Burda, Martin
(2007)
Essays in Semiparametric Econometrics and Panel Data Analysis.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
Abstract
Limited dependent variable (LDV) panel data models pose substantial challenges in maximum likelihood estimation. The likelihood function in such models typically contains multivariate integrals that are often analytically intractable. To overcome such problem in a panel probit model with unobserved individual heterogeneity and autocorrelated errors, in Chapter 1 - co-authored with Roman Liesenfeld and Jean-François Richard - we perform classical and Bayesian analysis of the model based on the Efficient Importance Sampling (EIS) technique (Richard and Zhang, 2006). We apply our method to the product innovation activity of a panel of German manufacturing firms in response to imports and foreign direct investment confirming their positive effects. Nonetheless, our key coefficient estimates are smaller than found in previous literature which can be explained by our flexible model assumptions. The remaining two chapters present my work on new estimation methods for models based on conditional moment restrictions. Such models are frequently stipulated by economic theory but only a few estimators based directly on them have so far been analyzed in the literature. Indeed, estimation of parameters therein poses a difficult ill-posed inverse problem. Rather, these models are typically converted into unconditional moment restrictions that are easier to handle. However, such conversion results in a loss of information compared to the original specification. Using the information-theoretic framework of so-called Generalized Minimum Contrast (GMC) estimation, in Chapter 2 I propose a new class of estimators based directly on conditional moment restrictions that encompasses the entire GMC family. Moreover, I show that previous literature covering a few special cases of the GMC class use an arbitrary uniform weighting scheme over the space of exogenous variables that can be improved upon with optimal local weighting. All currently available GMC estimators are based on moments containing finite-dimensional Euclidean parameters. To alleviate a potential misspecification problem resulting from strong parametric assumptions, in Chapter 3 I propose a new Sieve-based Locally Weighted Conditional Empirical Likelihood (SLWCEL) estimator containing also infinite dimensional unknown functions, thus extending a special case of Chapter 2 to the semiparametric environment. Much of Chapter 3 is devoted to analysis of SLWCEL's asymptotic properties.
Share
| Citation/Export: |
|
| Social Networking: |
|
Details
| Item Type: |
University of Pittsburgh ETD
|
| Status: |
Unpublished |
| Creators/Authors: |
|
| ETD Committee: |
|
| Date: |
26 September 2007 |
| Date Type: |
Completion |
| Defense Date: |
25 April 2007 |
| Approval Date: |
26 September 2007 |
| Submission Date: |
17 April 2007 |
| Access Restriction: |
5 year -- Restrict access to University of Pittsburgh for a period of 5 years. |
| Institution: |
University of Pittsburgh |
| Schools and Programs: |
Dietrich School of Arts and Sciences > Economics |
| Degree: |
PhD - Doctor of Philosophy |
| Thesis Type: |
Doctoral Dissertation |
| Refereed: |
Yes |
| Uncontrolled Keywords: |
Gibbs sampling; MCMC; panel probit; Generalized Empirical Likelihood; Sieve Minimum Distance |
| Other ID: |
http://etd.library.pitt.edu/ETD/available/etd-04172007-170017/, etd-04172007-170017 |
| Date Deposited: |
10 Nov 2011 19:37 |
| Last Modified: |
15 Nov 2016 13:40 |
| URI: |
http://d-scholarship.pitt.edu/id/eprint/7249 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |
![[feed]](/images/feed-icon-32x32.png){kind=icon}