Lecture 4:

1. Multicollinearity: High correlation between at least two independent variables. Correlatation:

• Strength of the relationship between two variables
• Positive or negative
• Values between -1 and +1

Special case for multicollinearity:

• Perfect multicollinearity, meaning perfect correlation and OLS is not possible. Example: include a dummy for every possible group/category instead of including one

More common:

• Multicollinearity, high correlation but not perfect correlation. Examples: include variables that are lagged values of one another of include variables that capture similar phenomena.
• Multicollinearity-intuition: In general nothing wrong with including correlated variables, but if x1 and x2 are perfectly/highly correlated, it is hard to identify the effect of x1 on y. Because whenever x1 changes, x2 changes with it.

What happens if we do OLS anyway?

• Hard to identify the individual impact of each x (which variable should take the credit for explaining variables in y)
• Larger standard errors and insignificance.
• Nonsensical coefficient signs and magnitudes (unbelievably large estimated and wrong/counterintuitive signs. So we cannot trust the results)

How can we check whether we have a multicollinearity problem?

• Check the data before estimation
• Correlation coefficients/Correlation matrixàCorrelation coefficients around 0.7-0.8 signal multicollinearity (But are absolute)  
• Variance inflation factor (VIF) à For one, but can also be for all. Measure by 1/(1-r2). Values higher than 10, multicollinearity is a problem
• For multicollinearity only look at the IV

How can the problem be solved?

• Increase the sample size, but, is that feasible?
• Drop one of the variables. If two variables measure the same thing, robustness check: first estimate the model with one variable, then with the other (similar results). But the question is whether you dropped an important/relevant variable
• Transform the highly correlated IVs. Log transformation, create a composite variable, combine collinear IVs. But does that make sense for your model? What about the interpretation of the regression coefficient?

2. Heteroscedasticity:  OLS assumption. Variance of the error term is constant over various values of the IVs. Dispersion of the error remains the same over the range of observations.
3. Heteroscedasticity: The OLS assumption does not hold. The error term does not have a constant variance. Dispersion of the error changes over the range of observations. à different variances

Problem:

• Dispersion of the error changes over the range of observations
• Why: Group of observations are different; follow different processes; different error terms

What happen if we do OLS anyways

• OLS assumption violated
• Biased standard errors
• Unreliable t-statistc
• Unreliable significance test
• Misleading conclusions about significance
• OLS estimators are not efficient

How to test for heteroscedasticity problem?

• Breusch-Pegan test
• White test
• Scatterplot of residuals: make a graph with the IV and the residuals of your regression. For each independent variable (multiple regression: Multiple plots). You want to see most of the scores concentrated in the center, no systematic patterns

How can we solve the problem?

• Weighted least squares: alternative estimator, each observation is weightedà observations with a higher (lower) variance get a lower (greater) weight in determining the regression coefficients.
• Calculate robust standard errors: adjust OLS standard errors for heterosceadaticity

Article:

• Corporate governance of foreign subsidiaries in MNEs (starting point)
• Focus on subsidiary boardà Key for a sound corporate governance
• How à Oversee performance on behalf of HQs, review strategic plans and internal policies, help integrate the subsidiary into the MNE, facilitate access to resources and knowledge
• Why are we doing this? à scant academic research on the roles of subsidiary boards and the factors that affect these roles, lack of empirical evidence, inconclusive results (we really don’t know much about it). Interesting and relevant for academics ads practitioners.
• In general: investigate the determinants of the roles of subsidiary board in MNEs, More specifically: Dual focus: strategy of the subsidiary and nationaly of the subsidiary directors
• Sample/data: survey data/questionnaires to CEOs. Bel-first database: subsidiaries operating in Belgium (1 host country) (HQS in 14 countries). Only the largest subsidiary (if more than one), more than 50 employees, not in financial industries. Final result à 428 subsidiaries with 83 responses
• Four roles of a subsidiary board:

1. Control: monitor decisions and evaluate performance of subsidiary
2. Strategy: provide advice
3. Coordination: transfer information/knowledge between HQs and subsidiary
4. Service: provide local knowledge, access to local resources

• Three strategic types of foreign subsidiaries:

1.  Local implementer: only in the local market, activities independently from the rest of the MNE
2. Specialized contributor: routine tasks, highly integrated into MNE operations
3. World Mandate: responsible for a broad scope of activities, involved in corporate strategy

• From theory to hypothesis:

1. Agency and resource dependence theory (H1a-d). The subsidiary is more involved in the control, strategy, coordination and service role, the local implementer subsidiary than in world mandate  and specialized contributor subsidiaries
2. Board internationalization and resource dependence theory (H2a-d). The board of the local implementer subsidiary is more involved in the control, strategy and coordination but less involved in service role. When more subsidiary directors are HQs country nationals

• In other words how to test the hypothesis?
• Dependent variables: constructed based on questions on control, strategy, service and coordination
• Independent variable: Local implementor or not à dummy, HQs country directors, proportion of directors who are HQs country nationals
• Other/control variables (what does literature say?)à subsidiary size, wholly owned, CEO tenure, HQs country
• From variable to analysis: OLS regression analysis, robust standard errors (heteroskedasticity)
• Are the hypothesis confirmed?

1. Hypothesis 1 is partly confirmed à subsidiary board, more involved in control, strategy and service roles, in local implementer subsidiariess
2. Hypothesis 2 is confirmed or strategy and coordination à board of the local implementer subsidiary, more involved in strategy and coordination roles when more directors are HQs country nationals.