Chapter 53
PANEL DATA MODELS: SOME RECENT DEVELOPMENTS*
MANUEL ARELLANO
CEMFI, Casado del Alisal 5, 28014 Madrid, Spain
BO HONORE
Department of Economics, Princeton University, Princeton, New Jersey 08544
Contents
Abstract 3231
Keywords 3231
1 Introduction 3232
2 Linear models with predetermined variables: identification 3233
2.1 Strict exogeneity, predeterminedness, and unobserved heterogeneity 3233
2.2 Time series models with error components 3241
2.2 1 The AR( 1) process with fixed effects 3242
2.2 2 Aggregate shocks 3245
2.2 3 Identification and unit roots 3246
2.2 4 The value of information with highly persistent data 3247
2.3 Using stationarity restrictions 3248
2.4 Models with multiplicative effects 3250
3 Linear models with predetermined variables: estimation 3255
3.1 GMM estimation 3255
3.2 Efficient estimation under conditional mean independence 3259
3.3 Finite sample properties of GMM and alternative estimators 3262
3.4 Approximating the distributions of GMM and LIML for AR(l) models when the number
of moments is large 3264
4 Nonlinear panel data models 3265
5 Conditional maximum likelihood estimation 3267
5.1 Conditional maximum likelihood estimation of logit models 3268
5.2 Poisson regression models 3269
6 Discrete choice models with "fixed" effects 3270
7 Tobit-type models with "fixed" effects 3272
7.1 Censored regression models 3272
7.2 Type 2 Tobit model (sample selection model) 3275
7.3 Other Tobit-type models 3276
7.4 Monotone transformation models 3278
7.5 Nonparametric regression and fixed effects 3280
7.6 Relationship with estimators for some cross sectional models 3280
8 Models with lagged dependent variables 3282
8.1 Discrete choice with state dependence 3283
8.2 Dynamic Tobit models 3286
8.3 Dynamic sample selection models 3287
9 "Random" effects models 3287
10 Concluding remarks 3290
References 3290