Econometric Theory
Applied Econometrics
International Finance

Econometrics
Applied Econometrics
International Finance
Macroeconomics
Finance

Econometrics and Macroeconomic Theory (Oral), 2001
Microeconomic and Macroeconomic Theory (Written), 2000

Discrete Choice Modeling with Nonstationary Panels and Robust Covariance Matrix Estimation

Professor Peter C.B. Phillips
Professor Donald W.K. Andrews
Professor Galina Hale

May 2004

M.Phil., Economics, Yale University, 2002
M.A., Economics, Yale University, 2001
M.A., (Summa Cum Laude), International Economics, Peking University, 1999
B.A., (with distinction) International Economics (with minor in Law), Peking University, 1996

Dissertation Fellowship, Yale University, 2003-2004
Cowles Foundation Summer Prize, 2002, 2003
Graduate Student Fellowship, Yale University, Summer 2001
Graduate Student Fellowship, Cowles Foundation, Summer 2000
University Fellowship, Yale University, 1999-2003
Excellent Graduate Student, Peking University, 1999
Economic Research First Rank Prize, Peking University, 1998
Tokyo-Mitsubishi Bank Award, Peking University, 1997
KALE-B Fund Scholarship, Peking University, 1995
Excellent Academic Achievement Awards, Peking University, 1992-1998
Freshman Award, Peking University, 1992
(Ranked 2nd in National College-Entrance Exam in Zhejiang Province, China)
Best Youth, Yiwu City, China, 1986
(Taking care of an old lady voluntarily ever since primary school)

Teaching Assistant, Econometrics 1 (Graduate), Fall 2002, Yale University
Teaching Assistant, Poverty Under Post-Industrial Capitalism, Spring 2002, Yale University
Teaching Assistant, Intermediate Macroeconomics, Fall 2001, Yale University
Instructor, Investment and Management, Spring1998, Peking University
Instructor, International Finance, Spring1997, Peking University

"Discrete Choice Modeling with Nonstationary Panels Applied to Exchange Rate Regime Choice," mimeo, Yale University, 2003 [job market paper].

"Long Run Variance Estimation Using Steep Origin Kernels without Truncation," (with Peter C.B. Phillips and Yixiao Sun), 2003, Cowles Foundation Discussion Paper 1437, submitted.

"Consistent HAC Estimation and Robust Regression Testing Using Sharp Origin Kernels with No Truncation," (with Peter C.B. Phillips and Yixiao Sun), 2003, Cowles Foundation Discussion Paper 1407, submitted.

"The KPSS Test with Seasonal Dummies," (with Peter C.B. Phillips), 2002, Cowles Foundation Discussion Paper 1373, also Economics Letters 77.

"Robust Inference in Cointegration"
"Nonstationary Discrete Choice: A Corrigendum," (with Peter C.B. Phillips and Ling Hu)
"The Limit Behavior of WHP Filter" (with Peter C.B. Phillips)

The Econometric Society
American Statistical Association
Institute of Mathematical Statistics

Co-organizer of the Graduate Summer Workshop at Yale, Summer 2003

Peking University. Research Head, investigated the operation of large state-owned enterprises (SOEs) in Shanghai, China, summer 1997

Remote Sensing (RS) Research Center. Intern, participated in a United Nations project (trained Indonesian people about the application of Citystar (a software product on GIS and RS)); assisted the development of Citystar updated version, 1996

Japan External Trade Organization, Beijing office. Intern, summer 1995

The dissertation has two parts. The first part develops a new asymptotic theory for nonstationary discrete choice panel data models and provides an application to exchange rate regime choice. The second part is concerned with long run variance and consistent HAC estimation and robust regression testing using a new class of power kernels with no truncation.

I. Discrete Choice Modeling with Nonstationary Panels Applied to Exchange Rate Regime Choice

Discrete choice panel data modeling has become a standard tool for empirical economic research. While traditional micro panel data empirical applications have been to large cross section (N) and fixed time series (T), growing interest in cross-country analysis of macroeconomic policy decisions, currency crisis prediction and emerging stock market behavior has promoted the use of large dimension panel data techniques. Often the time-series components exhibit strong evidence of nonstationarity. The goal of the present research is to provide new asymptotics for such cases, in particular, for nonstationary discrete choice panel data regression with individual effects. This development makes use of the limit theory for nonstationary panels in Phillips and Moon (1999) and provides for both sequential and joint limits for the maximum likelihood estimation of the discrete choice panel model. Some results obtained are directly applicable in the wider context of M-estimation. For instance, Wooldridge (1994)’s work is extended to deliver a limit theory for local extremum estimation for multi-indexed processes that is suitable for nonlinear nonstationary panel data analysis.

In the panel discrete choice setting it is shown that the maximum likelihood (ML) estimator is consistent without an incidental parameter problem, and has a limit theory with a fast convergence rate N^1/2T^3/4 (in the stationary case, the rate is N^1/2T^1/2) and a normal limit distribution for the coefficients and thresholds (in contrast, the limit distribution is known to be mixed normal in time series modeling, as shown in Park and Phillips (2000)). Choice probabilities and marginal effects are derived, and it is further shown that the limit behavior of the sample proportions of the various choices does not follow either "arc sine" (as in nonstationary binary choice time series models) or "extended arc sine" (as in nonstationary discrete choice time series models) limit laws. Instead, they converge to nonrandom quantities, which is typical in the stationary case. I also derive an asymptotic theory for tests of coefficient homogeneity.

Traditional panel analysis assumes cross section independence, which is restrictive in many empirical applications. I therefore extend the model to allow for common shocks in the errors as well as regressors, and I also incorporate observable global factors in the model. The asymptotic distribution of the ML estimator is now mixed normal, reflecting to the presence of common shocks, but has the same rate of convergence as in the independence case.

This approach is applied to model the choice of exchange rate regime by monetary authorities, providing an empirical analysis of the phenomenon of fear of floating (reported floats that actually intervene to smooth exchange rate fluctuations). This phenomenon was discussed in Calvo and Reinhart (2002) and has attracted a lot of attention in international finance. I show that consistent with the existing literature, fear of floating is positively associated with foreign denominated liabilities, monetary shocks, and global stock market volatilities.

II. Consistent HAC Estimation and Robust Regression Testing Using Power Kernels without Truncation (joint with Peter C.B. Phillips and Yixiao Sun)

Long run variance (HAC) estimation is now extensively used in applied macroeconomics and finance because of the need to robustify econometric testing. It is known that conventional HAC tests tend to overreject under the null hypothesis in finite samples and this has led to recent research (for example, by Kiefer and Vogelsang, and Jansson) that seeks to improve the size properties of the tests.

Our research proposes a new class of power kernel estimates for long run variance and HAC estimation. The power kernels have sharp or steep behavior in the neighborhood of origin and are constructed by exponentiating mother kernels, and they can be used without truncation or bandwidth parameters. When the exponent is passed to infinity with the sample size, these kernels produce consistent HAC estimates. The new estimates are shown to have limit normal distributions, and formulae for the asymptotic bias and variance are derived. With power kernel estimation, bandwidth selection is replaced by exponent selection and data-based selection is possible. Rules for exponent selection based on minimum mean squared error (MSE) criteria are developed. Optimal rates for steep origin kernels that are based on exponentiating quadratic kernels are shown to be faster than those based on exponentiating the Bartlett kernel, which produces the sharp origin kernel. It is further shown that, unlike conventional kernel estimation where an optimal choice of kernel is possible in terms of MSE criteria (Priestley, 1962; Andrews, 1991), steep origin kernels are asymptotically MSE equivalent, so that choice of mother kernel does not matter asymptotically.

Analysis and simulations indicate that power kernels lead to tests with improved size properties relative to conventional tests and better power properties than other tests using Bartlett and other conventional kernels without truncation. Some data-determined rule for practical use is provided.