LITERATURE REVIEW AND HYPOTHESES

The relationship between executive compensation and company performance has been widely analysed from theoretical and empirical perspectives. Agency theory (Jensen & Meckling, Reference Jensen and Meckling1976; Fama & Jensen, Reference Fama and Jensen1983) argues that the incentives of managers can be aligned with the objectives of the shareholders through the design of a contract that links the compensations of executives with business results (Evans & Weir, Reference Evans and Weir1995). However, Zingales (Reference Zingales1998) remarks that it is impossible to perfect a priori contract that fairly shares quasi-rents generated by a company to all its stakeholders, including executives and shareholders, and this quasi-rents will be captured by those groups with the most bargaining power.

This controversy is extended to empirical works (Sánchez-Marín, Baixauli-Soler, & Lucas-Pérez, Reference Sánchez-Marín, Baixauli-Soler and Lucas-Pérez2010). On the one hand, most of the studies have revealed a statistically significant positive relationship between changes in the wealth of shareholders and the compensations of executives (Jensen & Murphy, Reference Jensen and Murphy1990; Murphy, Reference Murphy1999; Conyon & Schwalbach, Reference Conyon and Schwalbach2000; Lee, Lee, & Wu, Reference Lee, Lee and Wu2010) and this relationship is consistent with agency theory. On the other hand, other studies offer evidence for the existence of a negative or negligible relationship between performance and the compensations of executives, concluding that managers put its own interests before the aims of the company owners (Core, Holthausen, & Larcker, Reference Core, Holthausen and Larcker1999; Duffhues & Kabir, Reference Duffhues and Kabir2008).

Devers et al. (Reference Devers, Cannella, Reilly and Yoder2007) argue that the mixed findings that pervade the field may result from a narrow focus that failed to incorporate other theoretical perspectives and methodologies. Recent theoretical developments have expanded agency theory explanations of the incentive alignment construct (Nyberg, Fulmer, Gerhart, & Carpenter, Reference Nyberg, Fulmer, Gerhart and Carpenter2010). In this vein, managerial power theory states that boards of directors approve compensation schemes that deviate from optimal contracting, since directors are subject to influence by management or are ineffective in overseeing managerial behaviour. As a result, top executives may camouflage or extract corporate rents by means of inefficient pay structures which are more in the interest of executives than shareholders. However, the evidence regarding the impact of CEO power on their compensations is insufficient and requires further studies (Tien, Chen, & Chuang, Reference Tien, Chen and Chuang2013).

Accordingly to the argument that top executives generally prefer a higher level of compensation that is less contingent upon performance, managerial power theory states that top executives have influence over the board of directors to approve compensation contracts that are more favourable to receive higher levels of pay and less sensitive to performance (Shin, Reference Shin2013). In a recent meta-analysis (Van Essen, Otten, & Carberry, Reference Van Essen, Otten and Carberry2012) on existing empirical studies focused on the managerial power theory, Van Essen, Otten, and Carberry. (Reference Van Essen, Otten and Carberry2012) conclude that top executives with greater power tend to receive higher levels of total pay and total compensation. However, the authors find that the results for performance-pay sensitivities provide less support for managerial power theory. Van Essen, Otten, and Carberry (Reference Van Essen, Otten and Carberry2012) explain that this unexpected finding could be due to the existence of a nonlinear relationship between CEO pay and performance, since the number of studies investigating nonlinear relationships is relatively limited.

Recently, few scholars (Leone, Shuang, & Zimmerman, Reference Leone, Shuang and Zimmerman2006; Canarella & Nourayi, Reference Canarella and Nourayi2008) have challenged the linear relationship between executive compensation and performance, and have examined a nonlinear and asymmetric influence of performance on compensations, depending on whether the company is profitable or not. Yet no theoretical or empirical consensus exists on how this asymmetry is reflected. Canarella and Nourayi (Reference Canarella and Nourayi2008) conclude that, from an accounting approach, a positive return by a company has a positive influence, although the executives are rewarded to greater extent for a positive return than they are penalized for a negative return. Leone, Shuang, and Zimmerman (Reference Leone, Shuang and Zimmerman2006) suggest that the salaries of managers are more sensitive to losses than to expected but unrealized profits. The authors state that these results are in line with the theoretical approach of the optimal contract. This approach is based on the idea that the implementation of an optimal contract reduces agency costs in relation to the expropriation of income from shareholders by the executives. Therefore, if the compensations of executives were more sensitive to expected profits than losses (the opposite case to that evidenced) then the executives may be rewarded for profits that do not finally appear. If the compensation were not reimbursed then an ex post settling up problem would be created.

While the limited previous studies have analysed the asymmetric influence of firm performance on executive compensations, depending on whether the company is profitable or not, none of these has considered a nonlinear relationship of the pay-performance sensitivity depending on the variation of corporate performance. This study sets its aim in this underdeveloped area of investigation in order to advance the debate about whether the structure of executive compensation aligned the interest of managers and shareholders before and during the most recent financial crisis.

The financial crisis has revealed weaknesses in the pay-setting process in the financial institutions since compensation schemes did not help mitigate agency problems and failed to closely align the interests of executives and shareholders (Faulkender et al., Reference Faulkender, Kadyrzhanova, Prabhala and Senbet2010). According to this argument, this study expands the explanation of the managerial power theory. This study states that whether the board of directors does not effectively oversee the design of incentive systems, the managers will have important influence over the compensation arrangements in order to serve their own interests rather than shareholders. In this case, it is expected that the pay-performance sensitivity is greater when the variation in the corporate performance is positive than when the variation is negative. Consequently, this study tests the existence of a nonlinear relationship that is asymmetrical in terms of pay-performance sensitivity to the sign of variation in bank performance. Accordingly, the following hypothesis is presented:

In this context, several empirical studies have estimated an econometric model where the explanatory variables are contemporary related with the explanatory variables. However, the assumption of not considering the dynamic behaviour of economic variables has been widely refuted. It is argued that the relationship between compensations and company profits is not only contemporary (Murphy, Reference Murphy1999; Duffhues & Kabir, Reference Duffhues and Kabir2008). Moreover in the short-term executive pay is taken into account the profit-sharing and bonuses if they are paid within 12 months of the end of the period. Therefore, the following hypothesis is proposed:

It is also argued that the executive compensation is persistent over time, which makes this variable dependent on its own past. Canarella and Nourayi (Reference Canarella and Nourayi2008) and Lilling (Reference Lilling2006) incorporated the first lag of the dependent variable in the equation and concluded that there was a positive and significant relation between the variable and the first lag. Consequently, the following hypothesis is expected:

In the Spanish banking system there are important differences in the corporate governance between commercial banks and savings banks. Commercial banks are privately owned institutions with a concentrated ownership structure and with a strong shareholder control (shareholder-oriented approach). Spanish savings banks are commercial non-profit organizations (Hansmann, Reference Hansmann1996), which have no owners or shareholders and are controlled by multiple interest groups: local and regional governments, employees, depositors and their founding entities (stakeholder-oriented approach). Several authors (Crespí, García-Cestona, & Salas, Reference Crespí, García-Cestona and Salas2004; García-Marco & Robles-Fernández, Reference García-Marco and Robles-Fernández2008; Sagarra, Mar-Molinero, & García-Cestona, Reference Sagarra, Mar-Molinero and García-Cestona2013), have argued that the dispersed ownership structure and the lack of shareholder control of the savings banks could give managers more freedom of action and leave much room for abuse and corrupt practices. Therefore, according to managerial power theory, executive managers of Spanish savings banks have more power over the board of directors to negotiate for higher pay than executive managers of commercial banks. In this case, the following hypothesis is suggested:

SAMPLE, VARIABLES OF THE MODEL, AND METHODOLOGY

Sample

The sample includes banks and savings banks issuing securities admitted to trading on regulated markets from Spain for the period 2004–2008. According to the Spanish Securities and Investments Board, 22 banks and 42 savings banks are required to publish an annual corporate governance report (CNMV, 2008). Credit cooperatives were not considered because of their low market share in the Spanish banking sector.

Data for the top executive pay has been extracted from the annual accounts and the annual corporate governance report for each bank. Additionally, the financial information from 2000 to 2007 for each bank has been obtained from Intertell database. The financial information for the year 2008 has been gathered from the annual accounts published by Spanish Securities and Investments Board. However, this information was collected until May 2009 and in this date 11 banks had not yet published their accounts, therefore, some information is missing.

The available information was used to construct an unbalanced panel of 61 banks (22 banks and 39 savings banks), which represents ~95% of the sample. After removing those financial institutions that did not have all the data and establishing that it is needed 5 years of continuous data, the final sample was a balanced panel of 195 observations for nine banks and 30 savings banks, which represents 61% of sample. Table 1 shows the descriptive statistics of the data included in each panel. In general, the data analysed does not show a normal distribution. Variables such as SEP and total assets are characterized by positively asymmetrical distribution and high kurtosis.

Table 1

Descriptive statistics

The table shows the descriptive statistics of the data included in unbalanced and balanced panel.

Methodology

As indicated by Gibbons and Murphy (Reference Gibbons and Murphy1990) agency theory suggests that the compensations of executives are related to performance in absolute or relative terms. This work has started from equation (1), which is a widely used model in the literature (Murphy, Reference Murphy1985; Jensen & Murphy, Reference Jensen and Murphy1990) that links executive compensation with the performance of a bank.

$$\ln {{(SEP)}_{i,t}}\, = \,{{{\rm{\beta }}}_{\rm{0}}}\, + \,{{{\rm{\beta }}}_{\rm{1}}}\,\times \,{{P}_{i,t}}\, + \,{{{\rm{\beta }}}_{\rm{2}}}\,\times \,{{P}_{i,t{\rm{ - 1}}}}\, + \,\mathop{\sum}\limits_{j\, = \,{\rm{1}}}^J {{{{\rm{\lambda }}}_j}} \,\times \,V.CONTRO{{L}_{i,t}}\, + \,{{u}_{i,t}}$$

Model (1) from a time dimension focus includes the first lag corresponding to an exogenous variable. This feature does not breach the basic hypothesis of the general linear model because all the explanatory variables remain deterministic. However, the cross-sectional and temporal nature of data means that consideration must be given to the existence of latent unobservable effects specific to each individual. If these effects are not considered explicitly in the model, the estimated coefficients of the explanatory variables will be biased, as they reflect some of the unobservable effects of each individual (Novales, Reference Novales2000).

$${ \ln (SE{{P}_{i,t}})\, = \, {{{\rm{\beta }}}_{\rm{0}}}\, + \,{{{\rm{\beta }}}_{\rm{1}}} \cdot {{P}_{i,t}}\, + \,{{{\rm{\beta }}}_{\rm{2}}} \cdot {{P}_{i,t{\rm{ - 1}}}}\, + \,{{{\rm{\beta }}}_{\rm{3}}} \cdot SUBINDUSTR{{Y}_{i,t}} \cr \quad + {{{\rm{\beta }}}_4} \cdot N{{E}_{i,t}}\, + \,{{{\rm{\beta }}}_{\rm{5}}} \cdot NE_{{_{{i,t}} }}^{{\rm{2}}} \, + \,{{{\rm{\beta }}}_{\rm{6}}} \cdot \ln {{(AT)}_{i,t}}\, + \,{{{\rm{\beta }}}_{\rm{7}}} \cdot CREDI{{T}_{i,t}} \cr \quad + {{{\rm{\beta }}}_{\rm{8}}} \cdot MA\_RIS{{K}_{i,t}}\, + \,\mathop{\sum}\limits_{J\, = \,{\rm{1}}}^4 {{{{\rm{\lambda }}}_J}} \cdot YEA{{R}_t}\, + \,{{{\rm{\eta }}}_i}\, + \,{{{\rm{\upsilon }}}_{it}} \cr}$$

Equation (2) specifies the general econometric model of this work, where the error term is divided (υ_i,t) into two components: the individual effects (η_i) which include the effect of non-observable variables in each bank, constant over time, and the stochastic disturbance. Traditionally, two transformations have been used of equation (2) to eliminate the problem of unobserved heterogeneity: deviations from the group averages (within estimator) and time differences (Gibbons & Murphy, Reference Gibbons and Murphy1990; Perry & Zenner, Reference Perry and Zenner2001).

Most studies assume that the explanatory variables of the model are exogenous as their values are taken as given, and so remain uninfluenced by the variable to be explained. However, in our model this hypothesis runs contrary to the empirical evidence, because from an accounting point of view, bank profit is affected by the salary costs of all employees (temporary and permanent). In this regard, recent studies suggest that analysing the behaviour of some corporate governance variables in isolation may not be appropriate (Andres & Vallelado, Reference Andres and Vallelado2008). Additionally, empirical evidence shows that the compensations of the executives are persistent over time. Based on model (2), a lagged-dependent variable is introduced and the result is the model (3). In this model, this study tests the hypothesis H₀₃ by analysing the sign and the significance of the coefficient of the delayed dependent variable.

$${ \ln (SE{{P}_{i,t}})\, = {{{\rm{\beta }}}_{\rm{0}}}\, + \,{{{\rm{\beta }}}_{\rm{9}}} \cdot \ln (SE{{P}_{i,t{\rm{ - 1}}}})\, + \,{{{\rm{\beta }}}_{\rm{1}}} \cdot {{P}_{i,t}}\, + \,{{{\rm{\beta }}}_{\rm{2}}} \cdot {{P}_{i,t{\rm{ - 1}}}}\, + \,{{{\rm{\beta }}}_{\rm{3}}} \cdot SUBINDUSTR{{Y}_{i,t}} \cr \quad+ \,{{{\rm{\beta }}}_{\rm{4}}} \cdot N{{E}_{i,t}}\, + \,{{{\rm{\beta }}}_{\rm{5}}} \cdot NE_{{_{{i,t}} }}^{{\rm{2}}} \, + \,{{{\rm{\beta }}}_{\rm{6}}} \cdot \ln {{(AT)}_{i,t}}\, + \,{{{\rm{\beta }}}_{\rm{7}}} \cdot CREDI{{T}_{i,t}} \cr \quad+ {{{\rm{\beta }}}_{\rm{8}}} \cdot MA\_RIS{{K}_{i,t}}\, + \,\mathop{\sum}\limits_{J\,{\rm{ = }}\,{\rm{1}}}^4 {{{{\rm{\lambda }}}_J}} \cdot YEA{{R}_t}\, + \,{{{\rm{\eta }}}_i}\, + \,{{{\rm{\upsilon }}}_{it}} \cr}$$

In model (3) it is assumed that the relationship between the compensations of executives in logarithms and the performance of the bank is linear. By following Leone, Shuang, and Zimmerman (Reference Leone, Shuang and Zimmerman2006), this study proposes model (4) in order to test the nonlinearity and the existence of asymmetry in the relationship between the variations in performance and SEP, depending on whether the company performances were better or worse than the previous year (hypothesis H₀₁).

$${ \ln (SE{{P}_{i,t}})\, = {{{\rm{\beta }}}_{\rm{0}}}\, + \,{{{\rm{\beta }}}_{\rm{9}}} \cdot \ln (SE{{P}_{i,t{\rm{ - 1}}}})\, + \,{{{\rm{\beta }}}_{\rm{1}}} \cdot {{P}_{i,t}}\, + \,{{{\rm{\beta }}}_2} \cdot {{P}_{i,t{\rm{ - 1}}}}\, + \,{{{\rm{\alpha }}}_1} \cdot {{P}_{i,t}} \cdot {{N}_{i,t}}\, + \,{{{\rm{\alpha }}}_{\rm{2}}}{{P}_{i,t{\rm{ - 1}}}} \cdot {{N}_{i,t{\rm{ - 1}}}} \cr \quad + \,{{{\rm{\beta }}}_{\rm{3}}} \cdot SUBINDUSTR{{Y}_{i,t}}\, + \,{{{\rm{\beta }}}_{\rm{4}}} \cdot N{{E}_{i,t}}\, + \,{{{\rm{\beta }}}_{\rm{5}}} \cdot NE_{{_{{i,t}} }}^{{\rm{2}}} \, + \,{{{\rm{\beta }}}_{\rm{6}}} \cdot \ln {{(AT)}_{i,t}} \cr \quad + \,{{{\rm{\beta }}}_{\rm{7}}} \cdot CREDI{{T}_{i,t}}\, + \,{{{\rm{\beta }}}_{\rm{8}}} \cdot MA\_RIS{{K}_{i,t}}\, + \,\mathop{\sum}\limits_{J\, = \,{\rm{1}}}^4 {{{{\rm{\lambda }}}_J}} \cdot YEA{{R}_t}\, + \,{{{\rm{\eta }}}_i}\, + \,{{{\rm{\upsilon }}}_{it}} \cr}$$

where N_i,t is the dummy variable that indicates negative variations, respectively, in the performance of the ith bank between two consecutive periods. That is:

$$ {{N}_{i,t}}\, = \:\left\{{\matrix{ \rm{0} \ \ {\it si} \ \ {{{\it P}_{\it i,t}}{{ - }}{{\it P}_{\it i,t{\rm{ - }}1}}\geq 0} \cr \!\!\!{\rm{1}} \ \ {si} \ \ \ {{{P}_{i,t}}{\rm{ - }}{{P}_{i,t{\rm{ - }}1}} \less 0} \right. $$

This dummy variable is multiplied by the corresponding performance. In this way, equation (4) introduces two variables that reflect the negative variations in performance between the periods t−1, t and t−2, t−1. Specifically, this study uses a characteristic (the sign of the variation in performance) to define the two regimes of behaviour, and in this way, it should improve the linear approximation.

In this context, the lag in the SEP is introduced as an explanatory variable, and this becomes related to the lag of the error term, so resulting in inconsistent estimates for both the within estimator and the first-difference estimator. An alternative method to avoid the discussed estimation problems for autoregressive models is to use an instrumental variable approach. Following the steps of Canarella and Nourayi (Reference Canarella and Nourayi2008) and Lilling (Reference Lilling2006), this study applied the generalized method of moments (GMM). Various forms of GMM estimators are feasible, depending on the definition of the conditions relative to the moments. Arellano and Bond (Reference Arellano and Bond1991) proposed the GMM difference estimator. It takes advantage of all the conditions of lineal moments arising from the assumption of non-correlation with the error term.

Blundell and Bond (Reference Blundell and Bond1998) demonstrated one of the limitations of this estimator by showing that if the autoregressive parameter is close to unity and there are few time periods, then simulations will produce estimates with a large bias in finite samples and coefficients that are imprecisely calculated. To overcome their weakness as instruments with variable lags at various levels, Arellano and Bover (Reference Arellano and Bover1995) and Blundell and Bond (Reference Blundell and Bond1998) proposed the system-GMM estimator. This estimator implies an additional restriction on the initial process conditions, so that all the available moments can be exploited through a system that uses lags in differences and levels of the dependent variable as instruments of the equation in levels and differences, respectively.

This article uses the system-GMM that has been implemented using a two-step estimator, so that estimates are efficient and asymptotically robust in the presence of heteroscedasticity. Additionally, standard second-stage error correction as proposed by Windmeijer (Reference Windmeijer2005) for small samples is applied.

CONCLUSION AND DISCUSSION

The recent controversy over the excessive executive pay associated with the corporate scandals and the financial crisis has called into a question the effectiveness of the executive incentive system to ensure that executives make decisions consistent with the shareholders’ interest. In this context, this paper contributes to this debate by providing empirical evidence regarding the relationship between executive pay and corporate performance in the Spanish banking sector. In particular, the aim is to examine the pay-performance sensitivity depending on the sign of variation in bank performance.

From a theoretical point of view, this study expands the explanation of the managerial power theory which states that top executives may influence the pay-setting process to approve compensation schemes that are more favourable to executives than shareholders. In accordance with this reasoning, this study suggests that executives may have influence to change the pay-performance sensitivity in their own interests depending on the sign of the variation of the performance.

From a methodology perspective, the article discusses the novel use of GMM that controls endogeneity problems and is consistent with the presence of unobserved heterogeneity in each bank. Given the small sample size, this study additionally implements standard Windmeijer error correction.

The main results reveal that SEP is positively associated with contemporaneous corporate performance, as well as present an asymmetric relationship in terms of sensitivities. The asymmetry in the pay-performance sensitivity is greater when the variation in results is positive than when the variation of the results is negative. This result is consistent with the managerial power theory and calls into question the role of the pay-performance incentives to align interest of executives and shareholders.

Regarding practical implications, the paper reveals that in spite of the numerous international recommendations in related to increase openness in executive compensation, there are numerous limitations in obtain public information about the compensations, mainly in long-terms executives pay and in individual terms, and without this information the shareholder and the rest of stakeholders can not control over excessive executive remuneration. Moreover, the results suggest that the board of directors and banking regulation should improve the supervision on the executive pay-setting process with the aim of ensuring that pay-performance incentive fulfils its purpose to align the interest of executive and shareholders.

This study has limitations that might open new areas for future research. The results of this study might be limited by the sample, the availability of information and the variables used in the empirical models. Future studies could move beyond Spanish savings banks and commercial banks to focus on cross-national studies to identify best practices and explore how context shapes managerial practices. In addition, the differences in incentive systems between Spanish savings banks and commercial banks could be further explored. To test the robustness of this study, future works should use more finely grained measurements of executive pay and could examine the nonlinear relationship between performance and the four pay components: salary, bonus, equity participation and post-employment compensations. Moreover, this study could be extended to pay-volatility sensitivities which would be another fruitful avenue for new and interesting theoretical and empirical developments. Finally, this study encourages further research on the question whether the compensation structures in financial firms contributed to the instability, excessive risk-taking, and the most recent global financial crisis.