Executive Pay and Corporate Financial Performance. An Explorative Data Analysis

The relationship between executive pay and corporate financial performance con-tinues to attract wide academic, media and policy attention. The very high salariesenjoyed by senior executives in corporations primarily in the North American coun-tries are often contrasted with the relatively low pay received by executives in Eu-rope and Asia. Empirical research on executive pay has mainly concentrated on thepay-performance relationship. Although the adopted data sets were very dfferentwithin and across countries, the results are very similar and show very low pay-for-performance elasticities. Despite the similar results, several methodological issues arestill uncovered. Almost all studies assume linear or semi-log linear pay functions with-out applying a test of the adequate functional form. Furthermore, most models do notallow for variations across corporations, industries, countries and time. It it assumedthat pay functions are homogenous across corporations and variations are capturedby the fixed effects in the constants and by the assumption about the errors. Themethodological deciencies in the executive pay studies might also be a cause for thesurprisingly similar results across very dierent data sets. The purpose of this paper isto circumvent these diffculties of possible misspecification by adopting an explorativedata analysis that uses nonparametric methods which impose rather weak restrictionson the model. Since they estimate the functions locally, the estimated curvature doesnot get affected by outliers in the horizontal direction. Applying smoothing methods we start with the most general model. Our methods allow for a gradually closer lookby specifying the various objectives of investigation or the model we deduce from theprevious results. We also can detect possible heterogeneity between sample subgroups.In particular, we study the pay function for various industry groups and expect thatthey vary across industries but do not support homogeneity. Our results show quiteclearly that all these methodolo