Analysis of Panel Data. Third Edition. Cheng Hsiao University of Southern California CAMBRIDGE UNIVERSITY PRESS

Transcription

1 Analysis of Panel Data Third Edition Cheng Hsiao University of Southern California CAMBRIDGE UNIVERSITY PRESS

2 Contents Preface to the ThirdEdition Preface to the Second Edition Preface to the First Edition page xvii 1 Introduction Introduction Advantages of Panel Data Issues Involved in Utilizing Panel Data Unobserved Heterogeneity across Individuais and over Time Incidental Parameters and Multidimensional Statistics Sample Attrition Outline of the Monograph 14 2 Homogeneity Tests for Linear Regression Models (Analysis of Covariance) Introduction '7 2.2 Analysis of Covariance '* 2.3 An Example 24 3 Simple Regression with Variable Intercepts Introduction Fixed-Effects Models: Least-Squares Dummy Variable Approach Random Effects Models: Estimation of Variance- Components Models Covariance Estimation Generalized Least-Squares (GLS) Estimation Maximum-Likelihood Estimation 45 ix xix xxi

3 X Contents 3.4 Fixed Effects or Random Effects AnExample Conditional Inference or Unconditional (Marginal) Inference Tests for Misspecification Models with Time- and/or Individual-Invariant Explanatory Variables and Both Individual- and Time-Specific Effects Estimation of Models with Individual-Specific Variables Estimation of Models with Both Individual and Time Effects Heteroscedasticity and Autocorrelation Heteroscedasticity Models with Serially Correlated Errors Heteroscedasticity Autocorrelation Consistent Estimator for the Covariance Matrix of the CV Estimator Models with Arbitrary Error Structure - Chamberlain jr-approach 69 Appendix 3A: Consistency and Asymptotic Normality of the Minimum-Distance Estimator 75 Appendix 3B: Characteristic Vectors and the Inverse of the Variance-Covariance Matrix of a Three-Component Model 77 4 Dynamic Models with Variable Intercepts Introduction The CV Estimator Random-Effects Models Bias in the OLS Estimator Model Formulation Estimation of Random-Effects Models Testing Some Maintained Hypotheses on Initial Conditions Simulation Evidence AnExample Fixed-Effects Models Transformed Likelihood Approach Minimum Distance Estimator Relations between the Likelihood-Based Estimator andthegmm Issues of Random versus Fixed-Effects Specification 119

4 Contents xi 4.6 Estimation of Dynamic Models with Arbitrary Serial Correlations in the Residuais Models with Both Individual- and Time-Specific Additive Effects 122 Appendix 4A: Derivation of the Asymptotic Covariance Matrix of Feasible MDE 129 Appendix 4B: Large N and T Asymptotics Static Simultaneous-Equations Models Introduction Joint Generalized Least-Squares Estimation Technique Estimation of Structural Equations Estimation of a Single Equation in the Structural Model Estimation of the Complete Structural System Triangular System Identification Estimation An Example 162 Appendix 5A Variable-Coefficient Models Introduction Coefficients that Vary over Cross-Sectional Units Fixed-Coefficient Model Random-Coefficient Model Coefficients that Vary over Time and Cross-Sectional Units The Model Fixed-Coefficient Model Random-Coefficient Model Coefficients that Evolve over Time The Model Predicting ß, by the Kaiman Filter Maximum-Likelihood Estimation Tests for Parameter Constancy Coefficients that Are Functions of Other Exogenous Variables A Mixed Fixed- and Random-Coefficients Model Model Formulation A Bayes Solution Random or Fixed Differences? Dynamic Random-Coefficients Models TwoExamples 212

5 xii Contents Liquidity Constraints and Firm Investment Expenditure Aggregate versus Disaggregate Analysis Correlated Random-Coefficients Models Introduction Identification with Cross-Sectional Data Estimation of the Mean Effects with Panel Data 223 Appendix 6A: Combination of Two Normal Distributions Discrete Data Introduction Some Discrete-Response Models for Cross-Sectional Data Parametric Approach to Static Models with Heterogeneity Fixed-Effects Models Random-Effects Models Semiparametric Approach to Static Models Maximum Score Estimator A Root-A/ Consistent Semiparametric Estimator Dynamic Models The General Model Initial Conditions A Conditional Approach State Dependence versus Heterogeneity Two Examples Alternative Approaches for Identifying State Dependence Bias-Adjusted Estimator Bounding Parameters Approximate Model Sample Truncation and Sample Selection Introduction An Example - Nonrandomly Missing Data Introduction A Probability Model of Attrition and Selection Bias Attrition in the Gary Income-Maintenance Experiment Tobit Models with Random Individual Effects Fixed-Effects Estimator Pairwise Trimmed Least-Squares and Least Absolute Deviation Estimators for Truncated and Censored Regressions 299

6 Contents xiii A Semiparametric Two-Step Estimator for the Endogenously Determined Sample Selection Model An Example: Housing Expenditure Dynamic Tobit Models Dynamic Censored Models Dynamic Sample Selection Models Cross-Sectionally Dependent Panel Data Issues of Cross-Sectional Dependence Spatial Approach Introduction Spatial Error Model Spatial Lag Model Spatial Error Models with Individual-Specific Effects Spatial Lag Model with Individual-Specific Effects Spatial Dynamic Panel Data Models Factor Approach Group Mean Augmented (Common Correlated Effects) Approach to Control the Impact of Cross-Sectional Dependence Test of Cross-Sectional Independence Linear Model Limited Dependent-Variable Model An Example - A Housing Price Model of China A Panel Data Approach for Program Evaluation Introduction Definition of Treatment Effects Cross-Sectional Adjustment Methods Panel Data Approach Dynamic System Panel Vector Autoregressive Models "Homogeneous" Panel VAR Models Heterogeneous Vector Autoregressive Models Cointegrated Panel Models and Vector Error Correction Properties of Cointegrated Processes Estimation Unit Root and Cointegration Tests Unit Root Tests Tests of Cointegration Dynamic Simultaneous Equations Models The Model 397

7 xiv Contents Likelihood Approach Method of Moments Estimator Incomplete Panel Data Rotating or Randomly Missing Data Pseudo-Panels (or Repeated Cross-Sectional Data) Pooling of Single Cross-Sectional and Single Time Series Data Introduction The Likelihood Approach to Pooling Cross-Sectional and Time Series Data AnExample Estimating Distributed Lags in Short Panels Introduction Common Assumptions Identification Using Prior Structure on the Process of the Exogenous Variable Identification Using Prior Structure on the Lag Coefficients Estimation and Testing Miscellaneous Topics Duration Model Count Data Model Panel Quantile Regression Simulation Methods Data with Multilevel Structures Errors of Measurement Nonparametric Panel Data Models A Summary View Benefits of Panel Data Increasing Degrees of Freedom and Lessening the Problem of Multicollinearity Identification and Discrimination between Competing Hypotheses Reducing Estimation Bias Generating More Accurate Predictions for Individual Outcomes Providing Information on Appropriate Level of Aggregation Simplifying Computation and Statistical Inference 469

8 Contents xv 13.2 Challenges for Panel Data Analysis Modeling Unobserved Heterogeneity Controlling the Impact of Unobserved Heterogeneity in Nonlinear Models Modeling Cross-Sectional Dependence Multidimensional Asymptotics Sample Attrition A Concluding Remark 473 References 475 Author Index 507 Subject Index 513