Spatial Econometric Analysis Kuan-Pin Lin Portland State Univerisity Spatial Autoregressive Model with Autoregressive Disturbances SARAR(1,1) = SPLAG(1)+SPAR(1) y Wy Xβ ε ε
Download ReportTranscript Spatial Econometric Analysis Kuan-Pin Lin Portland State Univerisity Spatial Autoregressive Model with Autoregressive Disturbances SARAR(1,1) = SPLAG(1)+SPAR(1) y Wy Xβ ε ε
Spatial Econometric Analysis 5 Kuan-Pin Lin Portland State Univerisity Spatial Autoregressive Model with Autoregressive Disturbances SARAR(1,1) = SPLAG(1)+SPAR(1) y Wy Xβ ε ε Wε υ E (υ | X,W ) 0 Var (υ | X,W ) E (υυ ') I 2 Stability : 1/ min , 1/ max 1 Spatial Autoregressive Model with Moving Average Disturbances SARMA(1,1) = SPLAG(1)+SPMA(1) y Wy Xβ ε ε Wυ υ E (υ | X,W ) 0 Var (υ | X,W ) E (υυ ') 2I Stability : 1/ min , 1/ max 1 Spatial Autoregressive Model with ARMA Disturbances SARARMA(1,1,1) = SPLAG(1)+SPAR(1)+SPMA(1) y Wy Xβ ε ε Wε Wυ υ E (υ | X,W ) 0 Var (υ | X,W ) E (υυ ') 2I Stability : 1/ min , , 1/ max 1 Model Estimation Maximum Likelihood Estimation Log-Likelihood Function n n ε ' J ' Jε 2 L(; y, X,W ) ln(2 ) ln( ) 2 2 2 2 ln I W ln J , ε ( I W )y Xβ SPLAG(1) + … J SPAR(1) (I-W) SPMA(1) (I+W)-1 SPARMA(1,1) (I+W)-1(I-W) Model Estimation Maximum Likelihood Estimation Quasi Maximum Likelihood (QML) Estimator ˆ max arg L(; y, X, W ) ˆ ) L ( ˆ ) ˆ ( Var ' 2 1 ˆ ) L( ˆ ) 2 L( ˆ ) L( ' ' 1 Model Estimation SARAR(1,1) y Wy Xβ ε y Zδ (I W ) υ 1 Z Wy ε Wε υ X , δ β ' ' E (υ | X, W ) 0 Var (υ | X, W ) 2 I Var (ε | X, W ) 2 (I W ) '(I W ) 1 Cov(Wy , ε) 2W (I W ) 1 (I W ) 1 0 Cov(Wε, υ) 2W (I W ) 1 0 Model Estimation SARAR(1,1): Generalized Method of Moments Moment Functions (Kelejian and Prucha, 1998, 2009) ˆ εˆ y ˆWy Xβˆ IV estimator βˆ υ εˆ Wεˆ ' 2 E ( υυ ) I υ Wυ E ( υυ ' ) W [ E (υυ' )]W ' 2WW ' E ( υυ' ) W [ E (υυ' )] 2W Model Estimation SARAR(1,1): Generalized Method of Moments Sample moment functions are the same two equations of one parameter as in the spatial error AR(1) model. The efficient GMM estimator follows exactly the same as the spatial error AR(1) model with the IV estimator of the spatial lag model. Model Estimation SARAR(1,1) The Model y Wy Xβ ε ( I W )[( I W )y Xβ] υ ε Wε υ Estimate , b and simultaneously: QML Estimate , b and iteratively: IV/GMM/GLS ˆ , ˆ IV or 2SLS β GMM ˆ GLS ˆ ˆ β, Crime Equation Anselin (1988) SARAR(1) Model (Crime Rate) = a + b (Family Income) + g (Housing Value) + + W (Crime rate) + e , e = We + u GMM vs. QML Estimator GMM Parameter GMM s.e QML Parameter QML s.e 0.45602 0.17491 0.36806 0.14947 -0.1221 0.13571 0.16669 0.17286 b -1.0438 0.37611 -1.0259 0.44610 g -0.2537 0.08706 -0.28165 0.18534 a 43.916 10.738 47.784 6.9048 Q/L 2.6706 -182.23 Applications Geographically Weighted Regression (GWR) Limited Dependent Variables Spatial Heterogeneity Spatial Autocorrelation Spatial Probit and Spatial Tobit Models Spatial Inference Spatial Prediction Best Predictors Spatial Model Comparison References K.P. Bell, N.E. Bockstael, 2000, Applying the Generalized-Moments Estimation to Spatial Problems Involving Microlevel Daqta, Review of Economic s and Statistics, 82, 72-82. H. Kelejian, and I. R. Prucha, 2010, Specification and Estimation of Spatial Autoregressive Models with Autoregressive and Heteroskedastic Disturbances. Journal of Econometrics, 157, 53-67. Das, D., H. Kelejian, and I.R. Prucha, 2003. Small Sample Properties of Estimators of Spatial Autoregressive Models with Autoregressive Disturbances. Papers in Regional Science, 82, 1-26. L.F. Lee, 2007. GMM and 2SLS Estimation of Mixed Regressive Spatial Autoregressive Models. Journal of Econometrics, 137, 489514.