Homoscedasticity Regression Analysis Yusep Suparman “Tugas hidup kita adalah memberi manfaat. Mulailah dari yang paling dekat yaitu diri sendiri” —Yusep Suparman 2 Let’s start! Homoscedastic means that the errors have the same or constant variances. 3 Formal formulation Previously we assumed that ⏷ ⏷ yi 0 1 x1i 2 x2i k xki i and cov i , x ji 0, i 1, 2, , n; j 1, 2, , k . Now we are assuming that ⏷ var i 2 ; i 1, 2, , n. the homoscedasticity assumption. 4 Violation of the assumption ⏷ It is called heteroscedaticity. var i i2 ; i 1, 2, , n at least one epsilon has a different variance ⏷ The effect on the OLS estimates var b j for j 1, 2, , k are not the minimum as well as inefficient → incorrect inferences see Verbeek (2002) p. 74-75 for the proof 5 What are the causes? Misspecified model Always reconsider your model specification if you detect heteroscedasticity! Detecting Homoscedasticity ⏷ Scatter plot residual against Xs ⏷ Scatter plot absolute residual against Xs Narrowing or widening patterns 7 Detecting Homoscedasticity (Breucsh-Pagan test) ⏷ Based on: i h z i α e 2 ⏷ Hypothesis H 0 : α 0, homoscedastic H 0 : α 0, heteroscedastic ⏷ Statistical test 2 nRe2 ~ 2j ⏷ Test criteria reject H 0 at significance level , if 2 12 , j Re2 is obtained from the auxilary regression ei2 0 z i α i i ei : residual from the primary regression z i : vector of observations of size j explaining error variances 8 Detecting Homoscedasticity (Breucsh-Pagan test) Procedure ⏷ Compute residual from the primary regression model . ⏷ Compute residual square. ⏷ Determine explanatory variables in the auxiliary. regression (often z i xi ) ⏷ Compute the R-square for the auxiliary regression. ⏷ Compute di chi-square statistic. ⏷ Determine a significant level and compute the test critical value. ⏷ Conclude. 9 Detecting Homoscedasticity (White test) ⏷ Hypothesis H 0 : α 0, homoscedastic H 0 : α 0, heteroscedastic ⏷ Statistical test 2 nRe2 ~ 2j ⏷ Test criteria reject H 0 at significance level , if 2 12 , j Re2 is obtained from the auxilary regression ei2 0 z i α i i ei : residual from the primary regression z i : vector of observations of size j explaining error variances. it include first moment and second moment as well as cross-product of the original regressor 10