Regress Y On X2 and X3 and Report The Results:: ECO 372 Assignment On Heteroscedasticity

Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 33

ECO 372 Assignment on Heteroscedasticity

1. Regress Y on X2 and X3 and report the results:

The above tables are generated when Y is regressed using the variables X2 and X3. Linear regression is used here. It must be mentioned beforehand that Y is Aggregate Demand, X2 is Income and X3 is price. The sample size is 60. The values for Beta (B) coefficients are .115 for X2 and .317 for X3, this means, provided other factors remain constant, a change in income by 1$ will lead to a change in Aggregate demand by $.115. Similarly a change in price by $1 will

1|Page

ECO 372 Assignment on Heteroscedasticity lead to a change in aggregate demand by $.317. A value for the intercept has been generated and this is -136.575. This can be interpreted as the value of aggregate demand when both income and price is equal to zero. The table above can be used to generate the following equation which will show the relationship between Y, X2 and X3:

Y= -136.575 + .115X2 + .317X3 t= (-5.748) (11.018) Se= (23.187) (0.010) (8.278) (0.038)

R Square= 0.889

2. Comment on the significance of the variables, overall significance and R2:

The value of R square is 0.889, this means that there is positive correlation between Y, X2 and X3 and 88.9% of the times this can be explained by the model. R square is a non decreasing function of the number of explanatory variables and as the number of regressors increases, R square almost always increases. On the other hand, adjusted R square, which is adjusted for the degree of freedom associated with the sums of squares, increases less than the unadjusted R square as the number of X variable increases. In short adjusted R square penalizes for the inclusion of irrelevant variables. This is something that R square fails to do. Here adjusted R square is slightly less than the unadjusted one.

2|Page

ECO 372 Assignment on Heteroscedasticity

Consider the t values of the intercept, Income and price. Testing for a 95% significance level and taking an = 0.05, a t-test is conducted to find out whether the computed t values fall within the critical regions and the null hypothesis can be accepted or rejected or not. The following are the conditions for the test:

The computed values of t for the intercept, income and price coefficients as seen from the table are -5.742, 11.018 and 8.278. As the sample size is well above 20 a t critical value of 2 is used here (marked t in the diagram). The t values for the slope coefficients of X2 and X3 will not fall in the acceptance region, which means they are significant. The t value for the intercept will be insignificant as it falls in the acceptance region

3|Page

ECO 372 Assignment on Heteroscedasticity Considering the overall significance, the F value that is computed in the ANOVA table must be considered. The value of F is computed to be 227.741, an F test is to be conducted to find out whether there is overall significance or not. Hence conducting an F test:

For a sample size of 60 and a degree of freedom of 2, at = 0.05 the critical value in the F distribution is 3.15, hence null hypothesis will be rejected and there will be overall significance. Whites Test A whites test is now performed to check for heteroscedasiticity. For this the error term is calculated. In this case the error term is calculated using the following formula: 1) Ui* = Yi-B2X2-B3X3 After this is obtained, Ui* is regressed against X2, X3, (X2)2, (X3)2 and X2X3. The following formula represents this test: 2) Ui*= 1 + 2X2 + 3X3 + 4(X2) + 5(X3) + 6(X2)(X3) The next page contains the tables for the output computed using SPSS.

4|Page

ECO 372 Assignment on Heteroscedasticity

Using the whites test where all variables, square of independent variables as well as the cross products of different variables have been used. Various changes can be observed in the test statistic. Notice the value of R, it is much larger at .998, this shows that there is not only positive correlation between the estimated error term and the variables, but the model explains the

5|Page

ECO 372 Assignment on Heteroscedasticity positive correlation between the dependant and independent variables almost 99.8% times. The standard error values are much larger when compared to those computed in the simple regression model. Notice the t values. The t values for Income is less that the t-critical value of the test, meaning that it is insignificant here. On the other hand the t-value of price is greater than the tcritical value and will be significant for the model. The t values of the square of income is significant while the t value for the square of price is insignificant. Greater investigations must be made by doing whites test in terms of income and price separately to find out the degree to which they affect Aggregate Demand. And then obtaining the R from this auxiliary regression and then multiplying it by the sample size (n)= n * R = (60*0.998)= 59.88 This follows the chi-square distribution with 5 degrees of freedom since in this auxiliary regression there are 5 regressors. This yields a chi-square critical value at 5% significance level of 16.7496.

H0: No Heteroscedasticity H1: Heteroscedasticity is present Since, n * R = (60*0.998) = 59.88 > X (5df) 16.7496 Therefore, as the computed value exceeds the chi-square critical values reject

H0 or the claim that there is no heteroscdasticity.

3. With respect to income, test for heteroskedasticity using Whites test and correcting it find the robust Standard error.
Previously the whites test was conducted using both price and income. Now the model will be tested for heteroskedasticity using Whites test Here Ui*I(Error term in terms of Income) is calculated using the following formula: 1) Ui*I=Yi X2 The error term Ui*I is regressed using whites test and testing only for income using the following formula: 2) Ui*I=1 + 2X2 + 4(X2)

6|Page

ECO 372 Assignment on Heteroscedasticity The following page contains the SPSS output for this test

As it can be seen the value of R is 0.974 which once again demonstrates a very strong positive correlation between the dependant variable Ui*I and the independent variables about 97.4% times. The standard error for Income is large at 359.934, but the square of this error term is almost negligible. Notice the t-values, both the t values are significant. Hence the model clearly shows that there is the presence of heteroskedasticity.

7|Page

ECO 372 Assignment on Heteroscedasticity

The R from this auxiliary regression is 0.974. The following are the conditions for this test: H0: No Heteroscedasticity H1: Heteroscedasticity is present Therefore, n * R = (60*0.974) = 58.44 > X (2df) 10.5966 As the computed value exceeds the chi-square critical value the claim made that there is no heteroscedasticity will be rejected. Using whites test to investigate Robust Standard Errors Var (B2) = Xi * Ui / (Xi) =1.058298048 Whites estimates of the variances are regarded to be consistent even in the presence of

heteroscedasticity.It is worth to mention that Whites heteroscedasticity corrected standard errors are considerably larger than the OLS standard errors. Hence it can be said that Whites test helps pick up those errors of independent variables often unaccounted for by OLS estimators

Correcting the problem A major flaw of the white procedure is that the estimators obtained are not very efficient as those methods that transform data to reflect specific types of heteroscedasticity. If it is assumed that the error variance is proportional to X such E(ui)=. Y/Xi = B1/Xi +B2 Xi+Ui / Xi =-60.23* 1/X + 0.154 */X Se (16.983) (0.014) Xi. Then,

These standard errors are regarded to be robust as well.

8|Page

ECO 372 Assignment on Heteroscedasticity

4. With respect to price level, test for heteroskedasticity using Whites test and correcting it find the robust Standard error.
Previously the whites test was conducted using both price and income. Now the model will be tested for heteroskedasticity using Whites test Here Ui*I(Error term in terms of Price) is calculated using the following formula: 1) Ui*P=Yi X3 The error term Ui*I is regressed using whites test and testing only for income using the following formula: 2) Ui*P=1 + 3X3 + 5(X3)

The next page contains the SPSS output for this test

9|Page

ECO 372 Assignment on Heteroscedasticity

As it can be seen that the value of R square is lower at 0.714 meaning that the positive correlation between the error term and price will be explained 71.4% of the times. Notice the value of the standard error for price, it is very healthy at 57.026. The t-tests show that the variable is significant and will be an explanatory variable in the model.

10 | P a g e

ECO 372 Assignment on Heteroscedasticity

Now testing for heteroscedasticity by running a new regression : (U) =1+ 2X3+ X3 The R from this regression is 0.714

H0: No Heteroscedasticity H1: Heteroscedasticity present Therefore, n * R = (60*0.714) = 42.84 > X (2df) 10.5966 As the computed value exceeds the chi-square critical value hence reject

H0 that there is no heteroscdasticity when the price level is considered only. Using whites test to investigate Robust Standard Errors Var (B2) = Xi * Ui / (Xi) = 13.635521

Correcting the error term

If we again assume that the error variance is proportional to X such E(ui)=. Y/Xi = B1/Xi +B2 Xi+ Ui / Xi = Se (Robust) 19693.587/Xi - 0.81Xi (3802.2186) (2.221)

Xi. Then,

As it can be seen the size of these robust errors are healthy and will be desirable for the model.

11 | P a g e

ECO 372 Assignment on Heteroscedasticity

Appendix/ Annex: The following contains all the spreadsheet data sheet, questions and software results. Questions: 5. Regress Y on X2 and X3 and report the results 6. Comment on the significance of the variables, overall significance and R2. 7. With respect to income, test for heteroskedasticity using Whites test and correcting it find the robust Standard error. 8. With respect to price level, test for heteroskedasticity using Whites test and correcting it find the robust Standard error. Things to be submitted: In Hard Copy: a) The report (just arrange the results appropriately answering the questions like a report from the software. Use whichever part you deem required.) b) Submit the data sheet, the questions and software results in the Annex. In Soft copy (via email by 19th of August, 12:00 midnight) a) The Excel/SPSS worksheet (if any) and the SPSS result sheet. Deadline: 20th of August, before the final exam! This deadline is non-extensible. However, I would be available for discussion on the following days: 9th August, 8:00 to 12:00 pm 13th August, 8:00 to 12:00 pm 19th August, 12:00 pm to 2:00 pm For trouble shooting e-mail me at my gmail address!!!

12 | P a g e

ECO 372 Assignment on Heteroscedasticity

13 | P a g e

ECO 372 Assignment on Heteroscedasticity

14 | P a g e

ECO 372 Assignment on Heteroscedasticity

15 | P a g e

ECO 372 Assignment on Heteroscedasticity

16 | P a g e

ECO 372 Assignment on Heteroscedasticity

17 | P a g e

ECO 372 Assignment on Heteroscedasticity

18 | P a g e

ECO 372 Assignment on Heteroscedasticity

19 | P a g e

ECO 372 Assignment on Heteroscedasticity

20 | P a g e

ECO 372 Assignment on Heteroscedasticity

21 | P a g e

ECO 372 Assignment on Heteroscedasticity

22 | P a g e

ECO 372 Assignment on Heteroscedasticity

23 | P a g e

ECO 372 Assignment on Heteroscedasticity

24 | P a g e

ECO 372 Assignment on Heteroscedasticity

25 | P a g e

ECO 372 Assignment on Heteroscedasticity

26 | P a g e

ECO 372 Assignment on Heteroscedasticity

27 | P a g e

ECO 372 Assignment on Heteroscedasticity

Regression

Notes Output Created Comments Input Data Active Dataset Filter Weight Split File N of Rows in Working Data File Missing Value Handling Definition of Missing User-defined missing values are treated as missing. Cases Used Statistics are based on cases with no missing values for any variable used. Syntax REGRESSION /MISSING LISTWISE /STATISTICS COEFF OUTS R ANOVA /CRITERIA=PIN(.05) POUT(.10) /ORIGIN /DEPENDENT YiXi0.5 /METHOD=ENTER invXiroot Xi0.5. F:\data3_done again.sav DataSet1 <none> <none> <none> 60 01-Aug-2011 19:41:28

Resources

Processor Time Elapsed Time Memory Required Additional Memory Required for Residual Plots

0:00:00.031 0:00:00.031 1812 bytes 0 bytes

28 | P a g e

ECO 372 Assignment on Heteroscedasticity

Variables Entered/Removed Variables Model 1 Variables Entered square root Xi, inverse xi root
a

Removed

Method . Enter

a. All requested variables entered.

Model Summary Adjusted R Model 1 R .903


a

Std. Error of the Estimate

R Square

Square .810

.816

2.58678

a. Predictors: square root Xi, inverse xi root b. For regression through the origin (the no-intercept model), R Square measures the proportion of the variability in the dependent variable about the origin explained by regression. This CANNOT be compared to R Square for models which include an intercept.

ANOVA Model 1 Regression Residual Total Sum of Squares 1719.559 388.103 2107.662
b

c,d

df 2 58 60

Mean Square 859.780 6.691

F 128.490

Sig. .000
a

a. Predictors: square root Xi, inverse xi root b. This total sum of squares is not corrected for the constant because the constant is zero for regression through the origin. c. Dependent Variable: Yi/root Xi d. Linear Regression through the Origin

Coefficients

a,b

29 | P a g e

ECO 372 Assignment on Heteroscedasticity


Standardized Unstandardized Coefficients Model 1 inverse xi root square root Xi a. Dependent Variable: Yi/root Xi b. Linear Regression through the Origin B -60.189 .134 Std. Error 16.781 .010 Coefficients Beta -.292 1.091 t -3.587 13.397 Sig. .001 .000

Regression

Notes Output Created Comments Input Data Active Dataset Filter Weight Split File N of Rows in Working Data File Missing Value Handling Definition of Missing User-defined missing values are treated as missing. Cases Used Statistics are based on cases with no missing values for any variable used. Syntax REGRESSION /MISSING LISTWISE /STATISTICS COEFF OUTS R ANOVA /CRITERIA=PIN(.05) POUT(.10) /ORIGIN /DEPENDENT Ui2 /METHOD=ENTER X3 X32. F:\data3_done again.sav DataSet1 <none> <none> <none> 60 01-Aug-2011 19:54:24

Resources

Processor Time

0:00:00.031

30 | P a g e

ECO 372 Assignment on Heteroscedasticity


Elapsed Time Memory Required Additional Memory Required for Residual Plots 0:00:00.032 1812 bytes 0 bytes

Variables Entered/Removed Variables Model 1 Variables Entered (X3)^2, Price(taka)


a

Removed

Method . Enter

a. All requested variables entered.

Model Summary Adjusted R Model 1 R .857


a

Std. Error of the Estimate

R Square

Square .725

.735

6.97425E6

a. Predictors: (X3)^2, Price(taka) b. For regression through the origin (the no-intercept model), R Square measures the proportion of the variability in the dependent variable about the origin explained by regression. This CANNOT be compared to R Square for models which include an intercept.

ANOVA Model 1 Regression Residual Total a. Predictors: (X3)^2, Price(taka) Sum of Squares 7.805E15 2.821E15 1.063E16 df

c,d

Mean Square 2 58 60 3.903E15 4.864E13

F 80.233

Sig. .000
a

b. This total sum of squares is not corrected for the constant because the constant is zero for regression through the origin.

31 | P a g e

ECO 372 Assignment on Heteroscedasticity


c. Dependent Variable: Ui^2 d. Linear Regression through the Origin

Coefficients

a,b

Standardized Unstandardized Coefficients Model 1 Price(taka) (X3)^2 a. Dependent Variable: Ui^2 b. Linear Regression through the Origin B 19693.587 -.181 Std. Error 3601.908 2.398 Coefficients Beta .868 -.012 t 5.468 -.075 Sig. .000 .940

Bibliography:
The following documents and publications were used while making this report. They are listed as follows: 1) "Python Econometrics."Wordpress. N.p., n.d. Web. 19 Aug. 2011. <http://adorio-research.org/wordpress/?p=1979>.

32 | P a g e

ECO 372 Assignment on Heteroscedasticity 2) "The t Distribution."Psychological Statistics at Missouri State University. N.p., n.d. Web. 18 Aug. 2011. <http://www.psychstat.missouristate.edu/introbook/sbk24m.htm>. 3) Thomas, Mark . "Economics 421 - Econometrics: White's Test for Heteroskedasticity."Economist's View. N.p., n.d. Web. 18 Aug. 2011. <http://economistsview.typepa 4) Levesque, Raynald . "Syntax."Sample Syntax. N.p., n.d. Web. 19 Aug. 2011. <pages.infinit.net/rlevesqu/SampleSyntax.htm>. 5) Coefficient of determination - Wikipedia, the free encyclopedia." Wikipedia, the free encyclopedia. N.p., n.d. Web. 18 Aug. 2011. <http://en.wikipedia.org/wiki/Coeff

33 | P a g e

You might also like