Theses, Dissertations, and Problem Reports. 3930.