Endogeneity in CEO power: A survey and experiment. Issue 3 (1st September 2016)

• Record Type: Journal Article
• Authors: Li, Frank
• Is Part Of: Investment analysts journal. Volume 45:Issue 3(2016)
• Page Start: 149
• Page End: 162
• Publication Date: 2016-09-01
• Subjects: Endogeneity problem -- CEO power -- CEO entrenchment -- firm performance -- GMM -- fixed effects -- instrumental variable -- lagged dependent variable; G34 -- G32 -- C58 -- D23 -- J33; Investment analysis -- Periodicals; Investments -- Africa, Southern -- Periodicals; Finance -- Periodicals; Finance; Investment analysis; Investments; Africa, Southern; Periodicals; 332.605
• DOI: 10.1080/10293523.2016.1151985
• Languages: English
• ISSNs: 2077-0227

Abstract

ABSTRACT: The endogeneity problem has always been one, if not the only, obstacle to understanding the true relationship between different aspects of empirical corporate finance. Variables are typically endogenous, instruments are scarce, and causality relations are complicated. As the first attempt to summarise different econometric methods that are commonly used to address endogeneity concerns in the context of corporate governance, we explore the relation between CEO power and firm performance, as an experiment, to illustrate how these methods can be used to mitigate the endogeneity problem and by how much. After carefully dealing with the endogeneity issues, we find strong evidence that the true relationship between CEO power and subsequent firm performance is negative, suggesting CEOs are overpowered in some firms. Furthermore, we show that all the prevailing econometric remedies are generally effective in mitigating the endogeneity problem to some degree (i.e., to correct the sign from positive to negative), but quantitatively the effects vary considerably. Among all the remedies, GMM has the greatest correction effect on the bias, followed by instrumental variables, fixed effect models, lagged dependent variables and the addition of more control variables. As for a combination of the methods, firm fixed effects, year fixed effects and the addition of more meaningful control variables appear to work as well, even without a valid instrumental variable.