Lecture #10 Studenmund (2006): Chapter 10 Objectives • What is heteroscedasticity? • What are the consequences? • How is heteroscedasticity be identified? • How is heteroscedasticity be corrected? All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.1 10.2 Homoscedasticity Case f(Yi) . . . Var(i) = E(i2)= 2 x11=80 x12=90 x13=10 0 income x1i The probability density function for Yi at three different levels of family income, X1 i , are identical. All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.3 Homoscedastic pattern of errors yi . . . . . . . . . . . . . . . .. . . . .. . . . . . . . . . . . . . . .. . . . 0 The scattered points spread out quite equally All right reserved by Dr.Bill Wan Sing Hung - HKBU xi 10.4 Heteroscedasticity Case f(Yi) . . x11 x12 . Var(i) = E(i2)= i2 x13 income x1i The variance of Yi increases as family income, X1i, increases. All right reserved by Dr.Bill Wan Sing Hung - HKBU Heteroscedastic pattern of errors 10.5 . yi . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . Small i associated with small value of Xi 0 The scattered points spread out quite unequally All right reserved by Dr.Bill Wan Sing Hung - HKBU large i associated with large value of Xi xi Definition of Heteroscedasticity: Var(i) = E(i2) = i2 Two-variable regression: Yi = 0 + 1 X1i + i xy ^ 1= = wi Yi = wi (0 + 1Xi + i) x2 2 10.6 Refer to lecture notes Supplement #03A ^ ) = unbiased ^ = + w => E( 1 1 i i 1 1 Var (^1) = E (^1 - 1)2 = E (wi i)2 = E (w12 12 + w22 22 + …. + 2w1w2 1 2 + …) = w12 12 + w22 22 + … …..+ 0 + ... = i2 wi2 = i2 x2 ( xi2)2 = if 12 22 32 … i.e., heteroscedasticity i2 x i2 ^ Var (1) = 2 xi2 All right reserved by Dr.Bill Wan Sing Hung - HKBU if 12 = 22 = 32 = … i.e., homoscedasticity 10.7 Consequences of heteroscedasticity 1. OLS estimators are still linear and unbiased ^ 2. Var ( i )s are not minimum. => not the best => not efficiency => not BLUE 2 2 i ^ ^ Two-variable 3. Var ( 1) = instead of Var( )= 1 x2 x2 case ^i2 ^2 ^2) 2 4. = n-k-1 is biased, E( 5. t and F statistics are unreliable. ^ RSS = ^ SEE = 2 Y = 0+ 1 X + i Cannot be min. All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.8 Detection of heteroscedasticity 1. Graphical method : plot the estimated residual ( ^i ) or squared (^i 2 ) against the ^ ) or any independent predicted dependent Variable (Y i variable(Xi). Observe the graph whether there is a systematic pattern as: ^ 2 Yes, heteroscedasticity exists ^ Y All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.9 Detection of heteroscedasticity: Graphical method ^ 2 ^ 2 yes yes ^ Y ^2 yes ^ Y ^2 ^ Y ^ 2 yes yes ^ Y ^ 2 no heteroscedasticity ^ Y All right reserved by Dr.Bill Wan Sing Hung - HKBU ^ Y 10.10 Yes, heteroscedasticity Yes, heteroscedasticity Yes, heteroscedasticity no heteroscedasticity All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.11 Statistical test: (i) Park test H0 : No heteroscedasticity exists (homoscedasticity) i.e., Var( i ) = 2 H1 : Yes, heteroscedasticity exists i.e., Var( i ) = i2 Park test procedures: 1. Run OLS on regression: Yi = 1 + 2 Xi + i , obtain ^i 2. Take square and take log : ln ( ^i2) 3. Run OLS on regression: ln ( ^ i2) = 1* + 2* ln Xi + vi 4. Use t-test to test H0 : 2* = 0 (Homoscedasticity) Suspected variable that causes heteroscedasticity If t* > tc ==> reject H0 ==> heteroscedasticity exists If t* < tc ==> not reject H0 ==> homoscedasticity All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.12 Example: Studenmund (2006), Equation 10.21 (Table 10.1), pp.370-1 PCON: petroleum consumption in the ith state REG: motor vehicle registration TAX: the gasoline tax rate Park Test Practical In EVIEWS Procedure 1: May misleading All right reserved by Dr.Bill Wan Sing Hung - HKBU Graphical detection All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.13 Procedure 2: Obtain the residuals, take square and take log All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.14 Y Yˆ ˆ Yˆ Y ˆ Horizontal variable Scatter plot All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.15 10.16 ˆ vs. EG 2 ˆ vs. REG All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.17 ˆi vs. Yˆ 2 ˆ i vs. Yˆ log(ˆi ) vs. Yˆ All right reserved by Dr.Bill Wan Sing Hung - HKBU Procedure 3 & 4 10.18 log(ˆ 2 ) 0* 1* REG * If | t | > tc => reject H0 => heteroscedasticity Refers AlltorightStudenmund (2006), Eq.(10.23), pp.373 reserved by Dr.Bill Wan Sing Hung - HKBU 10.19 (ii)Breusch-Pagan test, or LM test H0 : homoscedasticity H1 : heteroscedasticity Var ( i ) = 2 Var ( i ) = i2 Test procedures: (1) Run OLS on regression: Yi = 0 + 1X1i + 2X2i +...+ qXqi + i obtain the residuals, ^i (2) Run the auxiliary regression: ^i2 = 0 + 1 X1i + 2X2i +… +qXqi + vi (3) Compute LM=W= nR2 Or F= , R2u / q (1 - R2u) / n-k (4) Compare the W and 2df (where the df is #(q) of regressors in (2)) if W > 2df ==> reject the Ho if F*> Fcdf ==> reject the Ho All right reserved by Dr.Bill Wan Sing Hung - HKBU Yi = 0 + 1X1i + 2X2i + 3X3i + i All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.20 10.21 FC(0.05, 5, 44) = 2.45 2(0.05, 5) = 11.07 2(0.10, 5) = 9.24 W= Decision rule: W > 2df ==> reject the Ho BPG test for a linear model PCON=0+1REG+2Tax+ The W-statistic indicates that the heteroscedasticity is existed. All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.22 FC(0.05, 5, 44) = 2.45 2(0.05, 5) = 11.07 2(0.10, 5) = 9.24 W= Decision rule: W > 2df ==> reject the Ho The BPG test for a transformed log-log model: log(PCON)=0+1log(REG)+2log(Tax)+ The W-statistic indicates that the heteroscedasticity is still existed. Therefore, a double-log transformation may not necessarily remedy the Heterocsedasticity. All right reserved by Dr.Bill Wan Sing Hung - HKBU (iiia) White’s general heteroscedasticity test (no cross-term) (The White Test) H0 : homoscedasticity H1 : heteroscedasticity 10.23 Var ( i ) = 2 Var ( i ) = i2 Test procedures: (1) Run OLS on regression: Yi = 0 + 1 X1i + 2 X2i + i obtain the residuals, ^i , (2) Run the auxiliary regression: ^i2 = 0 + 1 X1i + 2 X2i + 3 X21i + 4 X22i + vi (3) Compute W (or LM) = nR2 (4) Compare the W and 2df (where the df is # of regressors in (2)) if W > 2df ==> reject the Ho All right reserved by Dr.Bill Wan Sing Hung - HKBU (iiib) White’s general heteroscedasticity test (with cross-terms) 10.24 (The White Test) H0 : homoscedasticity H1 : heteroscedasticity Var ( i ) = 2 Var ( i ) = i2 Test procedures: (1) Run OLS on regression: Yi = 0 + 1 X1i + 2 X2i + i obtain the residuals, ^i , (2) Run the auxiliary regression: ^i2 = 0 + 1 X1i + 2 X2i + 3 X21i + 4 X22i + 5 X1i X2i + vi (3) Compute W (or LM) = nR2 Cross-term (4) Compare the W and 2df (where the df is # of regressors in (2)) if W > 2df ==> reject the Ho All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.25 No cross-term All right reserved by Dr.Bill Wan Sing Hung - HKBU With cross-term 10.26 FC(0.05, 5, 44) = 2.45 2(0.05, 5) = 11.07 2(0.10, 5) = 9.24 W= Decision rule: W > 2df ==> reject the Ho White test for a linear model PCON=0+1REG+2Tax+ The W-statistic indicates that the heteroscedasticity is existed. All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.27 FC(0.05, 5, 44) = 2.45 2(0.05, 5) = 11.07 2(0.10, 5) = 9.24 W= Decision rule: W > 2df ==> reject the Ho The White test for a transformed log-log model: log(PCON)=0+1log(REG)+2log(Tax)+ The W-statistic indicates that the heteroscedasticity is still existed. Therefore, a double-log transformation may not necessarily remedy the Heterocsedasticity. All right reserved by Dr.Bill Wan Sing Hung - HKBU Another example 8.4 (Wooldridge(2003), pp.258) 10.28 W= 2(0.05, 9) = 16.92 2(0.10, 9) = 14.68 Decision rule: W > 2df => reject the H0 The White test for a linear model PCON=0+1REG+2Tax+ The test statistic indicates heteroscedasticity is existed. All right reserved by Dr.Bill Wan Sing Hung - HKBU Testing the log-log model 10.29 W= 2(0.05, 9) = 16.92 2(0.10, 9) = 14.68 Decision rule: W < 2df => not reject H0 The White test for a log-log model The test statistic indicates heteroscedasticity is not existed Using the log-log transformation in some cases may remedy the heteroscedasticity, (But not necessary). All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.30 Remedy :Weighted Least Squares (WLS) Suppose : Yi = 0 + 1 X1i + 2 X2i + i E(i) = 0, E(i j )= 0 ij Vqr (i2) = i2 = 2 [f (ZX2i)] = 2Zi2 If Var(i2)=2Zi If all Zi = 1 (or any constant), homoscedasticity returns. But Zi can be any value, and it is the proportionality factor. In the case of 2 was known :To correct the heteroscedasticity Transform the regression: Yi 1 X1i X2i i =0 + 1 + 2 + Zi Zi Zi Zi Zi => Y* = 0 X0* + 1 X1* + 2 X2* + i* All right reserved by Dr.Bill Wan Sing Hung - HKBU Then each term divided by Zi 10.31 Why the WLS transformation can remove the heteroscedasticity? In the transformed equation where 1 i (i) E ( Z ) = Z E (i) = 0 i i 1 1 i 2 2 (ii) E ( Z ) = Z 2E (i ) = Z 2 Zi22 = 2 i i i (iii) E ( i Zi j )= Zj 1 E ( i j ) = 0 ZiZj These three results satisfy the assumptions of classical OLS. All right reserved by Dr.Bill Wan Sing Hung - HKBU If the residuals plot against X2i are as following : ^2 ^i + 0 - 10.32 X2 X2 These plots suggest variance is increasing proportional to X2i2. The scattered plots spreading out as nonlinear pattern. Therefore, we might expect 2 = Z 22 Z 2 = X 2 =>Z =X i i i 2i i 2i Hence, the transformed equation becomes Yi 1 X1i = 0 + 1 X2i X2i X2i X2i i + 2 + X2i X2i => Yi* = 1 X0* + 2 X1* + 3 + * Where *i satisfies the assumptions of classical OLS All right reserved by Dr.Bill Wan Sing Hung - HKBU Now this becomes the intercept coefficient Example: Studenmund (2006), Eq. 10.24, pp.374 10.33 C.V.=0.3392 All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.34 The correction is built in EVIEWS Use the weight (1/REG) to remedy the heteroscedasticity All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.35 OLS result WLS result Refers to Studenmund (2006), Eq.(10.28), pp.376 All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.36 W= FC(0.05, 5, 44) = 2.45 2(0.05, 5) = 11.07 2(0.10, 5) = 9.24 Decision rule: W < 2df ==> not reject Ho Now, after the reformulation The test statistic value indicates that the heteroscedasticity is not existed. All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.37 Alternative remedy of heteroscedasticity: reformulate with per capita PCONi REGi 0 1 2TAX i i POPi POPi All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.38 FC(0.05, 5, 44) = 2.45 2(0.05, 5) = 11.07 2(0.10, 5) = 9.24 W= Decision rule: W < 2df ==> not reject Ho Now, after the reformulation The test statistic value indicates that the heteroscedasticity is not existed. All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.39 W < 2df => not reject Ho All right reserved by Dr.Bill Wan Sing Hung - HKBU If the residuals plot against X2i are as following : ^i ^2 + 0 - 10.40 X2 X2 This plot suggests a variance is increasing proportional to X2i. The scattered plots spreading out as a linear pattern Therefore, we might expect i2 = Zi2 hi2 = X2i => hi = X2i The transformed equation is Yi 1 X1i X2i i = 0 + 1 + 2 + X2i X2i X2i X2i X2i => Yi* = 1 X0* + 2 X1* + 3 X2* + * All right reserved by Dr.Bill Wan Sing Hung - HKBU Transformation: divided by the squared root term --“@sqrt(X)”10.41 Yi = 0 X2i 1 + 1 X2i X1i X2i + 2 X2i + X2i All right reserved by Dr.Bill Wan Sing Hung - HKBU i X2i 10.42 2(0.05, 5) = 11.07 2(0.10, 5) = 9.24 W < 2df => not reject Ho calculate: the C.V. = 0.36928 Compare to the transformation divided by the REG, the CV of That one is smaller. All right reserved by Dr.Bill Wan Sing Hung - HKBU Example: Gujarati (1995), Table 11.5, pp.388 Simple OLS result : R&D = 192.99 + 0.0319 Sales (0.194) (3.830) All right reserved by Dr.Bill Wan Sing Hung - HKBU SEE = 2759 C.V. = 0.9026 10.43 10.44 White Test for heteroscedasticity W= 2(0.05, 2)= 5.9914 2(0.10, 2)= 4.60517 All right reserved by Dr.Bill Wan Sing Hung - HKBU Observe the shape pattern of residuals: linear or nonlinear? All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.45 Transformation equations: 1. ( =>(1) 2. ^ Y i Xi ^ R&D i ) = -246.67 (-0.64) 1 Xi ˆ 2 ˆ C.V . , where Y n k 1 10.46 ˆ 2 + 0.036 Xi (5.17) = -246.67 + 0.036 Sales (-0.64) (5.17) Yi 1 ( ) = 0 Xi Xi Xi + 1 Xi SEE = 7.25 C.V. = 0.8195 Compare the C.V. To determine which weight is appropriated ^ (2) R&D = -243.49 + 0.0369 Sales (-1.79) (5.52) All right reserved by Dr.Bill Wan Sing Hung - HKBU SEE = 0.021 C.V. = 0.7467 Transformation: divided by the squared root term --“@sqrt(X)”10.47 ^ Yi 1 = 0 Xi Xi +1 Xi All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.48 ^ Y 1 i = 0 Xi Xi +1 Xi calculate: the C.V. = 0.8195 All right reserved by Dr.Bill Wan Sing Hung - HKBU After transformation by @sqrt(x), the W-statistic indicates there is no heteroscedasticity 10.49 2(0.05, 2)= 5.9914 2(0.10, 2)= 4.60517 W < 2df => not reject Ho All right reserved by Dr.Bill Wan Sing Hung - HKBU After transformation by @sqrt(Xi), residuals still spread out All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.50 Transformation: divided by the suspected variable (Xi) ^ Y 1 i + 1 = 0 Xi Xi All right reserved by Dr.Bill Wan Sing Hung - HKBU 10.51 10.52 ^ Y i Xi = 0 1 Xi + 1 Calculate the C.V. = 0.7467 All right reserved by Dr.Bill Wan Sing Hung - HKBU After transformation divided by the suspected X, the W-statistic indicates there is no heterose\cedasticity 10.53 2(0.05, 2)= 5.9914 2(0.10, 2)= 4.60517 W < 2df => not reject Ho All right reserved by Dr.Bill Wan Sing Hung - HKBU After transformation divided by Xi, residuals spread out more stable10.54 All right reserved by Dr.Bill Wan Sing Hung - HKBU