BUS613: Advanced International Business

The data diagnostics generated a number of outliers that could distract the regression results as well as the normality assumptions. First, the studentized residuals are calculated and observations whose absolute values of residuals are more than 2.5 are excluded from the regressions. The highest (9.65) and lowest (-13.54) residuals are observations of AMCOL International Corp. and Freescale Semiconductor INC, respectively. Eliminating the observations with high residuals decreased the sample size from the initial 1730 to 1685.

Normality of residuals is tested by a command (iqr) in STATA. Inter-quartile range (iqr) assumes the symmetry of the distribution of residuals. The command reveals if there are severe outliers that are either three inter-quartile-ranges below the first quartile or three inter-quartile-ranges above the third quartile. The residuals do not happen to have normal distribution and there is one severe outlier that allows the rejection of normality at 5% level. Another diagnostics test, Shapiro-Wilk W test, appears to be significant and therefore rejecting the normality of residuals.

We utilized Breusch-Pagan test and Cameron & Trivedi's decomposition of IM-test for variance of fitted values of our dependent variable (ROA). The former test fails to reject (prob>chi2 = 0.28) homoscedasticity. However, the latter test confirms the existence of heteroscedasticity (p < 0.001).