Một trong các giả thiết của mô hình hồi quy cổ điển là các biến độc lập phải phi ngẫu nhiên. Nếu chúng ta ước lượng mô hình có chuỗi thời gian mà các biến độc lập không dừng, thì khi đó giả thiết của OLS bị vi phạm ( kỳ vọng toán, phương sai và hiệp phương sai không đổi theo thời gian). Kiểm định nghiệm đơn vị là một tiêu chuẩn để kiểm định tính dừng. Dickey-Fuller đã đưa ra tiêu chuẩn kiểm định như sau: Giả thiết Ho: p=1 (chuỗi là không dừng), H1: p≠1 (chuỗi dừng). Nếu kết quả kiểm định cho P-value < α, bác bỏ H0, chấp nhận H1(Với α tương ứng với mức ý nghĩa 1%, 5%, 10%).
Bảng 4.11 Kết quả kiểm định tính dừng biến DIV
Panel unit root test: Summary Series: DIV
Date: 01/07/14 Time: 23:44 Sample: 2008 2012
Exogenous variables: Individual effects Automatic selection of maximum lags Automatic selection of lags based on SIC: 0
Newey-West bandwidth selection using Bartlett kernel Balanced observations for each test
Cross-
Method Statistic Prob.** sections Obs Null: Unit root (assumes common unit root process)
Levin, Lin & Chu t* -49.9737 0.0000 227 908
Null: Unit root (assumes individual unit root process)
Im, Pesaran and Shin W-stat -9.56933 0.0000 227 908 ADF - Fisher Chi-square 614.326 0.0000 227 908 PP - Fisher Chi-square 721.368 0.0000 225 900 ** Probabilities for Fisher tests are computed using an asymptotic Chi
Bảng 4.12 Kết quả kiểm định tính dừng biến MSO
Panel unit root test: Summary Series: MSO
Date: 01/07/14 Time: 23:46 Sample: 2008 2012
Exogenous variables: Individual effects Automatic selection of maximum lags Automatic selection of lags based on SIC: 0
Newey-West bandwidth selection using Bartlett kernel Balanced observations for each test
Cross-
Method Statistic Prob.** sections Obs Null: Unit root (assumes common unit root process)
Levin, Lin & Chu t* -694.262 0.0000 260 1040
Null: Unit root (assumes individual unit root process)
Im, Pesaran and Shin W-stat -136.775 0.0000 260 1040 ADF - Fisher Chi-square 1143.00 0.0000 260 1040 PP - Fisher Chi-square 1251.69 0.0000 259 1036 ** Probabilities for Fisher tests are computed using an asymptotic Chi
-square distribution. All other tests assume asymptotic normality.
Bảng 4.13 Kết quả kiểm định tính dừng biến INST
Panel unit root test: Summary Series: INST
Date: 01/071/14 Time: 23:46 Sample: 2008 2012
Exogenous variables: Individual effects Automatic selection of maximum lags Automatic selection of lags based on SIC: 0
Newey-West bandwidth selection using Bartlett kernel Balanced observations for each test
Cross-
Method Statistic Prob.** sections Obs Null: Unit root (assumes common unit root process)
Levin, Lin & Chu t* -96994.1 0.0000 218 872
Null: Unit root (assumes individual unit root process)
Im, Pesaran and Shin W-stat -11624.4 0.0000 218 872 ADF - Fisher Chi-square 1060.70 0.0000 218 872 PP - Fisher Chi-square 1184.08 0.0000 217 868 ** Probabilities for Fisher tests are computed using an asymptotic Chi
Bảng 4.14 Kết quả kiểm định tính dừng biến FORG
Panel unit root test: Summary Series: FORG
Date: 01/07/14 Time: 23:55 Sample: 2008 2012
Exogenous variables: Individual effects Automatic selection of maximum lags Automatic selection of lags based on SIC: 0
Newey-West bandwidth selection using Bartlett kernel Balanced observations for each test
Cross-
Method Statistic Prob.** sections Obs Null: Unit root (assumes common unit root process)
Levin, Lin & Chu t* -23526.7 0.0000 263 1052
Null: Unit root (assumes individual unit root process)
Im, Pesaran and Shin W-stat -3093.70 0.0000 262 1048 ADF - Fisher Chi-square 942.790 0.0000 262 1048 PP - Fisher Chi-square 1163.71 0.0000 262 1048 ** Probabilities for Fisher tests are computed using an asymptotic Chi
-square distribution. All other tests assume asymptotic normality.
Bảng 4.15 Kết quả kiểm định tính dừng biến STATE
Panel unit root test: Summary Series: STATE
Date: 01/07/14 Time: 23:56 Sample: 2008 2012
Exogenous variables: Individual effects Automatic selection of maximum lags Automatic selection of lags based on SIC: 0
Newey-West bandwidth selection using Bartlett kernel Balanced observations for each test
Cross-
Method Statistic Prob.** sections Obs Null: Unit root (assumes common unit root process)
Levin, Lin & Chu t* -12.4840 0.0000 89 356
Null: Unit root (assumes individual unit root process)
Im, Pesaran and Shin W-stat -4.15537 0.0000 88 352 ADF - Fisher Chi-square 209.531 0.0427 88 352 PP - Fisher Chi-square 214.888 0.0243 88 352
Bảng 4.16 Kết quả kiểm định tính dừng biến FS
Panel unit root test: Summary Series: FS
Date: 01/07/14 Time: 23:57 Sample: 2008 2012
Exogenous variables: Individual effects Automatic selection of maximum lags Automatic selection of lags based on SIC: 0
Newey-West bandwidth selection using Bartlett kernel Balanced observations for each test
Cross-
Method Statistic Prob.** sections Obs Null: Unit root (assumes common unit root process)
Levin, Lin & Chu t* -54.1197 0.0000 271 1084
Null: Unit root (assumes individual unit root process)
Im, Pesaran and Shin W-stat -23.7933 0.0000 271 1084 ADF - Fisher Chi-square 1046.37 0.0000 271 1084 PP - Fisher Chi-square 1491.40 0.0000 271 1084 ** Probabilities for Fisher tests are computed using an asymptotic Chi
-square distribution. All other tests assume asymptotic normality.
Bảng 4.17 Kết quả kiểm định tính dừng biến LEV
Panel unit root test: Summary Series: LEV
Date: 01/07/14 Time: 23:58 Sample: 2008 2012
Exogenous variables: Individual effects Automatic selection of maximum lags Automatic selection of lags based on SIC: 0
Newey-West bandwidth selection using Bartlett kernel Balanced observations for each test
Cross-
Method Statistic Prob.** sections Obs Null: Unit root (assumes common unit root process)
Levin, Lin & Chu t* -67.2686 0.0000 271 1084
Null: Unit root (assumes individual unit root process)
Im, Pesaran and Shin W-stat -15.5588 0.0000 271 1084 ADF - Fisher Chi-square 793.328 0.0000 271 1084 PP - Fisher Chi-square 1013.08 0.0000 271 1084 ** Probabilities for Fisher tests are computed using an asymptotic Chi
Bảng 4.18 Kết quả kiểm định tính dừng biến BETA
Panel unit root test: Summary Series: BETA
Date: 01/07/14 Time: 23:58 Sample: 2008 2012
Exogenous variables: Individual effects Automatic selection of maximum lags Automatic selection of lags based on SIC: 0
Newey-West bandwidth selection using Bartlett kernel Balanced observations for each test
Cross-
Method Statistic Prob.** sections Obs Null: Unit root (assumes common unit root process)
Levin, Lin & Chu t* -104.120 0.0000 271 1084
Null: Unit root (assumes individual unit root process)
Im, Pesaran and Shin W-stat -10.1756 0.0000 271 1084 ADF - Fisher Chi-square 663.037 0.0003 271 1084 PP - Fisher Chi-square 792.862 0.0000 271 1084 ** Probabilities for Fisher tests are computed using an asymptotic Chi
-square distribution. All other tests assume asymptotic normality.
Bảng 4.19 Tổng hợp kết quả kiểm định tính dừng các biến
Augmented Dickey-Fuller test statistic
P-value DIV 0.0000 FORG 0.0000 INST 0.0000 MSO 0.0000 STATE 0.0427 BETA 0.0003 F_S 0.0000 LEV 0.0000
Như vậy tất cả biến trong bài đều dừng ở mức ý nghĩa 1%, ngoại trừ biến STATE dừng ở mức ý nghĩa 5%.