Agency theory approach of the relationship between performance, compensation and value creation in the companies listed on Euronext Lisbon

The aim of the present study was to understand, in Portugal, relationships between the Chief Executive Officer (CEO) and the organizational performance, on the one hand, and relationships between the shareholder value creation and the CEO total compensation, on the other hand. This research is divided into two parts. The first part was examined whether organizational performance based on accounting measures influences organizational performance based on market measures and whether organizational performance based on accounting measures and market measures influences the CEO total compensation. The second part of the study analyses whether organizational performance, based on accounting measures and market measures, and the CEO total compensation influence the shareholder value creation. This research was based on agency theory assumptions in order to build the analysis model. The sample was composed of companies listed on Euronext Lisbon. The data analysis was performed using the structural equation modelling method. The results showed that organizational performance based on accounting measures influences organizational performance based on market measures, the CEO total compensation and the shareholder value creation.


Introduction
stated that the paradigm of corporate governance has drastically changed CEO compensation. In the past, investors did not have a say on how the CEO was paid. If the shareholders were unhappy, they had few options other than selling their own stocks. According to these authors, the recommendations of the executive board regarding the CEO compensation were approved by the respective remuneration committees with little independence, using the rival companies to assess the compensation competitiveness or to establish performance goals, if they even existed.
Currently, literature has shown that the assessment of the CEO compensation has drastically changed (e.g., Lilling, 2006;Ozkan, 2007;Shaw & Zhang 2010). Investors have started to have an active role and a say on compensation. According to Doubleday and Wagner (2009), the executive board faces an increasing scrutiny by shareholders, press, legislators and regulators and at the same time tries to balance the interests of the shareholders and CEO. The CEO has started to have a secondary role in a process that used to be led by them. A strong performance evaluation system assesses whether the interests of the shareholder are being safeguarded or not, thus providing support for decision making. These developments are changing CEO compensation, as companies optimize the connection between remuneration and performance through short and long term incentives.
The relationship between the CEO compensation and the shareholder value creation has caught the attention of investors over time. In accordance with Gong (2011), academics, legislators and the press have discussed the high salaries of CEO, questioning if these are consistent with the shareholders' interests. As claimed by this author, the existing studies do not show consistent results about the outcome of the performance of the CEO compensation, and these studies have raised concerns about how the compensation is able to align the interests of both the CEO and the shareholders. Several examples of the lack of alignment of these interests showed up during the so-called "subprime" crisis, highlighting the case of the Wall Street CEO compensation, which led to a strong debate in the US.
Given the gaps in the current research, the goal of this research is divided into two parts. The first part examines whether organizational performance based on accounting measures influences organizational performance based on market measures and whether organizational performance based on accounting measures and market measures influences the CEO total compensation. The second part of the study analyses whether organizational performance, based on accounting measures and market measures, and the CEO total compensation influences the shareholder value creation.
This research intends to provide, theoretically, a greater understanding about the CEO compensation level in the listed companies in Portugal and how it is affected by the different types of organizational performance measures (accounting and market) and also aims to contribute to the assessment on how the CEO compensation and organizational performance behaviour affect the shareholder value creation. In practical terms, it intends to clarify for the shareholders and other stakeholders the influence of organizational performance on the CEO compensation and the impact of latter on the shareholder value creation.
To address the objective of the study and related research hypotheses, this research is divided into three sections. In the first one, the theoretical concepts related to the agency theory are analysed and the theoretical foundations related to the variables, on which this study is based, are presented to support the key issues that need to be answered. In the second section, the objectives of the study will be described, plus the analysis model, the descriptive analysis, the data collection process and the selected statistical techniques for the treatment of the data. In the third section the results of the applied tests to the model are shown and, finally, the theoretical discussion of these results.

Agency Theory
The split that emerged after the industrial revolution between owners and managers, with the first holding the property but not a significant control over it, and the last controlling but not having a significant ownership, brought about a new relationship in companies (Berle & Means 1932), the so-called "agency relation" (Ross, 1973). Ross (1973) defined this relationship as one of the most ancient and common types of social interaction. As claimed by this author, an agency relation arises between two or more parts, when one of them, the agent, acts on behalf of, or as representative of the other part, called the principal, in the particular field of decisionmaking.
As reported by Eisenhardt (1989) and Bloom and Milkovich (1998), the agency relation presumes that both parts (principal and agent) are subject to three behavioural assumptions: rationality, effort aversion and risk aversion. Rationality exists because it enables agents and principals to assess the probability of future contingencies in order to protect their own interests (Baiman, 1990). The effort aversion exists because the agents want to dedicate the least possible work in the execution of their tasks without reducing their compensation (Bloom & Milkovich, 1998). The risk aversion exists because job security and the compensation of the agents rely on a unique entity (Balkin et al., 2000). Jensen and Meckling (1976) claimed that, if principal and agent maximize the utility for their own benefit, there is a good chance that the agent will not always act in the best interests of the principal, thus creating the so-called "agency problem". The agency problem arises when cooperative parties have different goals and perspectives about the work (Eisenhardt, 1989), which means that the cooperative behaviour that maximizes the general interests doesn't match the individual interest of each protagonist (Baiman, 1990). According to Chua et al. (2003), the agency problem is related to the divergence of interests, information asymmetry and limited rationality.
The principal may incur a loss, the so-called "agency costs", whenever the agent pursues different goals from the principal (Jensen & Meckling, 1976). As Tosi et al. (2000) mentioned, this is the outcome of the agency problem previously discussed, i.e., the agents don't like to take risks and have personal interests that may diverge from the principal's interests. These assumptions suggest that the agents may undertake actions based on their own agenda, which seeks to accomplish their individual goals (Tosi et al., 2000). Jensen and Meckling (1976) noted that agency costs emerge in any relationship that involves cooperative endeavour between two or more parties. These authors defined the agency costs as the sum of the expenses spent on a monitoring system, bonding costs and residual loss. The monitoring system is intended to limit the abnormal activities of the agent. The bonding costs ensure that the agent won't undertake certain actions that could harm the principal or, at least, guarantee that the principal will be reimbursed if those actions take place. The residual loss is the reduction of the benefits by the principal, as a result of the conflict of interests between the agent and the principal, even after the application of a monitoring system´s bonding costs.
The agency theory's unit of analysis is the contract that establishes the work relationship between the principal and the agent (Eisenhardt, 1989), containing the rights and responsibilities of each part, (Baiman, 1990). This contract includes remuneration agreements, information systems, functions assigned to the agent and rights of the principal (Baiman, 1990). It's also through the contract that the principal tries to link his or her own interests to the agent's interests (Tosi et al., 1997). Eisenhardt (1989) claimed that the theory focuses on assessing the most efficient contract, given the existing assumptions about the individuals, the organizations and the information. The developed theory about the efficiency of the agency contract points to two solutions that can be chosen by the principal to solve the agent-principal problem (Tosi et al., 2000). First, when the principal has full access to complete information about the agent's efforts, the most efficient contract is based on the observation of the agent's behaviour (monitoring). This choice prevents a contract being based on results which transfer, unnecessarily, the risks to the agent, who is reluctant to take risks. In the second choice, when the principal is unable to monitor the agent's efforts, and information asymmetry is high, the principal has to transfer the risks to the agent, through an incentive contract. Tosi et al. (1997) declared that the solution to the agency problem doesn't depend on a direct selection of monitoring systems and incentives. According to these authors, since the performance can be measured both by the actions and the results of these actions, the first choice of the principal should focus on creating a balance between base compensation (based on behaviour) and the additional incentives (based on results). This balance is known as "optimal contract", which maximizes the return of investment to the principal through modifications to the monitoring structure and compensation of the agent so that the desired changes in organizational success are achieved (Bloom & Milkovich, 1998).
Several management accounting tools can be used as monitoring, evaluation and incentives systems to mitigate the agency problem. Among the several internal monitoring strategies of the companies that mitigate the agency problem, Wickramasinghe and Alawattage (2007) highlight the performance evaluation systems. Luft (2009) added that performance evaluation may be used to give rewards like bonuses, capital instruments and promotions. The performance evaluation applies to both individuals and organizations (Burke & Litwin, 1992) and is, generally, operated by objective performance measures (Indjejikian & Nanda, 2002).

Variables
The variables used in this study are: CEO total compensation: The CEO compensation is measured according to Miller et al. (2002) definition of "CEO total compensation", which includes all the forms of compensation, like the base compensation, bonus, stock options, restricted stocks and other remuneration types.
Organizational Performance Based on Accounting Measures: In agreement with Richard et al. (2009), accounting measures are the most common and simplest way to assess organizational performance and many authors have employed this kind of measures to determine it (e.g., Murphy, 2001). The accounting measures used in the present study were selected based on the measures identified by Richard et al. (2009). The chosen measures were the ones that achieved significant results with CEO compensation in previous research (e.g., Perry & Zenner, 2001). The two selected measures were: net income and Earnings Before Interest, Taxes, Depreciation, and Amortization (EBITDA). Richard et al. (2009) mentioned that, despite the credibility of accounting measures having been proved, these measures are subjected to distortions originating from the adopted accounting politics, human error and fraud. These authors highlighted that, due to its dependence on auditable information, accounting measures reflect what happened and are very limited in predicting about future performance. In order to mitigate the gaps of these measures in evaluating organizational performance, the variable organizational performance based on market measures has been added.
Organizational Performance Based on Market Measures: According to Richard et al. (2009), the advantages of market measures are to predict the future and incorporate the value of intangible assets (e.g., brand image) in a more efficient way than by just using accounting information. Many authors (e.g., Core et al., 2003) used these measures to assess organizational performance. The market measures used in the present study were selected based on the measures identified by Richard et al. (2009) and combined, considering the study of Murphy et al. (1996). The chosen measures were the ones that achieved significant results with CEO compensation in previous studies (e.g., Johnson et al., 2008). Consequently, the selected measures were: basic Earnings Per Share (EPS) and stock price. Basic EPS correspond to International Accounting Standard 33 definition and stock price correspond to quotation price of the common stocks on a regulated market at year end. These measures make part of the "profit" dimension identified by Murphy et al. (1996). Like accounting measures, market measures also have some disadvantages. To Merchant (2007), the market does not always reflect the effective performance. That means that some of these evaluations are very influenced by future expectations that will never become reality. The author added that paying bonuses based on market measures is risky, as these payments can be made based on results that will never come true.
Shareholder Value Creation: Shareholder value creation is measured using the Market Value Added (MVA) method. MVA is calculated by subtracting the amount invested by the shareholders to the market value of capital (Brigham & Ehrhardt, 2008). According to Hillman and Keim (2001), the market value of capital is equal to the market capitalization of a company and the amount invested by the shareholders is the subscript capital.

Formulation of Research Hypotheses
It is expected that the variables organizational performance based on accounting measures and organizational performance based on market measures influence positively and significantly the CEO total compensation. This presumption is based on the assumption that companies use optimal contracts (e.g., Bloom & Milkovich, 1998) to align the interests of the CEO with organizational performance and with the interests of shareholders (e.g., Tosi et al., 1997), thus mitigating the agency problem (e.g., Jensen & Meckling, 1976). Since the interests of shareholders depend directly on organizational performance (e.g., a high net income represents, theoretically, a higher dividend or investment), it is expected that organizational performance has a major impact on the CEO total compensation. This expectation is supported by the Portuguese corporate governance report model, which refers in §70, point III, Section D, Part I, that the remuneration should be structured "(...) in order to align the interests of members of the administration with the long-term interests of society (...)" and based on "performance evaluation" (Regulation No. 4/2013). Murphy (2001), Perry and Zenner (2001) and Kateratorn (2013) used accounting measures to prove the existence of a relationship between organizational performance and CEO compensation. Murphy (1998), Core et al. (2003) and Johnson et al. (2008) demonstrated the same by using market measures.
Based on the literature referred to, the following research hypotheses were formulated: In addition to the influence of organizational performance on the CEO total compensation, there is the influence of organizational performance on shareholder value creation. Since it is predictable that organizational performance based on accounting and market measures influences the CEO total compensation and the CEO total compensation influences shareholder value creation, then it is correct to predict that organizational performance based on accounting and market measures influences the shareholder value creation. Thus, it is expected that the organizational performance based on accounting measures and the organizational performance based on market measures influence, in a positive and significant way, the shareholder value creation. This correlation is based on the assumption of Merchant (2007) that the market assimilates all available information on a given company and reflects it in its assessment of the same, which varies the shareholder value creation. The relationship between the accounting and market measures and the shareholder value creation has been proven previously, for example, by Milunovich and Tsuei (1996), Uyemura et al. (1996) and Vijayakumar (2008). Finally, as referred to above, it is expected that the CEO total compensation influences the shareholder value creation in a positive and significant way. This influence is based on the assumption of optimal contract use of CEO compensation schemes to mitigate the agency problem and align their interests with the interests of shareholders (e.g., Tosi et al., 1997). This relationship has been proved earlier by Sheikholeslami (2001), Fatemi et al. (2003) and Baum et al. (2004).
Based on the contributions in the literature, the following research hypotheses have been added: H 4 : Organizational performance based on accounting measures influences the shareholder value creation. H 5 : Organizational performance based on market measures influences the shareholder value creation. H 6 : CEO total compensation influences the shareholder value creation.

Objective and analysis model
The objective of the present research is divided into two parts. The first part intends to verify whether organizational performance based on accounting measures influences organizational performance based on market measures (H 1 ) and whether organizational performance based on accounting measures and market measures influences the CEO total compensation (H 2 and H 3 ). The second part of the study analyses whether organizational performance, based on accounting measures and market measures, and the CEO total compensation influence the shareholder value creation (H 4 , H 5 and H 6 ). This study is also based on the assumptions of the agency theory on the principal/agent (Shareholder/CEO) relation, predicted by Eisenhardt (1989), the agency problem and agency costs (with monitoring, bonding costs and residual loss), predicted by Jensen and Meckling (1976), and the agency contract (that establishes the work relationship between the principal and the agent), predicted by Tosi et al. (1997), to build the analysis model. The hypothetical analysis model can be divided into two parts, as already mentioned concerning the objective of the research. The first part is formulated based on the contributions from studies that successfully linked the selected measures of organizational performance and the CEO compensation. These measures are the net income (e.g., Murphy, 2001), EBITDA (e.g., Kateratorn, 2013), basic EPS (e.g., Johnson et al., 2008) and the stock price (e.g., Core et al., 2013). Based on the classification from the study of Richard et al. (2009) and the dimensions of the study of Murphy et al. (1996), the different organizational performance measures were grouped in order to assess two variables not directly observable: organizational performance based on accounting measures and organizational performance based on market measures. Through the definition of compensation by Miller et al. (2002) the variable CEO total compensation was assessed. These contributions support the hypotheses H 1 , H 2 and H 3 of this study.
The second part of the model is built based on the contributions of the studies that positively and significantly link the selected measures of organizational performance and the CEO compensation to the shareholder value creation, which is measured by the MVA (e.g., Baum et al., 2004;Vijayakumar, 2008). These contributions support the hypothesis H 4 , H 5 and H 6 of this study.

Data collection and sample
Data were collected from the Annual Reports and Accounts, including the Consolidated Financial Statements (in accordance with the International Financial Reporting Standard 10), Reports on Corporate Governance (when this is an annex to the Annual Report and Accounts, published separately), both of them with mandatory disclosure to the public in accordance with The reporting period covers five years, in accordance with the deadline for making public the annual reports and accounts (No. 1 , Article 245 of the CVM). This selected period begins in 2010 and finishes in 2014, the last year having the annual report and accounts published at the time of collecting the data used in this study. Excluded from the sample were the entities that were not listed on the Euronext Lisbon during the periods of 2013 and 2014 and the Sports Companies, given their specificity about the goals to be achieved (e.g., sporting success) or the motivation of investors in stocks acquisition (e.g., affectivity).
The sample is composed of several types of entities that integrate the main sectors of activities (see Table 1).
The analysis of Table 1 shows that the sector with the highest percentage is Industrials (30%), followed by Consumer Services (21%) and Financial Services (12%). The Oil & Gas sector includes only one company in the sample, but the relative weight is 2.3%. It is also noteworthy the annual average growth of 6.4% in the Oil & Gas sector and 4% in the Utilities sector. On the contrary, the Telecommunications sector showed a very significant negative annual variation (-60%), followed by the Financial Services sector (18%).

Statistical Methods and Treatment of Data
The Structural Equation Modelling (SEM) is the method selected for processing the data in this study. According to Schreiber et al. (2006), the SEM are statistical techniques that can be used to reduce the number of variables observed in a smaller number of latent variables through the analysis of covariance between the observed variables. In order to process the data, the model's position and its variables were analysed given the assumptions of the SEM, the quality of adjustment was tested and, finally, the estimates of the model parameters were calculated, i.e., the weight of the regressions of the direct and indirect relationships between the variables and the correspondent level of significance. In the analysis for decisionmaking about whether or not to validate the research hypotheses being studied, a 5% significance level was assumed. The model design and statistical tests were made using the AMOS (v. 19, SPSS Inc, Chicago, IL) and SPSS (v. 19) software.

Diagnostic of the Assumptions of Structural Equation Modelling
Like most statistical techniques, SEM is based on assumptions that must be respected so that the results are reliable (Finney & DiStefano, 2006). Schreiber et al. (2006) recommend that the assumptions about the absence of outliers are analysed, plus the sample size, the multivariate normality of the data and also the absence of multicollinearity among manifest variables. These assumptions are analysed separately in the paragraphs below.
According to Marôco (2014), outliers can be registration errors or actual extreme values, suggesting the Mahalanobis Distance measure (D 2 ) to diagnose multivariate outliers. The Mahalanobis Distance consists of the calculation of two probabilities (p 1 e p 2 ). An observation that registers a reduced value in both probabilities can be classified as multivariate outlier. These observations were removed to make the subsequent analysis. Considering the reference value suggested in the literature (e.g., Schreiber et al., 2006), 5%, 27 multivariate outliers were detected in this study. These observations were removed to make the subsequent analysis.
Regarding the assumption about sample size, Schreiber et al. (2006) reported that there is no specific rule about the exact number of observations that a sample should contain in the SEM. However, there is a general consensus of 10 observations per parameter to be estimated in a model. Thus, the ratio of observations should not be less than 10/1. The study model consists of 18 parameters (10 regressions, 1 covariance and 7 variances) and the number of observations is 184. Considering the criterion mentioned by the authors, the sample size fulfills the assumption to apply the SEM technique, since the ratio of participants per parameter is 10.22 (184/18) to 1.
In the analysis of data distribution, according to Marôco (2014), the most used distribution measures are asymmetry and kurtosis. These measures are operationalized through the calculation of univariate asymmetry, univariate kurtosis and multivariate kurtosis. Finney and DiStefano (2006) reported that when univariate asymmetry and univariate kurtosis approximate values of 2 and 7, respectively, they indicate the existence of problems related to normality. For these authors, a value of multivariate kurtosis greater than 10 indicates severe violation of normality. The results show that the manifested variables present values higher than the reference values suggested by the authors, indicating a severe violation of the normal distribution. Since the data do not meet the assumption of the normal distribution, the function Maximum Likelihood cannot be used to estimate the model.
In cases of violation of the normal distribution, Marôco (2014) suggests, among other methods, the use of Bootstrap estimation. This method involves repeating random samples, replacing the original sample with a new Bootstrap sample and thus calculating standard errors for the hypotheses tested (Schumacker & Lomax, 2004). This form of Bootstrap, known as non-parametric Bootstrap, does not depend on population distribution (Enders, 2002).
Bootstrap estimation requires the absence of missing values (Jose, 2013) and, to meet this requirement, its imputation was made. From the analysis, it was found that the missing values of the sample are concentrated in variable EBITDA. This variable was not calculated for the entities classified by Euronext Lisbon as Financial Services, since the specificities of this activity make EBITDA an irrelevant ratio (e.g., financing expenditures are part of the operating activities). For this reason, the number of missing values is equal to the number of entities classified as Financial Services that are part of the sample in each of the five periods being analysed, which totals 22 missing values, corresponding to 12% of the total sample for EBITDA variable. These values were imputed by the Expectation Maximization algorithm. According to Olinsky et. al. (2003), Expectation Maximization is the most used method to deal with missing values. This method performs an iteration, through a process, to initially estimate the missing information and then the parameters. The iteration process ends when the estimated parameters converge on some pre-established criteria.
Finally, the assumption of multicollinearity refers to the high correlations between manifest variables (Grewal, Cote & Baumgartner, 2004). For Hair et al. (2011), the information of an indicator may prove to be redundant due to high levels of multicollinearity in the model, which may result in indicators without statistical significance. O'Brien (2007) pointed out that the degree of multicollinearity of the indicators should be analysed to determine the redundancy by calculating the Variance Inflation Factor (VIF) or Tolerance. A VIF value of 5 or greater or a Tolerance value of less than 0.200 indicates potential multicollinearity problems (O'Brien, 2007;Hair et al., 2011). Given the criteria recommended by Hair et al. (2011), this assumption is not violated since all the variables obtained a VIF value less than 5 and a Tolerance value greater to 0.200.

Adjustment Quality Diagnosis
According to Schermelleh-Engel et al. (2003) and Schreiber et al. (2006), the selected tests for quality adjustment assessment of the SEM in this study are the Chi-square test (χ 2 ), inferential test, the ratio χ 2 /degrees of freedom (df), the Root Mean Square Error of Approximation (RMSEA) and the Standardized Root Mean Square Residual (SRMR), descriptive indexes of general adjustment, and the Non-Normed Fit Index (NNFI) and the Comparative Fit Index (CFI), which are descriptive indexes based on the comparison of models. Additionally, it's calculated the corrected χ 2 Bollen-Stine test for data that do not follow a normal distribution (Bollen & Stine, 1992). The following table summarizes the results of selected tests. According to the results shown in Table 2, quality adjustment tests point out that the model has a good fit (Schermelleh-Engel et al., 2003;Schreiber et al., 2006;Marôco, 2014). The statistical test results of χ 2 is 4.911 with a p-value associated 0.297, i.e., the matrix of population covariance does not differ significantly from the covariance matrix estimated by the model (null hypothesis), and the result of the ratio χ 2 /df is 1.228. As the χ 2 test is sensitive to the violation of the normal distribution assumption, and the manifest variables in this study violate this assumption, it was calculated the corrected χ 2 Bollen-Stine test (Bollen & Stine, 1992) that obtained a p-value of 0.306 in two thousand Bootstrap samples, confirming the adequacy of the model. The descriptive indexes of general adjustment RMSEA and SRMR obtained a result of, respectively, 0.035 (P[rmsea≤0.05]=0.510; confidence level 90%) and 0.026. The descriptive indexes of model comparison NNFI and CFI obtained the results of, respectively, 0.989 and 0.997.

Standardized Estimates and Significance of Trajectories
This point begins with the presentation of the final structural equations model of this study. The model consists of two latent variables, six manifest variables and seven unique factors or errors (e). It also has a correlation identified between the errors. Measurement errors of the net income and basic EPS variables are correlated by the reason that the last uses the first in the calculation formula. The research hypotheses (H) are also shortly identified and the directional paths that represent them.  Figure 1, the values that follow the directional paths point to the standardized coefficients (β) of the parameters of the relationship they represent. The values placed in the upper right side of the latent and manifest variables indicate the squared multiple correlation coefficients (R 2 ). The significance level of the parameters estimated by the model was obtained using the estimation process for Bootstrap, which simulated two thousand samples.

Referring to
The results point to a correlation between measurement errors of the manifest variables net income and basic EPS of 0.34 (R = 0.340, p <0.001), as expected. Factorial weights of organizational performance based on accounting measures in EBITDA and net income are, respectively, 0.72 and 0.52 (λ = 0.72 and λ = 0.52), and the proportion of explained variance is 51% and 27% (R 2 = 0.51 and R 2 = 0.27). The factor weights of organizational performance based on market measures in basic EPS and stock price are, respectively, 0.52 and 0.92 (λ = 0.52 and λ = 0.92), and the proportion of the explained variance is 27% and 84% (R 2 = 0.2 and R 2 = 0.84). All factors have high factor weights (λ ≥ 0.5), indicating factorial validity, and appropriate individual reliabilities (R 2 ≥ 0.25) (Marôco, 2014). Standardized path coefficients and multiple correlation coefficients, including its statistical significance, will be analysed separately. Table 3 contains the standardized coefficients and the respective significance levels, which statistically evaluate the formulated hypotheses in this research. According to Table 3 data, the effect of organizational performance based on accounting measures on organizational performance based on market measures (β = 0.650, p < 0.001) and the effect of organizational performance based on accounting measures in the CEO total compensation (β = 0.910, p < 0.001) are positive and statistically significant. The effect of organizational performance based on market measures in CEO total compensation is not positive or statistically significant (β = -0.230, p = 0.130). The effect of organizational performance based on accounting measures in shareholder value creation is positive and statistically significant (β = 0.810, p = 0.034). The effect of organizational performance based on market measures on the shareholder value creation is positive but not statistically significant (β = 0.240, p = 0.483). Finally, the effect of CEO total compensation in shareholder value creation is negative and not statistically significant (β = -0.370, p = 0.107).
The squared multiple correlation coefficients and the respective levels of significance that assess the variability of a given variable with respect to the predictors used for this variable are shown in Table 4. Analysing the data presented in Table 4, it was concluded that the model explains 42% of the variability of organizational performance based on market measures (R 2 = 0.423, p < 0.001), 61% of the CEO total compensation variability (R 2 = 0.611, p < 0.001) e 58% of the shareholder value creation variability (R 2 = 0.576, p = 0.003).

Discussion
The research hypotheses H 1 , H 2 and H 4 were confirmed. Hypothesis 1 confirms that the market absorbs all available information on a given company and reflects it in its assessment of the same (e.g., Merchant, 2007). Therefore, the better or worse the organizational performance based on accounting measures, the better or worse is the organizational performance based on market measures.
Hypothesis 2 corroborates the assumption that optimal contract of the agency theory establishes a link between organizational performance (in this case, based on accounting measures) and the CEO total compensation (e.g., Tosi et al., 1997;Bloom & Milkovich, 1998). This result means that the companies listed on Euronext Lisbon recognize the existence of the agency problem (e.g., Jensen & Meckling, 1976;Baiman, 1990), originated by the CEO/ Shareholder relationship (e.g., Ross, 1973), and use the agency contract (e.g., Tosi et al., 1997) to mitigate the costs associated with this problem (e.g., Jensen & Meckling, 1976;Tosi et al., 2000). To achieve that, these entities use accounting measures (e.g., Luft, 2009). In this situation, the shareholder selects the performance evaluation system that specifies the measures that will be based on the compensation of the CEO and the function linking these measures to compensation. Thus, the better or worse the organizational performance based on accounting measures, the higher or lower the CEO total compensation. This hypothesis also confirms that the analysed entities comply with the recommendation of the Comissão do Mercado de Valores Mobiliários (CMVM) on the dependence of the CEO's compensation on the CEO's performance (Regulamento No. 4/2013). The result of this research hypothesis is consistent with the conclusions of the studies of Murphy (2001), Perry and Zenner (2001) and Kateratorn (2013).
Research hypothesis 4 indicates that a positive or negative variation on the shareholder value creation is explained in part by the organizational performance based on accounting measures. As the effect of organizational performance based on accounting measures on organizational performance based on market measures, the effect predicted by this hypothesis is based on the assumption that the market assimilates all available information on a given company and reflects it in its assessment (e.g., Merchant, 2007), which causes variations on the shareholder value creation. This result adds to the results of studies such as Milunovich and Tsuei (1996), Uyemura et al. (1996) and Vijayakumar (2008). Overall, the organizational performance based on accounting measures assumes an essential position in the analysed entities. Both the CEO total compensation and the shareholder value creation depend on the behaviour of this variable. This dependence, when excessive or when not shared with other measures, such as market measures, can damage the fixation of the CEO total compensation and the perception of the shareholder on the intrinsic value of the entities. This injury results from distortions caused by accounting policies adopted, human error and fraud (e.g., Richard et al., 2009).
The research hypotheses H 3 , H 5 and H 6 were not confirmed because they did not present statistically significant results. That is, it was not possible to prove the effect of organizational performance based on market measures on the CEO's total compensation. Likewise, the effect of organizational performance based on market measures on the shareholder value creation was not corroborated. The effect of the CEO's total compensation on shareholder value creation has also not been proven.
In relation to variability, the results point out that more than half of the variability of the CEO total compensation and the shareholder value creation is explained by its predictors in the model. The predictors of the CEO total compensation are organizational performance based on accounting measures and market measures and predictors of shareholder value creation are the organizational performance, based on accounting measures and market measures, and CEO total compensation. The explained variability of the CEO total compensation, slightly above 60%, reflects the balance between monitoring and incentives (e.g., Bloom & Milkovich, 1998), i.e. the agency contract (e.g., Tosi et al., 1997) uses sufficient incentives to align the agent's interests with the interests of the principal, without transferring too much risk and remuneration variability to the agent, reflecting a balance between monitoring and incentives (e.g., Baiman, 1990). With respect to the variability explained by the predictors of organizational performance based on market measures, basic EPS, stock price and organizational performance based on accounting measures, corresponds only to 42%, i.e. less than half. This suggests replacing or adding organizational performance predictors based on market measures.

Conclusion
The objective of this study was divided into two parts. The first part examined whether organizational performance based on accounting measures influences organizational performance based on market measures (H 1 ) and organizational performance, based on accounting measures and market measures, influences CEO total compensation (H 2 and H 3 ). The second examined whether organizational performance, based on accounting measures and market measures, and CEO total compensation influence the shareholder value creation (H 4 , H 5 and H 6 ).
This study used agency theory assumptions to build the analysis model. The sample was composed by the companies admitted to trading on the regulated market Eurolist by Euronext Lisbon. Data were collected from the Annual Reports and Accounts, including the Consolidated Financial Statements, Reports on Corporate Governance and the Euronext Lisbon webpage.
Data were processed using the SEM method.
The hypotheses H 1 , H 2 and H 4 were confirmed. Hypothesis 1 confirmed that the evaluation of a company by the market is conditioned by the accounting measures. Hypothesis 2 confirmed that the organizational performance based on accounting measures determines the CEO total compensation. Hypothesis 4 has confirmed that the organizational performance based on accounting measures conditions the shareholder value creation.
This research contributes, theoretically, to strengthen the explanation of the factors that affect organizational performance based on accounting measures behaviour, the CEO total compensation behaviour and the shareholder value creation behaviour (e.g., Lilling, 2006;Merchant, 2007;Shaw & Zhang, 2010). This research contributes, in practical terms, to help in clarifying matters for the stakeholders of the listed companies in Portugal, in particular shareholders, who are its owners. The results of this study show that the listed companies in Portugal comply with the CMVM recommendation on the dependence of the compensation of the CEO's performance, at least with regard to organizational performance based on selected accounting measures.
The main limitation of this study is related to the form of disclosure of the CEO compensation. The form of disclosure for the CEO compensation imposed by CMVM does not favour the collection of data. The fact that a standard model is not available for companies to publish information related to the compensation of members of the Company's Board of Directors makes it difficult to collect and impairs the collected information itself.
For future research, it is suggested to undertake methodologically similar studies with a modification in the number of manifest variables. The manifest variables explained by the organizational performance based on market measures, basic EPS and stock price, could be increased or replaced taking into account the negative results obtained in this study. The number of manifest variables explained by the organizational performance based on accounting measures and market measures could be increased and eventually lead to more robust results of the model as advocates Marôco (2014). Additionally, as future studies intends to extend the study to the other European countries.