Econometric Analysis of Panel Data • Panel Data Analysis – Linear Model • One-Way Effects • Two-Way Effects – Pooled Regression • Classical Model • Extensions Panel Data Analysis • Linear Model Representation yit xit' β it it ui vt eit t 1, 2,..., T (Ti ); i 1, 2,..., N ( N t ) (1) y i Xi β ui iTi v i ei or (2) y t Xt β u t vt i Nt et v1 u1 v u 2 2 vi , ut vTi u Nt xi' 1 x1,i1 yi1 ' y xi 2 x1,i 2 i2 yi ,X i ' yiTi xiTi x1,iTi x1' t x1,1t y1t ' y x 2t x1,2t 2t yt , Xt ' yNt t x Nt t x1, Nt t x2,i1 x2,i 2 x2,iTi x2,1t x2,2t x2, Nt t xK ,i1 ei1 1 e xK ,i 2 i2 2 ,β , ei xK ,iTi eiTi K xK ,1t e1t e xK ,2 t 2t , et xK , Nt t eNt t Linear Panel Data Model (1) • One-Way (Individual) Effects yit xit' β ui eit y i Xi β ui iTi ei (t 1, 2,..., Ti ; i 1, 2,..., N ) y Xβ u e u1iT1 y1 X1 e1 1 y X e u i 2 T2 2 y 2 , X 2 , β 2 , u , e u N iTN K y N XN e N Linear Panel Data Model (1) • One-Way (Time) Effects yit xit' β vt eit y i Xi β v i ei (t 1, 2,..., Ti ; i 1, 2,..., N ) y Xβ v e y1 X1 v1 e1 1 y X v e y 2 , X 2 , β 2 , v 2 , e 2 y X v K N N N e N Linear Panel Data Model (1) • Two-Way Effects yit xit' β ui vt eit y i Xi β ui iTi v i ei (t 1, 2,..., Ti ; i 1, 2,..., N ) y Xβ u v e Linear Panel Data Model (2) • One-Way (Individual) Effects yit xit' β ui eit y t Xt β u t et (i 1, 2,..., N t ; t 1, 2,..., T ) y Xβ u e y1 X1 1 u1 e1 y X u e y 2 , X 2 , β 2 , u 2 , e 2 yT XT K uT eT Linear Panel Data Model (2) • One-Way (Time) Effects yit xit' β vt eit y t Xt β vt i Nt et (i 1, 2,..., N t ; t 1, 2,..., T ) y Xβ v e v1i N1 y1 X1 1 e1 y X e v i 2 N2 2 y 2 , X 2 , β 2 , v , e y X vT i NT T T K eT Linear Panel Data Model (2) • Two-Way Effects yit xit' β ui vt eit y t Xt β ut vt i Nt et (i 1, 2,..., N t ; t 1, 2,..., T ) y Xβ u v e Panel Data Analysis • Between Estimator yit xit' β ui eit yi xi' β ui ei 1 yi Ti 1 t 1 yit , x T i Ti 1 ui u N ' i 1 t 1 x , ei T i Ti ' it Ti e t 1 it • If i 1 ui , then the pooled or population-averaged model is more efficient. N Panel Data Analysis • Linear Pooled (Constant Effects) Model yit xit' β ui eit yit xit' β u eit yit w it δ eit w it x ' it β 1 , δ u (t 1, 2,..., Ti ; i 1, 2,..., N ; NT i 1 Ti ) N y Wδ e Pooled Regression Model • Classical Assumptions – Strict Exogeneity E (eit | W) 0; Cov(wit , eit ) 0 – Homoschedasticity Var (eit | W) e2 – No cross section and time series correlation Var (e | W) e2 I NT Pooled Regression Model • Extensions – Weak Exogeneity E (eit | w i1 , w i 2 ,..., w iTi ) E (eit | Wi ) 0 E (eit | w i1 , w i 2 ,..., w it ) 0 E (eit | w it ) 0 – Heteroschedasticity Var (eit | Wi ) it2 Var (eit | Wi ) t2 Var (eit | Wi ) i2 Pooled Regression Model • Extensions – Time Series Correlation (with cross section independence for short panels) Cov(eit , eis | w it , w is ) ts , t s Cov(eit , e js | w it , w js ) 0, i j Var (eit | w it ) tt t2 Var (ei | Wi ) i Var (e | W) Ω 11 12 21 22 i Ti 1 Ti 2 1T i 2T i TiTi 1 0 0 2 Ω 0 0 0 0 N Pooled Regression Model • Extensions – Cross Section Correlation (with time series independence for long panels) Cov(eit , e jt | w it , w jt ) ij , i j Cov(eit , e js | w it , w js ) 0, t s Var (eit | w it ) i2 12 I 12 I 21I 22 I Var (e | W ) Ω IT N 1I N 2 I 1N I 2N I 2 N I Pooled Regression Model • Extensions – Cross Section and Time Series Correlation Var (eit | w it ) i ,tt i2 Cov(eit , eis | w it , w is ) i ,ts i ts , t s Cov(eit , e jt | w it , w jt ) ij , i j Cov(eit , e js | w it , w js ) ij ts , t s 12 R 12 R 2 R R 21 2 Var (e | W ) Ω R N 1 R N 2 R 1 R 21 T 1 1N R 2N R N2 R 12 1 T 2 1T 2T 1 Alternative Pool Regression Models • Between (Group Means) Estimator yit xit' β u eit yi xi' β u ei • First-Difference Estimator yit yit 1 (xit' xit' 1 )β (eit eit 1 ) yit xit' β eit • Within (Group Mean Deviations) Estimator yit yi (xit' xi' )β (eit ei ) Pooled Regression: OLS • Classical Model Estimation (OLS) N δˆ OLS ( W ' W ) 1 W ' y i 1 Wi' Wi 1 N ' W yi i i 1 ˆ (δˆ ) ˆ 2 ( W ' W) 1 ˆ 2 N W ' W Var OLS e e i 1 i i ˆ e2 eˆ ' eˆ / ( NT K ) eˆ y Wδˆ 1 ˆ (δˆ ) is inconsistent • Variance estimator Var because of heteroscedasticity and autocorrelation. OLS Pooled Regression: OLS • Panel-Robust Variance-Covariance Matrix – Adjusting general heteroscedasticity and serial correlation within panel Var (δˆ ) E[(δˆ δ)(δˆ δ) '] ( W' W) 1 W' E (ee ') W( W' W) 1 1 i 1 W Wi i 1 W E (e e )Wi i 1 W Wi N ' i N ' i 1 ' i i N ' i 1 ˆ (δˆ ) N W ' W N W 'eˆ eˆ ' W N W ' W Var i 1 i i i 1 i i i i i 1 i i 1 1 N T N T T N T i 1 t i 1 w it w it' i 1 t i 1 si1 w it w is' eˆit eˆis i 1 t i 1 w it w it' eˆ i y i Wi δˆ , eˆit yit w it δˆ 1 Pooled Regression: GLS • The Model y Wδ e E (e | W) 0 Var (e | W) Ω R 12 12 21 22 N 1 N 2 1N 2N N2 1 R 21 T 1 12 1 T 2 1T 2T 1 • Generalized Least Squares (GLS) 1 ˆ 1W W'Ω ˆ 1y, Var ˆ 1W ˆ (δˆ ) W'Ω δˆ GLS W'Ω GLS 1 – If cross sections are independent (short panels) N ˆ 1W δˆ GLS i 1 Wi' i i – where ˆ is i 1 1 ˆ ) N W' ˆ ˆ ˆ 1W W y , Var ( δ i1 i i GLS i i i i 1 N ' i the consistent estimator of i 1 Pooled Regression: GLS • Heteroscedasticity 1 T 2 ˆ t 1 eˆit T 2 i • Cross Section Correlation 1 T ˆ ij t 1 eˆit eˆ jt T • Time Series Correlation ˆts Pooled Regression: GLS • Examples of Time Series Correlation – Equal-Correlation – AR(1) 1 if t s ts if t s if t s 1 ts |t s| if t s – Stationary(1) if t s 1 ts if | t s | 1 0 otherwise – Nonstationary(1) if t s 1 ts ts if | t s | 1, ts st 0 otherwise Model Extensions • • • • Time-invariant regressors Random regressors Lagged dependent variables Dynamic models Example: Investment Demand • Grunfeld and Griliches [1960] Iit i Fit Cit it – i = 10 firms: GM, CH, GE, WE, US, AF, DM, GY, UN, IBM; t = 20 years: 1935-1954 – Iit = Gross investment – Fit = Market value – Cit = Value of the stock of plant and equipment Example: Investment Demand • Pooled Model (Population-Averaged Model) Iit Fit Cit it • Classical OLS • Panel-Robust OLS • Feasible GLS – Heteroscedastcity – Autocorrelation
© Copyright 2026 Paperzz