ECON
LINREG2

# A popular test for heteroscedasticity is the breusch

• Notes
• 29

This preview shows pages 19–22. Sign up to view the full content.

A popular test for heteroscedasticity is the Breusch-Pagan 6 test. Given that E [ U 2 j | X j ] ' g ( ( 0 % ( 1 X j ) for some unknown function g (.). (38) the Breusch-Pagan test tests the null hypothesis H 0 : ( 1 ' 0 ] E [ U 2 j | X j ] ' g ( ( 0 ) ' F 2 , say (39) against the alternative hypothesis H 0 : ( 1 0 ] E [ U 2 j | X j ] ' g ( ( 0 % ( 1 X j ) ' R ( X j ), say . (40) Under the null hypothesis (39) of homoskedasticity the test statistic of the Breusch-Pagan test has a distribution 7 , and the test is conducted right-sided. χ 2 1 12. How close are OLS estimators ? The ice cream data in Table 1 is not based on any actual observations on sales and temperature; I have picked the numbers for and quite arbitrarily. Therefore, there is no way X j Y j to find out how close the OLS estimates are to the unknown parameters α ˆ α ' & 0.25, ˆ β ' 1.5 and β . Actually, we do not know either whether the linear regression model (2) and its assumptions are applicable to this artificial data. In order to show how well OLS estimators approximate the corresponding parameters I

This preview has intentionally blurred sections. Sign up to view the full version.

8 Via the EasyReg International menus File 6 Choose an input file 6 Create artificial data. Rather than generating one random sample of size n = 1000 and then using subsamples of sizes n = 10 and n = 100, these samples have been generates separately for n = 10, n = 100 and n = 1000. 20 have generated random samples 8 for three sample sizes: n = 10, n = 100 and ( Y 1 , X 1 ),...,( Y n , X n ) n = 1000, as follows. The explanatory variables have been drawn independently from the X j χ 2 1 distribution, the regression errors have been drawn independently from the N(0,1) U j distribution, and the ‘s have been generated by Y j Y j ' 1 % X j % U j , j ' 1,2,..., n . (41) Thus, in this case the parameters α and β in model (2) are α = 1 and β = 1, and the standard error of is σ = 1. Moreover, note that the Assumptions I * -IV * hold for model (41). U j The true R 2 can be defined by R 2 0 ' 1 & E [ SSR ] E [ TSS ] ' 1 & ( n & 2) σ 2 ' n j ' 1 E [( Y j & ¯ Y ) 2 ] . In the case (41), σ 2 ' 1, μ Y ' E ( Y j ) ' 1 % E ( X j ) ' 2, ' n j ' 1 E [( Y j & ¯ Y ) 2 ] ' E ' n j ' 1 ( Y j & μ Y ) & ( ¯ Y & μ Y ) 2 ' E ' n j ' 1 ( Y j & μ Y ) 2 & n ( ¯ Y & μ Y ) 2 ' ( n & 1)var( Y j ) and var( Y j ) ' E [( X j & 1 % U j ) 2 ] ' E [( X j & 1) 2 ] % E [ U 2 j ] ' E [( X j & 1) 2 ] % 1 ' 3, because is distributed and therefore has the same distribution as and it can be shown X j χ 2 1 U 2 j , that for standard normal random variables Thus, the true R 2 in this case U j , E [( U 2 j & 1) 2 ] ' 2. is R 2 0 ' 1 & n & 2 3( n & 1) ' 2 n & 1 3 n & 3 . 0.7037 for n ' 10 0.6700 for n ' 100 0.6670 for n ' 1000 The estimation results involved are given in Table 2:
21 Table 2 : Artificial regression estimation results ˆ β ˆ α SER ( ' ˆ σ ) R 2 n estimate : 1.11748 0.55912 0.919045 0.8842 10 ( t & value ): (7.817) (1.675) estimate : 1.03309 0.96028 0.992502 0.8284 100 ( t & value ): (21.753) (8.237) estimate : 1.02360 0.98518 0.983608 0.6899 1000 ( t & value ): (47.124) (26.037) Even for a sample size of n = 10 the OLS estimator is already pretty close to its true value 1, ˆ β and the same applies to , but is too far away from the true value α = 1. However, for n = ˆ σ ˆ α 100 the OLS estimators and deviate only about ±4% from their true values α = β = 1, and ˆ β ˆ α ˆ σ deviates about -1% from its true value 1. In the case n = 1000 these deviations reduce to about ±2%. The R 2 's are too high, and only for n = 1000 is the R 2 reasonably close to its true value.

This preview has intentionally blurred sections. Sign up to view the full version.

This is the end of the preview. Sign up to access the rest of the document.

{[ snackBarMessage ]}

### What students are saying

• As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

Kiran Temple University Fox School of Business ‘17, Course Hero Intern

• I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

Dana University of Pennsylvania ‘17, Course Hero Intern

• The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

Jill Tulane University ‘16, Course Hero Intern