Econometrics
Financial Economics

May 1999 (Written) Microeconomic Theory and Macroeconomic Theory
May 1999 (Oral) Financial Economics
October 1999 (Oral) Econometrics

Econometrics of Nonstationary Panel Data Applied to CEO Compensation Analysis

Professor Peter C.B. Phillips
Professor Donald W.K. Andrews
Professor Michael Keane

Completed August 2002

Ph.D., Yale University, expected to be awarded in December 2002
M.Phil., Yale University, December 2001
M.A., Yale University, December 1999
M.S. in Economics (with distinction), University of Tampere, Tampere, June 1998
M.S. in Physics (with distinction), Tampere University of Technology, August 1998
GPA: At Yale in economics 2.93/3.00, in Finland 2.88/3.00 and 4.58/5.00

At Yale University
   Emil Aaltonen Foundation Fellowship 1998-2001
   Yrjö Jahnsson Foundation Fellowship 1999-2001
   Jenny and Antti Wihuri Foundation Fellowship 1998, 2000, 2001
   Cowles Foundation Graduate Student Fellowship 1999
   Cowles Foundation Prize 2000, 2001
   Yale University Fellowship 1998-2002
In Finland
   Yrjö Jahnsson Prize for the Best Economics Master’s Thesis at University of Tampere, 1998
   Member of Finnish team in International Physics Olympiad, 1992
   Member of Finnish team in International Olympiad of Russian Language, Moscow, 1991

Instructor, Valmennuskeskus, 1998 and Boomi, 1996, 1997, Finland
   Taught prospective business and economics majors economics for entrance tests
Teaching Assistant, Department of Physics, Tampere University of Technology, 1996, 1997, Finland
   Led introductory physics sections

Research Assistant, Semiconductor Lab, Tampere University of Technology, Finland, 1996-1998
    Optimized semiconductor lasers using numerical simulations and processed solar cells
Summer Intern, R&D Center, Nokia Mobile Phones, 1995

"CEOs Are Really Paid Like Bureaucrats," June 2002

"Estimating Average Long Run Relations in Nonstationary Panels with Large Cross Sections," June 2002

"Limit Distributions for Linear Regressions in Nonstationary Panels when n and T Are of Same Order of Magnitude," August 2001

"Linear Regressions in Nonstationary Panel Mixtures of Unknown Type," September 2001

Nonstationary panels with a possibly large cross section relative to the time series dimension are common in financial and macroeconomic applications including cross-country growth and savings analyses. The first chapter of this dissertation proposes a new way of estimating relationships in these panels. The second chapter uses the new estimator to evaluate the link between CEO compensation and company performance. The third chapter demonstrates some problems associated with the pooled least squares procedures and other typical panel estimators in nonstationary panels with large cross sections.

The first chapter, Estimating Average Long Run Relations in Nonstationary Panels with Large Cross Sections, discusses estimation of and inference for the average long run relation parameter proposed by Phillips and Moon (1999) in unit root nonstationary panels when the cross-sectional dimension n is possibly much larger than the longitudinal dimension T. The chapter proposes a new estimator that is constructed as a cross-sectional arithmetic average of kernel long run covariance matrix estimators of the time series of the panel. Cross-sectional pooling helps to improve the bias properties of the estimator relative to other common estimators. The estimator is consistent and asymptotically normal at rate root-n in spurious and heterogeneously cointegrated panels and at rate root-nT/K in homogeneously and near-homogeneously cointegrated panels both under sequential and joint limits, where K is a kernel bandwidth. Biases in the asymptotic distributions can be eliminated if n/T² --> 0, which covers a much wider range of panels than existing estimators that typically assume that the time series dimension is larger or at least of the same order of magnitude as the cross-sectional dimension. A relatively large value of the bandwidth is usually needed for the validity of the asymptotic approximations but this seems to have little or no adverse effect on the estimator variance.

Because standard variables in CEO pay-performance sensitivity regressions are apparently nonstationary, the second chapter, CEOs Are Really Paid Like Bureaucrats, uses the new estimator to evaluate the average long run relation between CEO compensation and company performance in major U.S. corporations. The results of the chapter indicate that the connection between CEO compensation and company performance is economically very weak. Estimated sensitivities of total compensation, which takes into account changes in executive stock and stock option values, to company market value are significantly lower than in the literature. The sensitivity of total compensation to firm accounting performance is of the same order of magnitude as to market performance. Sensitivities of narrower compensation measures to company market and accounting performance are economically negligible. Little support for relative performance evaluation is found, i.e., CEOs do not seem to be evaluated relative to other companies in the market, and some econometric problems in its estimation are demonstrated. Although point estimates of pay-performance sensitivities differ across industries, statistical tests cannot detect inter-industry differences in the sensitivities.

The third chapter, Limit Distributions for Linear Regressions in Nonstationary Panels when n and T Are of Same Order of Magnitude, demonstrates some of the problems that the pooled least squares estimator and other linear panel estimators suffer from in nonstationary panels. Phillips and Moon (1999) discussed linear regressions in unit root nonstationary panels and derived probability limits and limit distributions for estimators. They assumed n/T --> 0 in their joint limit distribution theory. The chapter derives joint limit distributions for the pooled least squares estimator in spurious and cointegrated panels when n/T --> k, where k is finite, and shows that although the limit distributions are still normal, there is generally a bias in the distribution. Allowing n/T --> k does not affect the variances of the limit distributions but merely shifts them by finite nonrandom factors that are proportional to root-k.