Executive remuneration determinants: New evidence from meta-analysis

Abstract This meta-analysis takes stock of 121 C.E.O. pay studies published between 1998 and 2018 with the objective of identifying the main drivers of C.E.O. pay from a global perspective and contributing to the agency vs managerial debate on this ground. The meta-results disclose a positive C.E.O. pay–performance correlation (the highest correlation coefficient corresponds to Earnings per share with a 34%) as the agency theory prescribes and the governance policies promote. However, firm size still predominates as the main driver of C.E.O. pay (correlation coefficient is around 44%) according to managerial premises. Moreover, our results reconcile both approaches because results of the meta-regressions suggest that larger companies and more independent boards strengthen the pay–performance association. Additional analyses of moderating factors on C.E.O. pay forces do not provide robust conclusions, though, they suggest: (1) weak impact, if any, of both the Cadbury Report and the S.O.X.; and (2) lack of homogeneity in the banking industry despite its specific regulation.

Analysis (M.A.R.A.) on the main C.E.O. pay forces providing new evidence on the agency and M.P.T. frameworks.
This study could be of interest not only for regulators and standard setters but also for other stakeholders, because the inverse causal relationship also applies: Abowd (1990), after examining C.E.O. pay in 250 large firms during 1981-1986, found that increases in the link compensation-shareholder return enhanced firm performance. Thus, the efficacy of governance regulations would benefit shareholders and third parties interested in the firm's performance. Researches can also benefit from a structured and systematic review of the, to date, published results.

Literature review and research questions
Two main drivers of C.E.O. pay, i.e., firm performance and firm size, and their related theories (agency and managerial theory, respectively) constitute the core of the academic debate and provide the basis for the first research question of this study.
On the one hand, scholars in economics and finance commonly follow the agency theory (Fama & Jensen, 1983;Jensen & Meckling, 1976), which advocates that the optimal contract is the one that links C.E.O. pay with performance, controlling for firm risk measures, because it closely aligns the interests of shareholders (principal) and the managers (agent) and, consequently, it reduces agency problems.
How to operationalise the construct performance is also debatable: The use of market-based variables, such as return to shareholders, market to book value or Tobin's Q seem to better attach shareholder and manager interests. Conversely, Bertrand and Mullaintathan (2001) claim that the stock market evolution is not entirely controlled by managers and, therefore, the use of accounting variables, i.e., R.O.A. and R.O.E. should be desirable. Worth noting, the extensive earnings management literature evidences the danger of employing accounting measures that might suffer from manipulation.
On the other hand, rather than contradicting the agency theory, the M.P.T. explains why the C.E.O. compensation is, in many cases, part of the problem, rather than the solution itself (Bebchuk & Fried, 2003;2004) and the reason for a pay-size correlation higher than the pay-performance one (Tosi et al., 2000;Van Essen et al., 2015). Executives prefer to link their remuneration to the firm size because they exert more control over the firm growth (through new investments and/or acquisitions) and managing bigger firms also leads to more power and prestige. In addition, the increasing organisational complexities and human capital needs of growing companies seem to better justify their remuneration (Chalmers, Koh, & Stapledon, 2006).
Then, C.E.O.'s preferences (higher pay and lower pay-performance association) would prevail upon the shareholder's ones (lower pay and higher pay-performance association) in those firms with weaker governance mechanisms and higher C.E.O. power over-the-board. In this context, C.E.O.s might influence board decisions on the rewarding agreements that are prone to satisfy C.E.O.s rather than shareholder's interests. Hence, remuneration might be higher and tighter to firm size than it should be desirable.
In consequence, according to M.P.T., we expect 'good' boards, that is, boards that are not too big to face problems of coordination and communication (Jensen, 1993;Yermack, 1996); active (Vafeas, 1999) and independent (Fama & Jensen, 1983;Core, Holthausen, & Larcker, 1999) result in less C.E.O. pay and higher pay-performance association. However, some studies reported a positive association between Board Independence and C.E.O. pay because the external members seem to be more influenced by C.E.O.s (Wade, O'Reilly, & Chandratat, 1990;Lambert, Larcker, & Weigelt, 1993;Boyd, 1994). In this vein, C.E.O. duality (C.E.O. that also chairs the board) is also expected to influence on the rewarding agreements.
Researchers have also explored a wide assortment of C.E.O.-related variables: Commonly, C.E.O. Age and C.E.O. Tenure control for the superior skill management associated with higher experience and expertise that triggers, ultimately, higher executive compensation.
Ownership concentration is negatively associated with C.E.O. pay, because a high number of shareholders hinders a good coordination and supervisory function (Schwalbach, 1990;Core et al., 1999). The relationship between firm's leverage and C.E.O. pay also commonly turns up to be negative, because financial institutions refuse financing firms without minimal governance principles (Jensen, 1986).
However, empirical evidence confirming the M.P.T. is not conclusive (Murphy, 2002). For instance, the simultaneous increase in C.E.O. pay and either the increase in the board independence (Conyon, 2006;Hall & Murphy, 2003) or shortened C.E.O. tenure (Kaplan, 2008) raise doubts over managerial premises. Conversely, Core et al. (1999) concluded that C.E.O. earns greater compensation when governance structures are less effective.
The M.A.s provided by Tosi et al. (2000) and Van Essen et al. (2015) conclude that firm size explained more significantly the variance in total C.E.O. pay than performance measures. This may support the transcendence of managerial preferences supported by M.P.T.
However, none of them includes relevant governance measures (Dodd-Frank Act of 2010, Section 952; FRC, 2010 or E.U. Directive 2013/36/E) and the research conducted by Van Essen et al. (2015) is only focused on U.S. firms. Therefore, since this study aims to update and test in a multinational setting the M.P.T./agency theory, we posit the following research question: RQ1: According to prior reported results, which are the variables that exhibit the highest association with C.E.O. pay?

Moderating factors
Governance codes and regulation: the Cadbury and Greenbury reports and the S.O.X During the period that covers this M.A., the evolution of pay-performance might have been shaped by the successive C.G. Codes and regulations. In particular, we explore the influence of the following moderating factors: 1. The issuance of C.G. recommendations. Ozkan (2011) revealed that the aim of the British Cadbury (1992) and Greenbury (1995) Reports to more closely link C.E.O. pay to firm performance had not been totally effective. Likewise, the results of  suggested that the impact of Cadbury reforms had been disappointing, the pay-performance link had been reduced, and the pay-size link had been reinforced. 2. The approval of mandatory legislation. In addition to the typical 'complain or explain' approach of G.C. codes, in the U.S., mandatory legislation such as S.O.X.  (Chen, Jeter, & Yang, 2015) and significant increases in total compensation (Wang, 2010).
Due to the limited availability of publications in other countries, we restrict the exploration to how British Cadbury and Greenbury Codes and the American S.O.X. moderate reported findings, by investigating this research question: Since the banking industry is subject to specific regulation, their published results might offer homogeneity. However, in most studies, the sample does not specify whether they include the financial industry or not and, therefore, we cannot disentangle the correlation coefficients of the banking industry from the remaining industries. Thus, we explore the following research question in order to identify whether the C.E.O. pay drivers are homogeneous in the financial sector: RQ3: Does the industry (financial vs total industry) explain the heterogeneity of prior reported results?

Dependent variables
C.E.O. pay investigations have explored the two components of the Total compensation, that is, Cash and Non-cash components, the latter pursuing to increase pay-performance sensitivity. However, Buck, Liu, and Skovoroda (2008) confirmed that, while increasing C.E.O.'s total rewards, the presence of non-cash incentives is associated with reductions in pay-performance sensitivity. Brick, Palmon, and Wald (2006) also split the dependent variable into C.E.O. compensation and Executive compensation and concluded that both variables were correlated, revealing possible problems of complicity inside the companies.
In consequence, we have explored whether the operationalisation of the construct C.E.O. pay through the above mentioned variables also moderate prior findings, through the following research question: 'compensation performance', 'pay sensitivity' and 'pay elasticity', the initial search reported 1,343 articles. After discarding duplicates and studies from different subjects, the initial sample consists of 225 articles. Table 1 displays the exclusion criteria which trigger a final sample of 104 publications. In some publications there are more than one statistical analysis over independent samples that are suitable for the M.A. Hence, the meta-data is nourished by the results of 121 regressions executed over either absolute values (98) or incremental values (23) of the exploratory variables.
The final sample (details in Table 2) covers a wide range of countries. Unsurprisingly, the Anglo-Saxon countries 1 (U.S., U.K., Australia, Canada) predominate (49 studies) being the U.S., the most analysed environment (31 investigations). In second place, the continental Europe is the focus of 33 studies. Finally, in the Asian region, China heads the list (12 studies out of 30).
The vast majority of studies (81 out of 104) cover periods prior to the 2008 global financial crisis and only 23 investigations explore samples collected during and after this event.

Research design
All variables related to size, leverage, performance and C.G. characteristics of the companies are specified in Appendix 1. This study applies M.A. techniques introduced by Hunter, Schmidt, and Jackson (1982) as it follows: 1. The Pearson correlation coefficients (r) 2 reported in the studies of the sample are the data source to estimate the global effect size of the relationship between executive-pay and the explanatory variables. 3 In order to assess those coefficients, the scale developed by Cohen (1988) has been applied. 2. We assumed homogeneity in the results if 75% or more of the observed variance was explained by the sampling error and if the statistic of the Q test was not significant. Otherwise, when effect magnitudes were heterogeneous, we tried to identify the moderating variables chosen from our previous narrative review. 3. The so-called 'filed drawer problem' 4 (or publication bias towards significant results) results in higher M.A. coefficients than they should be otherwise (Wolf, 1986, p.37). In order to address this issue, we computed the Safe N (Rosenthal, 1979). 4. To identify whether the exploratory variables significantly influence the pay-performance association, we run a meta-regression following the M.A.R.A. procedure (Lipsey & Wilson, 2001) with the modifications proposed by Harbord and Higgins (2008), which include the improvement of the algorithm for the estimation of the between-study variance by residual maximum likelihood (R.E.M.L.) and the modifications suggested by Knapp and Hartung (2003). The dependent variable is the effect size of the association between performance and C.E.O. pay, computed from t-statistics and degrees of freedom of primary studies (Greene, 2008).

Effect size of the main variables
Columns 1 to 5 in Table 3    # Zrm is significant at 5%, that is, the 95% interval of confidence does not include zero.
performance and our variable of interest. In particular, 67 studies explore the impact of Share return on C.E.O. pay: 40 offered a positive significant relationship, one displayed negative coefficient and 26 resulted in no-significant results. The correlation coefficient offers a positive value of 0.069 included in the 95% confidence interval. We can claim that there is no publication bias, because we would need 21,195 studies (Safe N) with null results to fail to reject the null hypothesis of no significant effect of Share return on C.E.O. pay. The hypothesis of homogeneity is consistently refused: Only 5.701% of the observed variance is due to sampling error and the X 2 coefficient is significant at 1%, therefore the differences within the published outputs are due to unobserved/underlying variables that could explain the diversity in the results. Similar explanations apply to the remaining firm performance measures. Notably, the variable that offers the highest correlation coefficient with C.E.O. pay is Earnings per share (0.338). Among the set of governance characteristics, the Board Size coefficient is positive and significant in 22 out of 32 studies and displays the highest correlation value (0.183). On the contrary, the more active the board the lower the C.E.O. pay (the coefficient of Board Meetings is -0.103), although we need to be cautious in drawing any conclusion because the number of published studies reporting on this variable is low (nine cases) and the Safe N is also the lowest of the M. Finally, Leverage is positively related to our variable of interest, although it offers the lowest coefficient (0.009).
According to Cohen's (1988) scale, all the effect sizes (untabulated) are low but earnings per share and board size (that exhibit medium values).
Panel B of Table 3 reports the M.A. results on published sensitive analyses, that is, the regressions performed over incremental values, in order to determine the elasticity of C.E.O. pay in relation with the exploratory variables, following the model developed by Murphy (1985) and Coughlan and Schmidt (1985). All coefficients in Panel B display a significant association with C.E.O. pay, being the elasticity related to R.O.A. and Shareholder Return the one that offers the highest values. Yet again, all variables lack from homogeneity within published outputs due to underlying (not considered) variables. Therefore, the exploration of moderating variables is pertinent. Table 4 shows the conclusions driven by the significant results (not reported for brevity) of applying the moderating factors in order to answer the research questions RQ2-RQ4.

Results related to moderating variables
Regarding firm performance measures, in the U.K. context, only the published outcomes on the Share return-C.E.O. pay association and in the post-Cadbury period exhibit significant homogeneity. Notably, the coefficient is positive but lower than in the pre-Cadbury group of studies (as in . In the same vein, in the U.S. setting, the R.O.E.-C.E.O. pay association is higher in the pre-S.O.X. than in the following years although only the results in the pre-S.O.X. period are homogeneous. Thus, our results do not suggest a clear effectiveness of the governance measures in increasing the pay-performance association. None of the tested variables moderates firm's size measures but total industry (the group that excludes the publications focused on the banking industry). Hence, despite the specific banking regulation, we failed to find a consistent behaviour of any variable attached to C.E.O. pay in the financial industry.
Within the set of governance characteristics, the study reveals homogeneity only in the results related to: (1) C.E.O. Ownership-C.E.O. pay association with a higher coefficient than the one referred to the pre-S.O.X. period; and (2) the Executive (executive positions other than C.E.O.) pay and the board meetings.
Additionally, in undisclosed tables, we have performed the M.A. using additional moderating variables such as the geographical region (E.U. vs Non-E.U. countries or U.S. vs non-U.S. based studies) the legislative setting according to both Anglo-Saxon vs Non-Anglo-Saxon countries and Civil vs Code Law countries, and type of compensation (Total compensation vs Cash compensation), but we failed to find homogeneous groups.

Results of the meta-regression
We run the meta-regression on a model where the dependent variable consists of the effect size of the C.E.O. pay-performance association, being the independent variables all the determinants considered in the meta-analysis and displayed in Table 3. The set of exploratory variables adopts a dichotomy form, which equals 1 if they are included in the models tested in the primary studies and 0 otherwise. In addition, we also include some variables such as the Number of variables in the tested regressions and the Median year sample window, both in absolute values (as in Van Essen et al., 2015).
MARA results (Table 5) indicate that the association between performance and C.E.O. pay is moderated when the firm's size and the number of board independent members are included, showing both positive and significant coefficients. That is, the larger companies and the more independent boards strengthen the association between firm's performance and C.E.O. pay. Moreover, ownership concentration, although not significant, displays a positive association with C.E.O. pay-performance relationship (contrary to Bebchuk & Fried, 2004). Thus, further research on this field would help to a better understanding of the ownership structure and C.E.O. power over the board.

Discussion of the main effects results
We extend prior M.A. (Tosi et al., 2000) by analyzing results published in the last two decades that have witnessed a worldwide proliferation of Governance codes.
This study contributes to the agency theory-M.P.T. debate, from a global perspective, in the following terms: Our results reveal that in absolute values: (1) the payperformance association is significant although all effect sizes are low but earnings per share that exhibits a medium value (according to the scale by Cohen, 1988); and (2) firm size still displays the highest correlation with C.E.O. pay (around 44%). However, deeper insights from elasticity analyses suggest that C.E.O. pay is more sensitive to variations in performance variables (both market and accounting based measures) than to firm size changes. A plausible explanation for this finding is that ongoing governance measures are modifying rewarding schemes though they exhibit stickiness to size variables due to the organisational complexity and risk exposure of large firms (Diez Esteban, Garc ıa-G omez and L opez-Iturriaga, 2013).
Since the M.A. does not test the causal effect, deeper insights through M.A.R.A. results also support that pay-performance increases with board independence (as agency theorists predict) and firm size (i.e., preserving C.E.O.s interests according to M.P.T.). Hence, our results reconcile rather than alternate both theories. Moderator variables offer miscellaneous results and do not support robust conclusions. They barely suggest: (1) that governance measures (in particular the Cadbury Code and S.O.X.) had, if any, a detrimental effect in the pay-performance association; and (2) the C.E.O. pay in the banking industry does not offer a homogeneous pattern. In addition, dichotomies related to institutional or geographical regions, such as Anglo-Saxon vs Non-Anglo-Saxon; E.U. vs non-E.U.; Civil vs Common Law countries failed to further explain heterogeneity in prior findings.
These results help to understand the somehow contradictory empirical evidence and to provide solid foundations for future hypotheses developments. This contribution is also relevant to the regulatory bodies and standard setters because our results reveal that, in order to harmonise shareholders and managers' interests, the pay-performance association still needs further factual implementation of governance measures.

Limitations and further research
Among the limitations of this study, the use of meta-analytic structural equations could offer additional results and reduce the possibility of omitted variables bias. Endogeneity concerns could also be addressed, in particular, how remuneration policies influence performance (De Andres et al., 2018).
Future research should be devoted to test alternative theories. Also, additional investigation over unexplored areas, such as Latin America or Russia, might help to understand peculiarities and differences of governance systems across countries. Notes 1. As identified in http://www.oxfordreference.com/view/10.1093/oi/authority.20110803095413570 2. The r correlation coefficient was reported in 35 studies. In the remaining cases, t-statistics, b parameters, p-values or standard errors were used to estimate partial correlation (following Rosenthal & DiMatteo, 2001). 3. When necessary, in order to avoid problems generated by high standard deviation in p values, we computed the Pearson correlation coefficient normalised by Fisher's Transformation (Zr). 4. Studies with 'no significant results' are likely unpublished due to either the editors' preferences or the researches inhibition from sending papers when they failed to verify the formulated hypotheses.