Three Essays on Executive Compensation - KOPS - Universität ...

Which Pay for what Performance? Evidence from Executive
Compensation in Germany and the United States
in our specification. EBIT and net income are correlated with firm size measured
by total assets (the correlation is around 0.7), because in the cross section larger
firms generate higher earnings and income than smaller firms.
Multicollinearity
inflates the standard errors of the regression coefficients and thus causes downwardbiased
t-statistics. In our specification, however, multicollinearity is not the reason
why total assets is mostly insignificant in Table 2.4. There is not much variation
in total assets over time during the five-year sample period and variation in the
cross section is taken by the fixed effects 42 . A straight-forward approach to avoid
multicollinearity would be to drop firm size from the regression.
However, this
changes the specification and may cause an omitted-variable problem which is not
preferable over some degree of multicollinearity. Another approach in the literature
is orthogonalization of correlated variables. Specifically, this would imply to first
regress EBIT or net income, respectively, on total assets and then use the residual
of this regression as a performance measure in equation (2.1) instead. However, as
Kennedy (1982) and Pearce and Reiter (1985) show, the estimated coefficient of the
residual would be the same as the coefficient of EBIT or net income in the original
specification without orthogonalization (also the standard error of the coefficient
would not change). Thus the interpretation of the effect of EBIT or net income on
compensation would not change. Only the estimated coefficient of total assets would
change such that this coefficient shows the effect of total assets on compensation as
if there was no EBIT or net income in the regression 43 and thus from a different
specification without firm performance 44 . Since we do not interpret the effect of firm
size on compensation because it is largely captured by the fixed effects, we prefer
to keep our original specification with EBIT or net income, respectively, and total
assets as a control variable 45 .
42 It is a well established result that in the cross section firm size is a strong predictor for executive
compensation. Larger firms pay higher compensation also in Germany; see for example Haid and
Yurtoglu (2006). However, executive fixed effects capture this firm size effect. The variation in
total assets for a given firm during our five-year sample period is not a strong predictor for changes
in compensation in that firm. When we estimate the specifications in Table 2.4 without executive
fixed effects, our results remain qualitatively the same but total assets have significant explanatory
power for executive compensation.
43 See the critique on this issue in Clarke and Stone (2008).
44 The coefficient would be estimated from a misspecified model, because we do not believe that
executives are paid according to firm size alone and not for performance at all.
45 Yet another alternative would be to use as a regressor in equation (2.1) the ratio of EBIT
divided by total assets. In a robustness test we find that this ratio is positive and significant
and all other results remain qualitatively unchanged. However, anecdotal evidence from German
annual reports suggests that executive compensation is rather based on EBIT than on return on
assets measured by EBIT over total assets. Hence we prefer to use EBIT as a performance measure
79