Cressey hypothesis ( 1953 ) and an investigation into the occurrence of corporate fraud : an empirical analysis conducted in Brazilian banking institutions

is article lls a technical-scienti c gap that currently exists in the Brazilian literature on corporative fraud, by combining the theoretical framework of agency theory, of criminology, and of the economics of crime. In addition, it focuses on a sector that is usually excluded from analyses due to its speci c characteristics and shows the application of multinomial logit panel data regression with random e ects, which is rarely used in studies in the area of accounting. e aim of this study is to investigate the occurrence of corporative fraud, as well as cases of fraud in Brazilian banking institutions, by using detection variables related to the Cressey fraud triangle. Research into fraud and methods of detecting fraud has grown in management literature, especially a er the occurrence of various corporative scandals in the 1990s. Although regulatory agencies have increased their investments in monitoring and control, fraud investigations and convictions are still common in the day-to-day administration of banks, as can be seen in the Brazilian Central Bank and the National Financial System Resource Council’s databases of punitive proceedings. We believe that this article will have a positive impact in the area of accounting sciences, since it involves corporative fraud in a multidisciplinary form and because it provides the incentive to use a quantitative tool that can help increase the development of similar studies in the area. is study tested the theory that the dimensions of the fraud triangle condition the occurrence of corporative fraud in Brazilian banking institutions. irtytwo representative variables of corporative fraud were identi ed in the theoretical-empirical review, which were reduced to seven latent variables by the principal component analysis. Finally, the seven factors formed the independent variables in the multinomial logit models used in the hypothesis tests, which presented promising results.


INTRODUCTION
e number of studies on fraud and fraud detection methods has grown in management literature, especially since the occurrence of corporate scandals in the 1990s (Wang, Winton, & Yu, 2010).ese studies include those that investigate the probability of occurrence of accounting and corporate fraud, such as those by Beasley (1996), Brazel, Jones, and Zimbelman (2009), Crutcheley, Jensen and Marshall (2007), Erickson, Halon and Maydew (2006), Lou and Wang (2009), Troy, Smith, and Domino (2011), and Wang et al. (2010).
In the studies mentioned, one point of interest can be veri ed: banking institutions have not been analyzed in isolation.Because of the importance of these institutions for the economic context, it is clear that the losses from a fraud in a large bank will be felt by the economy in general, since these institutions act as nancial intermediaries and providers of external capital for other economic activities.erefore, understanding and nding ways of preventing and detecting corporate fraud in banking institutions is crucial for society as a whole.
In light of this, the hypothesis created by Cressey (1953) stands out, enabling fraudulent behavior by managers in corporations to be examined using an analysis with three dimensions: pressure, opportunity, and rationalization.
is hypothesis reports that individuals who occupy positions of trust in the nancial area ( nancial trust) can violate this trust if they have some particular nancial problem that cannot be shared.ese individuals, in their capacity as agents of the corporation, believe that this particular problem can be resolved secretly, even if for this they have to violate the trust placed in them.By justifying the breaking of trust as a means of resolving their nancial problems, these individuals may use nancial resources to their own bene t, even if they self-sustain a false feeling of seriousness in their actions (Cressey, 1953).Cressey's hypothesis (1953), which is also known as the fraud triangle, considers three dimensions of fraudulent behavior: pressure, opportunity, and rationalization.Pressure, also known as motivation, requires the existence of nancial problems that cannot be shared; opportunity would be to secretly resolve these problems by violating nancial trust; and rationalization of the fraudulent act would be viewing it as necessary and justi able for resolving the nancial problems.Cressey's fraud triangle (1953) has been used both by the literature and in accounts auditing practice to investigate the occurrence of accounting and corporate fraud.In the literature, the studies by Brazel et al. (2009), Lou and Wang (2009), and Troy et al. (2011) stand out, which manage to elaborate econometric models based on Cressey's triangle (1953) to detect fraud.In practical terms, Cressey' s hypothesis (1953) has been used by standardizing organizations as a tool for detecting fraud (Higson & Kassem, 2013) It bears mentioning that Brazilian studies on corporate fraud are still in their infancy (Murcia & Borba, 2005;Silva, 2007).Most of the Brazilian studies have di erent scopes from that of this paper, concentrating on analyzing red ags, whether via auditor perception or via the creation of new structures (Murcia & Borba, 2007;Murcia, Borba, & Schiehll, 2008), using data on international companies to calculate the probability of fraud (Wuerges & Borba, 2014) or mapping patterns of corporate fraud (Imoniana & Murcia, 2016).erefore, carrying out studies to calculate the probability of corporate fraud occurring in the Brazilian context is of academic relevance.
In light of this theoretical gap and of the importance of the Cressey fraud triangle in the theoretical and empirical contexts, the general aim of this study is established, which is to investigate the occurrence of corporate fraud, as well as indications of fraud, in Brazilian banking institutions, by using detection variables from agency theory, criminology, and the economics of crime.

Agency Theory
The agency relationship is one of the oldest and most common forms of codified social interactions (Ross, 1973) and is present in complex societies.Examples of agency relationships include: bossworker, doctor-patient, adviser-administrator, and relationships between parents and children.Functional dependency, among other reasons, determines that agency relationships are extremely common (Mitnick, 1975).These relationships include those of the owners and managers of corporations, derived from the separation between ownership and control, which is the object of study of agency theory.
According to Demsetz (1983) and Veblen (2001Veblen ( /1921)), the separation between ownership and control sought greater management e ciency.e delegation of decisionmaking authority, according to Barnea, Haugen, and Senbet (1985), is an essential characteristic of modern corporations, in which shareholders delegate their authority to a professional who has managerial skills.However, as Berle and Means (1932) and Demsetz (1983) observe, one of the main problems with regards to the relationship derived from the separation between ownership and control is that of ensuring that managers in fact work towards the aim of achieving the owners' objectives, since their interests are not always convergent.
Agency theory seeks to analyze the relationship between agents and principals.For this, a contract metaphor is used, in which the relationship between agents and principals is formalized in contractual terms (Jensen & Meckling, 1976) that can contain clauses to outline the agent's behavior, with the intention that he/she acts to ful ll the principals' expectations.e di erences between the parties' aims give rise to the con icts known as agency problems.Clearly, agency problems emerge when the con icts of interest between agents and principals, or between the principals themselves, a ect the operation of the company's businesses (Barnea et al., 1985).
ese agency problems, whatever the relationship, can damage the e cient functioning of a company.In order to minimize these problems, the owners incur agency costs, which can be subdivided into manager monitoring costs, spending on contractual guarantees from the agent, and residual losses (Jensen & Meckling, 1976;John & Senbet, 1998).
Monitoring costs consist of limiting divergences from the principal, via the creation of appropriate incentives for the agent, which will limit any abnormal activities.ese costs can be subdivided into mechanisms for incentivizing and monitoring agents' actions.
Incentive costs involve the agents' remuneration structure and system of financial incentives, while monitoring mechanisms are related to the corporate governance system, including the use of internal and external audits and formal systems of controls (Jensen & Meckling, 1976).
Commitment expenses are resources that agents will receive in the form of a guarantee that they will not make decisions that damage the principal, as well as for guaranteeing that the principal will be compensated if the agent makes such decisions (Jensen & Meckling, 1976).
e residual cost/loss is the monetary value resulting from the reduction in the principal's well-being which occurs in situations in which of the agent's decisions present divergences with regards to the decisions that would maximize the principal's well-being. is loss occurs when the cost of the total execution of a contract exceeds its bene ts (Fama & Jensen, 1983).
However, even with the occurrence of the agency costs described above, and with the creation of contracts with restriction clauses and incentives for managers' actions, monitoring of managers' behavior is imperfect, due to the fact that managerial actions are not observable (Denis, Denis, & Sarin, 1999).Like Denis et al. (1999), Jensen and Meckling (1976) observed that the management literature indicates the existence of imperfect monitoring in the agency relationship, resulting from the failure of monitoring costs to resolve agency problems.us, even with monitoring mechanisms, agents may not work in favor of the principals' interests.

Economics of Crime
e rst indications of the application of economic concepts in the area of criminology were observed in the studies by Cesare Beccaria (1819/1764) and Jeremy Bentham (2000Bentham ( /1781)), which were forgotten until the 1960s.Gary Becker (1968), a Nobel Prize laureate in 1992, reignited the discussion on the theory of crime via the economic prism and emphasized that his e ort in determining an economic structure for criminal behavior can be seen as a resurrection and modernization of the pioneering studies.Becker (1968)  behavior can dispense with special anomie theories, psychological inadequacies, or the inheritance of special characteristics, and he extended the usual choice analysis of economists.For the author, criminals are like any other person and behave as rational utility maximizers.us, a person becomes "criminal" when the function between cost and bene t of illicit activities is greater than other activities involving legal alternatives (Becker, 1968).
Based on the work of Becker (1968), economists have invaded the eld of criminology, using the comprehensive individual rational behavior model.is model, assuming that individual preferences are constant, can be used to predict how alterations in the probability of the severity of sanctions and socioeconomic factors can a ect the amount of crime (Eide, Rubin, & Sheperd, 2006).
In the literature on the economics of crime, studies are observed that analyze the relationship between previous performances and corporate crimes (Alexander & Cohen, 1996), the e ect of gender on corporate crimes (Ste ensmeier, Schwartz, & Roche, 2013), and a veri cation of the impact of psychological variables on white collar crime, using individuals who have not committed crimes as a control sample in their experiment (Blickle, Schlegel, Fassbender, & Klein, 2006).ese studies, and others related to corporate fraud and agency theory, will enable variables to be identi ed to measure the probability of corporate fraud.

Corporate Fraud
Fraud, in its wider sense, can cover any gains obtained through crime, which uses error as its main modus operandi (Wells, 2011).However, although all fraud involves some type of error, not all errors are necessarily frauds.Coenen (2008) observed that the legal de nition of fraud is generally presented as false intentional representation regarding a material point and which causes a loss to a victim.
Corporate fraud is related to the corporate environment and can be conceptualized as fraud committed by or against a corporation (Singleton & Singleton, 2010).Costa and Wood Jr. (2012, p. 465) conceptualize corporate fraud as a series of illicit actions and conducts carried out in a conscious and premeditated manner by members of an organization's senior management, which take place in a process, aiming to serve own interests and with the intention of harming third-parties.
e United States Justice Department de nes corporate fraud in three main areas: accounting fraud or nancial fraud, insider trading, and obstructive conduct (American Institute of Certi ed Public Accountants, 2006, cited by Rezaee & Riley, 2010).
Fraud in nancial statements can be conceptualized as the deliberate misrepresentation of a company's nancial conditions, carried out by distorting or intentionally omitting values or disclosure in the accounting statements in order to mislead the users of this information (Association of Certi ed Fraud Examiners, 2008, cited by Singleton & Singleton, 2010).With regards to corporate insider trading, this is mainly related to the improper appropriation of corporate assets by senior executives.As for obstruction of justice conduct, this refers to criminal convictions for giving a false testimony at the Securities and Exchange Commission (SEC) and in uencing or threatening other witnesses.
In this study, the type of fraud addressed is corporate, whose areas are related to nancial fraud and insider trading.Coenen (2008) argues that fraud constitutes the intentional neglect of a system and a deliberate attempt to violate this system to make personal gains, and that most company systems are not created to detect and prevent fraud.us, in an attempt to contribute by showing ways to detect corporate fraud, the Cressey fraud triangle (1953) will be used, which is detailed in the next item.

Cressey's fraud triangle.
One of the most brilliant students of Sutherland, Donald R. Cressey studied at the University of Indiana during the 1940s (Wells, 2011).At this teaching institution he took a doctorate in criminology and became interested in the behavior of fraudsters. is interest led him to write his doctoral thesis, for which he used interviews carried out with 200 prisoners convicted of fraud.With the results of the research, Cressey (1953, p. 30) formulated a nal hypothesis, known today as the fraud triangle.is hypothesis assumes that: Trusted persons become trust violators when they conceive of themselves as having a nancial problem that is non-shareable and are aware this problem can be secretly resolved by violation of the position of nancial trust, and are able to apply to their own conduct in that situation verbalizations which enable them to adjust their conceptions of themselves as users of the entrusted funds and properties.e elements of this triangle are shown in Figure 1.The Cressey hypothesis (1953) and an investigation into the occurrence of corporate fraud: an empirical analysis conducted in Brazilian banking institutions Pressure, also known as incentive or motivation, refers to something that happens in the personal life of the fraudster and that creates a stressful need, thus motivating the fraudster (Coenen, 2008;Singleton & Singleton, 2010).Cressey (1953) showed that in all of the cases found in the interviews, non-shareable problems precede the criminal violation of nancial trust.For the author, the violator considers various di erent situations to produce problems that are structured as non-shareable.ese problems are related with the status required by the o ender or with maintaining his/her status.
e analysis carried out by Cressey (1953) is consistent with the literature on fraud by indicating that the conditions related to immorality, emergencies, increased needs, reversals in the business environment, and a high standard of living are important for violations of trust.However, relevance only veri es whether these conditions produce non-shareable problems for people that occupy a position of trust. is situation will only have the e ect of creating, in the person of trust, the desire for speci c results -the pressure -, related with the solution to the problem, and which can be produced by the criminal violation of nancial trust (Cressey, 1953).
Opportunity presupposes that fraudsters have the knowledge and chance to commit fraud.e logic is that the individual will commit fraud as soon as he/she holds a position of trust, knows the weaknesses in the internal controls, and obtains su cient knowledge regarding how to successfully commit the crime (Singleton & Singleton, 2010).
For Cressey (1953), technical knowledge is acquired before the existence of the non-shareable problems and, consequently, the individual's ability to perceive that the non-shareable problem can be resolved by violating the position of trust involves the application of general information to speci c situations.us, when pressure, which is the existence of non-shareable problems, is added to such opportunities derived from the individual's knowledge, the potential for fraud is greater (Singleton & Singleton, 2010).
Rationalization is a cognitive process of selfjusti cation (Markin, 1979;Rahn, Krosnick, & Breuning, 1994;Scheufele, 2000). is concept is widely discussed by sociologists, psychologists, and psychiatrists.In his hypothesis, Cressey (1953) perceived that fraudsters rationalize their trust-violating conduct as acceptable and justi able behavior by the intention to resolve a given problem classi ed as non-shareable.So, rationalization is the process in which an employee mentally determines that fraudulent behavior is the correct attitude, considering that the company can absorb the consequences of this act or that no shareholder or stakeholder will be materially a ected by the execution of the fraud (Coenen, 2008;Singleton & Singleton, 2010).According to Cressey (1953), the rationalization used by violators is necessary and essential for the criminal violation of nancial trust, as it is by way of this that the individuals nd pertinent and real reasons to act; that is, they convince themselves that carrying out the violation of nancial trust is a justi able and acceptable act.
us, according to Cressey (1953), the occurrence of fraud is conditioned by the joint existence of three dimensions: pressure, opportunity, and rationalization.In accordance with that assumption, this study used Cressey's hypothesis (1953) to elaborate and test hypothesis 1.

Rationalization Opportunity Pressure
Michele Rílany Rodrigues Machado & Ivan Ricardo Gartner H1: the three dimensions of the fraud triangle together condition the occurrence of corporate fraud in Brazilian banking institutions.

EMPIRICAL ANALYSIS METHODOLOGY
e study is classi ed as empirical with a quantitative approach.To carry out the work, the rst step was to identify the banking institutions for which the Brazilian Central Bank (CB) made available data involving quarterly nancial information (QFI).Two hundred thirty-one institutions were found with data from January 2001 to December 2012.It bears mentioning that from December 2012 the CB ceased to publish quarterly information on the institutions under its supervision, following Circular letter n. 3,630 of 2013 (Brazilian Central Bank [CB], 2013).
As there was the need to also collect data from the Annual Information Forms (AIFs) and the Reference Forms made available to the market, the decision was made to work with only publicly-held banking institutions with AIFs covering no fewer than three whole years.Due to this limitation, 44 banking institutions composed the research sample.As the data were gathered and organized quarterly, a panel of 2,112 lines of observations was obtained.
In order to identify the existence or not of corporate frauds, the decisions reported by the CB and by the National Financial System Resource Council (NFSRC) were adopted.e former issues rst instance decisions while the latter is tasked with judging, in the second and third instance, the punitive administrative proceedings applied by the CB.In the period from 2001 to 2012, 123 punitive proceedings were found, spread over 27 banking institutions and corresponding to 61.36% of the 44 in the sample.e organizations that did not form part of the o enders report formed the control sample for the study, therefore enabling di erent patterns to be identi ed between two groups of institutions: with and without corporate frauds.
It is worth mentioning that the choice of punitive administrative proceedings was based on internationally published empirical studies on corporate/accounting frauds, such as those by Beasley (1996), Brazel et al. (2009), Erickson et al. (2006), Lennox and Pittman (2010), Troy et al. (2011), andWang et al. (2010).In these articles, the authors used the Accounting and Auditing Enforcement Releases, published by the SEC, which contemplate press releases on administrative and/or civil proceedings.
Note that for the classi cation of the proceedings the concept of corporate fraud was observed, which is fraud committed by or against a corporation (Singleton & Singleton, 2010). is concept enables the analysis of each sentence imposed on the banking institutions analyzed.So, cases were analyzed in which the claim was related with foreign exchange operations, rural credit and debts, investment funds, share repurchases, and the accounting of these institutions.

Econometric Modeling
e economic modeling developed is based on the empirical analysis procedures proposed in the studies by Beasley (1996), Brazel et al. (2009), Crutchley et al. (2007), Erickson et al. (2006), Lennox and Pittman (2010), and Wang et al. (2010), which were previously cited in this study and are geared towards measuring the probability of occurrence of corporate fraud.e calculation of this probability derives from qualitative choice econometric models, such as logit and probit, in which the dependent variable is binary, with 1 referring to the presence of the attribute and 0 referring to the absence of the attribute, and the attribute being de ned by the occurrence of the event, in this case corporate fraud.e logit and probit qualitative choice models can also be extended to multinomial approaches, in which the dependent variable can take one of several attributes in order to cover a greater number of possible occurrences for the studied phenomenon.
is possibility of greater detailing of the dependent variable enables a better adjustment of the data collected in this research to the empirical analysis methodology, since the occurrence of corporate fraud associated with the administrative proceedings of banking institutions is characterized by the existence of intermediary behaviors, given that an institution that is being investigated can be convicted or cleared in the administrative proceeding.In summary, the dependent variable can assume the following behaviors:  Absence of the attribute: banking institution without administrative proceeding;  Presence of the attribute: banking institution investigated in an administrative proceeding;  Presence of the attribute: banking institution cleared in an administrative proceeding;  Presence of the attribute: banking institution convicted in an administrative proceeding.According to Hilbe (2009), the multinomial probability distribution is an extension of the binomial distribution.
e multinomial model to test the functional relationship between corporate fraud and the independent variables can be speci ed as: in which where the set of variables x will be represented by the elements of the fraud triangle and by the control variables, which will be subsequently described.
Equation 1 presents the relationship between the dependent variable, y, and the independent variables, x. e attribute in reference, which will be compared to the others, is represented by b, while the number of categories is presented as m.To solve the equations j, the following equation is used to calculate the predicted probabilities: e collected data present the behaviors of the banking institutions studied over the years, forming transversal and longitudinal cross-sections.ese cross-sections can be analyzed by forming a simple pooling, when the estimated parameters are constant for all for all the observation units and for all the periods, or by forming a panel of data, when the estimated parameters are variables for each observation unit over time, which in the multinomial case are estimated by random e ects.A study conducted by Karlson (2011) compared the two approaches and found that the simple poolings tend to underestimate the parameter estimates compared to the random e ects models.It bears mentioning that in both models the estimation method used is maximum likelihood.
In light of this, in order to test the established hypotheses we opted for the estimation of the two types of multinomial logit models: the simple pooled one and the random e ects one, so as to then compare them.In order to provide an overview of how this study was operationalized in terms of econometric modeling, Figure 3 was elaborated.

Polychotomous dependent variable
Multinomial logit model Cramer an Ridder (1991) occurrence test to select between the pooled or panel model H0:


As in the owchart shown in Figure 3, depending on the speci city of the dependent variable and in light of the layout of the research data, the pooling of 44 banking institutions in accordance with their longitudinal data series was considered, in order to verify whether the model should following simple pooling or panel data under random e ects.e rst step in applying the methodology was to The Cressey hypothesis (1953) and an investigation into the occurrence of corporate fraud: an empirical analysis conducted in Brazilian banking institutions calculate the Cramer and Ridder (1991) test, with the aim of verifying whether this model can be used as simple pooling. is test assumes a high multinomial logit with (S +1) states and two pooling candidate states/levels, s 1 and s 2 .e null hypothesis assumes that s 1 and s 2 have the same regressor coe cients, except the e ect; that is: us, for the models in which the Cramer and Ridder (1991) test indicated that the multinomial logit model could be analyzed, the irrelevant alternatives hypothesis was tested.In this hypothesis (IIA), the appropriateness of using the multinomial model is ensured.us, the Hausman-MacFadden (1984) test and the Small-Hsiao (1985) test were applied, and alternatively, the suestbase Hausman test to verify the independence of the irrelevant alternatives, and consequently, the assumption of independence between the error terms.
For the models in which the Cramer and Ridder (1991) test indicated that simple pooling was not the best alternative, a multinomial logit model with random e ects was used.For this model, the Stata v.13 program was used, which made it possible to measure the multinomial logit models with random e ects using the gsem command.
is command enables models with multilevel data to be adjusted.e adjustment of these models makes it possible to simultaneously treat the level e ect in groups, for example by including random e ects such as nonobservable e ects in the group of companies.
An additional comparison between the multinomial logit, traditional, and with random e ects models was carried out by analyzing the information criteria -Akaike information criterion (AIC), Schwartz Bayesian information criterion (BIC), and likelihood ratio test (LR) -, with the aim of guaranteeing the choice of that which best adjusts to the data.
It is important to highlight that the multinomial logit model with xed e ects was not considered in the analysis because up to the nal date of carrying out this study its integration into the program used -Stata v.13 -had not been concluded, although in 2011 the researcher Klaus Pforr submitted the programming for this insertion.Other statistical software was not studied, with the exception of SPSS and Gretl.So, it cannot be a rmed that the multinomial model with xed e ects has not been implemented in di erent programs.

Independent variables.
e set of independent variables, X i,t , was mostly de ned with the help of the literature on agency theory and the economics of crime and empirical studies on fraud and earnings management in banking institutions.e description of these variables, of the data collection and research sources, as well as the control variables used, are presented in tables 1, 2, and 3.
Table 1 shows the independent variables used to measure the pressure dimension of the Cressey (1953) fraud triangle: The auditing rules, Resolution n. 1,207 of 2008 (CFC, 2009), and SAS n. 99 (AICPA, 2002), elaborated in accordance with the Cressey fraud triangle (1953), establish two types of sources of incentive or pressure that produce fraudulent nancial information when management is under pressure -external and internal sources -to achieve earnings targets or expected nancial results.
As internal and external sources of pressure, two items stand out that can serve as indicators of the pressure dimension: the performance to be achieved by the agents and perceived remuneration, which in uences the standard of living of these agents.As internal sources of pressure, the studies by Alexander and Cohen (1996) and Macey (1991) are used, which originate from the area of criminology and the economics of crime.According to Macey (1991), self-interested managers become involved in criminal conduct in the name of their organizations not to bene t the shareholders but to maintain their positions.Also according to the author, the threat of a lower than ideal performance can lead to managers preferring a higher level of risk in order to increase the company's performance.is can lead the manager to manipulate the nancial statements with the aim of increasing the institution's performance, thus achieving the performance targets set by the owners and resulting in them maintaining their current position.us, in order to measure the institutions' performance, the nancial intermediation revenue was used, as well as the variation in the institution's net income, the return on assets and on investments, and the level of market share in relation to the leading banking institution.For these variables, a positive and signi cant behavior is expected in relation to the probability of fraud; the higher its value, the greater the possibility of fraud occurring.As external sources of pressure, variables related to the structure of the managers' remuneration were used, since this is interconnected with their personal standard of living.
The managers' remuneration structure, or compensation policies, according to agency theory, aims to give incentives for the agent to select and plan actions that increase the wealth of shareholders (Jensen & Murphy, 1990).us, using articles on this theory, the following variables were outlined: xed remuneration, pro t share, and an indicator of remuneration perceived below the market average.Fixed remuneration and pro t share, as well as remuneration policies, could indicate an inverse relationship with the probability of fraud.Source: Elaborated by the authors.
The Cressey hypothesis (1953) and an investigation into the occurrence of corporate fraud: an empirical analysis conducted in Brazilian banking institutions As remuneration perceived below the market average, a positive relationship with the probability of fraud is expected, given that managers can use fraudulent means to "correct" perceived economic inequality.It bears mentioning that as a pressure point Cressey (1953) highlighted the relationship between employees and employers, in which the former can feel undervalued in relation to their status in the organization.is feeling can derive from perceived economic inequalities, such as payment and a feeling of being overloaded with tasks or being undervalued (Wells, 2011).Table 2 presents the independent variables used to measure the opportunity dimension.e opportunity dimension includes weak corporate governance structure as well as other working conditions that enable the manager to commit fraud (Brazel et al., 2009).
Corporate governance mechanisms enable the owners of a corporation to exercise control over the activities of insiders and managers, so that their objectives are protected (John & Senbet, 1998).Thus, if these mechanisms are not adequate and present fragilities, the possibilities of corporate fraud occurring is increased.To measure such mechanisms, the number of independent board members was used, as well as the size of the scal council, the size of the board of directors, the auditing rms considered as Big Five, company seals of corporate governance issued by the São Paulo Stock, Commodities, and Futures Exchange (BM&FBOVESPA), and the size of the institution's executive board.
It bears mentioning that Arthur Andersen, Deloitte, Ernst & Young, KPMG, and Price were considered as the Big Five.Currently the auditing companies are called the Big Four, as a result of the demise of Arthur Andersen in 2002.As the period of this study covers that year, the decision was made to follow the same treatment given by Lennox and Pittman (2010), by including this company and using the original denomination -Big Five.
Note that the independent members, board of directors size, and scal council size variables were squared to capture the fact that coordination, communication, and decision-making problems make the board's performance di cult when the number of directors increases (Yermack, 1996).us, the size of the board of directors and the number of independent members help in the e ectiveness of the corporate governance up to a certain point, since as the board grows the incremental cost of adding members will be greater than the monitoring bene t, thus constituting a convex function.
Also, as independent variables, the allocation for doubtful accounts and company size were included.Lou and Wang (2009) found that complex transactions are accompanied by a high inherent risk due to the involvement of a high degree of management judgment and subjectivity.
e allowance for doubtful accounts will identify earnings management opportunities for the manager, given the use of subjective criteria in its accounting. is variable is used to detect earnings management in banking institutions (Deboskey & Jiang, 2012), precisely because of the discretion and subjectivity that managers have to estimate it and the di culty involved in the auditing of this variable (Deboskey & Jiang, 2012).
When the size of the organization increases, managers have a greater amount of resources at their disposal, as well as there being an increase in the complexity of operations and in the agency con icts between owners and managers (Ryan & Wiggins, 2001).us, when the size of the company increases together with the complexity of its operations and the con icts derived from the agency relationship, managers can use this environment to execute corporate fraud.
Table 3 shows the independent variables used to measure the rationalization dimension of the Cressey fraud triangle (1953) and also those used as control variables.In an attempt to measure the rationalization dimension, the demographic characteristics of the executives were used.ese characteristics, although considered incomplete and imprecise proxies of the executives' cognitive structures, are taken as valid because of the di culty in obtaining conventional psychometric data on senior executives (Hambrick, 2007).For this, the following demographic variables were used: age of the chief executive o cer (CEO), his/her training in any area, his/her training on courses related to the area of business, the predominance of females on the scal council, on the board of directors, and in management, and a punishment factor.
For Troy et al. (2011), younger managers are more likely to rationalize accounting fraud as an acceptable decision.us, Zahara et al. (2007) emphasize that younger managers tend to take risks as a way of more quickly achieving career progression.As for older ones, they tend to be more analytical in their decision making, executing decisions with more care, seeking more information, and carrying out a more precise diagnosis of the information gathered.erefore, age appears to be an indicator of an individual/manager's moral development.
With regards to educational training, studies such as those by Gioia (2002), Hambrick and Mason (1984), and Rest and oma (1985) recognize the role of education in the ethical behavior of managers, as well as empirically proving this relationship has a positive association with moral development (Rest & oma, 1985).Also, Troy et al. (2011) a rm that managers without knowledge in the area of business will tend to rationalize fraud as an acceptable decision, unlike those trained in the area of business, who will be more aware of potential repercussions and penalties of unethical behavior.In light of the claims of the various authors, it is perceived that managers' educational training can be directly associated with the act of committing fraud.
To analyze managers' gender, support was sought from the studies by Kelley et al. (1990), Reynolds (2006), andZahra et al. (2007), which indicate that male managers are more likely to accept unethical behavior to achieve their objectives.Moreover, Ste ensmeier et al. ( 2013) determined that female executives can be more ethical in their decision making.erefore, it can be inferred that a female predominance can negatively in uence in the probability of corporate fraud occurring.e use of a punitive factor was derived from the studies by Cressey (1953) and Becker (1968).In his model, the latter pondered that criminal individuals consider the e ect of punishment in their decision to commit a crime or not.us, the e ect of an increase in punishment is expected to negatively impact the probability of corporate fraud.
In order to measure the e ect of punishment on the probability of corporate fraud occurring, the decision was made to use the normative alterations that increased the punishments applied to crimes against the National Financial System (NFS).e alterations in the legislation were quali ed by a dummy.It bears mentioning that the legal instruments that discipline crimes against the NFS were researched and of these only Law n. 9,613 of 1998 (Brazil, 1998) against crimes of money and illicit laundering in the financial system was altered.The alteration occurred via Law n. 12,683 of July 2012 (Brazil, 2012), which raised the penalties applied to institutions involved in crimes and their representatives.
For the model, four control variables were also de ned: type of bank, convergence with the international accounting rules, state control, and subprime crisis.
e rst variable was used to identify the type of bank classi ed by the CB according to their activities -multiple or commercial bank.
e second variable was employed with the aim of distinguishing the period in which the conversion to the international accounting rules came into e ect, in accordance with Resolution n. 3,786 of 2009(CB, 2009).It bears mentioning that, according to CB information, the accounting norms established by the National Monetary Council and by the CB, embodied in the Accounting Plan for Institutions of the National Financial System (COSIF), present divergences in relation to the international accounting rules issued by the IASB, representing partial convergence with the international accounting rules.
In order to moderate the e ect of the control and ownership of the banking institutions, the inclusion of variables that identify whether the control is state or foreign was considered.ese two types of categories have di erent characteristics from the other banking institutions.Foreign banks need to deal with di erent environments and regulations: regulations in their country of origin and those of the foreign institution.State banks can operate with government subsidies, besides having a more complex governance due to the presence of one more agent: the politician (Silva, 2004).However, in this study foreign control was not di erentiated, given that for the banking sector, 87.38% of total assets in 2012 are of banks with Brazilian capital, besides the limited number of banking institutions with foreign control, corresponding to six of the 44 institutions analyzed.
e fourth control variable was included to analyze the e ect of the subprime crisis.us, the variable will indicate the crisis period occurring from mid-2007 to April of 2009 (Maciel et al., 2012).
It is worth noting that in light of the research period, January 2001 to December 2012, the variables derived from the nancial statements and the xed remuneration of management were monetarily corrected.For this, the monetary correction index was used, as well as the general index of prices-internal availability (IGP-DI), calculated monthly by the Getúlio Vargas Foundation. is procedure enabled an analysis of the impact of these variables in a space of time without the in uence of in ation.
It is perceived that the number of independent variables presented in tables 1, 2, and 3 is too high, totaling 32, and this number of variables may have an impact on the accuracy of the model to be estimated for hypothesis 1, due to the problems derived from the multicollinearity of the independent variables.One option for minimizing this problem is the application of data reduction techniques, such as factor analysis.
Factor analysis is a multivariate technique that provides tools for examining the structure of the interrelationships in a large number of variables, de ning sets of strongly inter-related variables, known as factors (Hair Jr., Black, Robin, Anderson, & Tatham, 2009).To apply this technique, rst the following tests were carried out: (i) Bartlett sphericity, to analyze whether the variables are intercorrelated, and (ii) Kaiser-Meyer-Olkin (KMO), to measure the adequacy of the sample.For Hair Jr. et al. (2009), the closer to 1 the KMO value is, the better the sample will t the factor analysis.e author states that the researcher should consider a general value above 0.50 to apply the factor analysis.
e technique applied enabled the generation of components, called factors, which were rotated by the orthogonal varimax method to simplify their analysis.
ese factors were allocated instead of the 32 variables reported in tables 1, 2, and 3 and used to test hypothesis 1 of this study, via the application of the multinomial logit model.
It is important to mention that the logit model to be estimated can lose its predictive character due to the application of the factor analysis technique.e inclusion of new observations in the database will imply the need to employ the technique again, whose action can alter the parameters estimated in this study.However, it is worth highlighting that the results of the study enable important variables to be identi ed to calculate the probability of fraud occurring.ese variables can be used both in new academic studies and by market professionals, such as regulators, auditors, and investors.

EMPIRICAL ANALYSIS
e use of factor analysis to reduce the data resulted in the data in Table 4.As the results of Table 4 show, the Bartlett sphericity test indicates that the null hypothesis was rejected at a 1% level of signi cance, therefore the analyzed variables are correlated.e KMO test statistic, with a value of 0.718, shows that the proportion of the variance in the data can be considered as common to all of the variables, thus validating the use of the factor analysis in this study.Note that the subprime crisis variable was removed from the sample as it did not present a high correlation with any other variable used.e variable for CEO with a stricto sensu post-graduation course in any area was also excluded as it presented a correlation of 100% with the variable for CEO with stricto sensu post-graduation in the area of business, in order to avoid redundancy.As observed in Table 4, seven factors were extracted as they presented Eigen values above 1.00.ese factors, in their totality, are able to explain 69.91% of the cumulative variance.
In Table 4 it is veri ed that the variables from each dimension -pressure, opportunity, and rationalization -, numbered in tables 1, 2, and 3, were distributed between the factors, except factors 3 and 6, in which variables from the rationalization dimension were grouped.Due to this dispersion of variables, it was decided not to rename them, thus preserving the individual characteristics of the variables with high factor loads, grouped in the di erent factors.erefore, factor 1 contemplates variables from the pressure and opportunity dimensions; factor 2 contains variables from the opportunity and rationalization dimensions and control variables; factor 4 contains variables from the pressure and opportunity dimensions; factor 5 contemplates variables from the three dimensions and one control variable; and factor 7 is composed of variables related to the pressure and rationalization dimensions.
The factors presented in Table 4 were used as independent variables to process equations 1 and 2, which were previously described and whose results are available in Table 5.Note that the multinomial logit panel model was operationalized by the Stata v.13 so ware.From examining the results in Table 5, it is veri ed that the results of the Cramer and Riddler test (1991) indicate that the groupings among the alternatives cannot be carried out at a 1% level of signi cance.us, the econometric analysis of hypothesis 1 cannot be carried out via simple pooling of the data and it is necessary to treat them as a panel and consider the existence of nonobserved heterogeneity between the banking institutions.

Michele Rílany Rodrigues Machado & Ivan Ricardo Gartner
at is, the banks present peculiar characteristics that di erentiate them over time.So, the multinomial logit model with random e ects represents the most appropriate model for estimating the parameters of the functional relationships established in the study.
e additional comparison tests also indicate that the logit model with random e ects best adjusts to the data. is can be observed in the results of the LR test and via the AIC and BIC information criteria.e LR test shows a p-value of less than 0.05, implying that the constrained model is more adequate.e values of the AIC and BIC information criteria for the multinomial logit model with random e ects were lower than those of the traditional logit model, which enables it to be inferred that the model with random e ects appear to be more adjusted to the data in the study.
Considering the probability of indications of corporate fraud, in Table 5 (column B1) it is observed that factor 1 presents a positive and signi cant relationship.In this factor, the variables with the highest factor load are: level of market share, size of the executive board, size of the institution, quarterly remuneration, and deviation from remuneration paid by the market, in that order, and with a factor load above 0.85 (Table 4, column A).It bears mentioning that these variables were characterized as representative of the pressure and opportunity dimensions.
e level of market share and the variables interlinked with remuneration represent measures for company performance and compensation structure, which denotes elements of the pressure dimension.As for the size of the executive board and the size of the institution, these can be considered as explaining the opportunity dimension.
Due to factor 1 having indicated a positive behavior with the probability of indications of fraud, it is perceived that the variables derived from the pressure dimension present a coherent behavior with the ndings of Alexander and Cohen (1996) and Macey (1991), both from the area of criminology.In these, below-ideal performance can lead managers to prefer a higher level of risk to increase company performance, such as by manipulating results.With regards to the variables classi ed as remuneration items, Macey (1991) argues that in order to achieve their objective of satisfying or maintaining a particular level of income, managers can achieve this either through work and ability, or by becoming involved in criminal activities.
For the size of the executive board variable, it is veri ed that an executive board with a greater number of members can represent a measure of power compared with board of directors (Brazel et al., 2009) and can also imply an increase in monitoring costs and problems for the board of directors to coordinate these directors.As for the size of the institution, this can result in an environment that is more conducive to indications of corporate fraud.According to Alexander and Cohen (1996), what promotes an environment with a greater number of opportunities for committing fraud is the size of the organization.us, for the variables that denote measured items of the opportunity dimension in factor 1, it is observed that given that its factor loads are positive and the factor is positively and signi cantly related with the probability of indications of fraud, the results are consistent with the empirical results of studies on accounting fraud, agency theory, and criminology.
For the probability of corporate fraud occurring, it is veri ed that factor 2 (Table 5, column B2) shows a positive and signi cant relationship with the occurrence of corporate fraud.For this factor, it is observed that the variables with the highest factor load, above 0.84 (Table 4, column B), are the size of the scal council and companies under state control.With these, the former was considered as an element of the opportunity dimension and the latter as a control variable.e behavior of the size of the scal council was expected to be the opposite to that of factor 2. However, given the positive factor load, this was not found.is variable aims to monitor management acts, o er an opinion on particular issues, and express the shareholders' position (Trapp, 2009).It would therefore be a element of monitoring management actions.
As for the factor load for the state control variable, this shows an interesting result.Because it is positive, it is consistent with the relationship between factor 2 and the probability of corporate fraud occurring, which is positive and signi cant. is result is consistent with the most recent cases of corporate fraud in Brazil, such as those of Petrobrás and Correios (Brito, 2014;Ministério Público Federal, 2014), indicating that these companies provide greater opportunities to execute corporate fraud.
Also for the probability of occurrence of corporate fraud, it is observed that factors 4 and 5 indicate negative and signi cant behavior in relation to this probability.
As variables with loads higher than 0.62 (Table 5, column D), factor 4 presents the return on the assets of the banking institution and the size of the board of directors.As this factor is negatively related with the probability of fraud occurring, an alignment is perceived between the result obtained with the size of the board of directors, given that it is positively correlated with this factor.is result is consistent with agency theory, in that the composition of the board of directors is a fundamental corporate governance mechanism in market economies, as it exercises control over the executive board (Byrd, Parrino, & Pritsch, 1998;John & Senbet, 1998).As for return on assets, this indicated an unexpected behavior.Classi ed as a pressure element and especially as a measure of the institution's performance, its in uence was expected to be inverse to the probability of fraud, however this did not occur.e nancial intermediation revenue variable presented an expected result, consistent with factor 4 and therefore not contributing with its negative relationship to the probability of fraud occurring, which supports the writings of Alexander and Cohen (1996) and Macey (1991).
Finally, factor 5 gathered the following variables of the opportunity and rationalization dimensions of the Cressey fraud triangle (1953), as well as a control variable: independent members, predominance of females on the executive board, punishment, and convergence with the international accounting rules.Factor 5 presented a negative and signi cant relationship for the probability of occurrence of corporate fraud.In this, the variables with a factor load above 0.50 (Table 4, column E) showed a positive correlation with the factor, therefore contributing to the negative relationship found for the probability of corporate fraud occurring.Note that the results obtained for independent members are consistent with agency theory, given that a greater proportion of independent members -outsiders -on the board of directors reduces the probability of corporate fraud occurring, as seen in Beasley (1996).
As for the predominance of females on the executive board and punishment, these are in accordance with the writings from the area of criminology.e results for females are similar to those of Ste ensmeier et al. ( 2013) and the claims of Kelley et al. (1990), Reynolds (2006) and Zahra et al. (2007) concerning the behavior of females with regards to fraud.According to these authors, female managers are less susceptible to committing fraudulent acts than male managers.e results for punishment are bolstered by the studies by Becker (1968), Eide et al. (2006), andMurphy (2012) in the area of criminology and the economics of crime.According to these studies, punishment is an important situational factor in the decision-making process with regards to committing a criminal act or not.
As it was observed in the results of the research, variables from the three dimensions of the Cressey fraud triangle (1953), even when grouped in factors, were signi cant for measuring the probability of corporate fraud.is is consistent with the arguments of Cressey (1953), in which the absence of any one of the dimensions would prevent the violation of nancial trust; that is, committing fraud.Therefore, because of the results obtained with factors 2, 4, and 5, hypothesis 1, which states that the three dimensions of the fraud triangle together condition the occurrence of corporate fraud in Brazilian banking institutions, cannot be rejected.
Based on the results put forward in Table 5, which were discussed in the previous paragraphs, an econometric model is presented for measuring the probability of corporate fraud occurring in banking institutions, in its nal speci cation: 4

CONCLUSIONS
is article investigated the occurrence of corporate fraud in Brazilian banking institutions in the period between January 2001 and December 2012 using detection variables extracted from agency theory and the economics of crime, grouped according to the dimensions of the Cressey fraud triangle: pressure, opportunity, and rationalization.From agency theory, variables were identi ed that enabled the measurement of the pressure and opportunity dimensions pertaining to the instruments for monitoring manager actions, such as remuneration incentives and corporate governance.From the economics of crime, technical and empirical studies were used that enabled the identi cation of variables for measuring the pressure and rationalization dimensions, such as indicators of pressure for company performance and manager demographic characteristics, with age, educational level, training in the area of business, and gender standing out.
e study con rmed the general hypothesis of the Cressey fraud triangle, in which breaking nancial trust is conditioned by the simultaneous existence of the three dimensions of the fraud triangle: pressure, opportunity, and rationalization.
is article is relevant in that it lls a gap in the literature in the area by carrying out a di erentiated analysis of frauds, contemplating all those that occur in the context of an institution and not being limited to only those of an accounting nature.Likewise, this study enabled the measurement of the probability of occurrence of corporate fraud by dissociating fraudulent banking institutions from those that only showed indications of corporate fraud, this situation not being found in neither the Brazilian or international empirical studies used.
Besides the theoretical relevance presented in the previous paragraph, it is worth highlighting that the variables used in constituting the factors, with statistical signi cance, can be treated as indicators of possible occurrences of corporate frauds.Moreover, identifying these variables will enable both regulatory bodies and investors to analyze the possibilities of fraud occurring, whether for the regulatory bodies to curb it or for investor decisions with regards to maintaining or carrying out new investments.
is research also contributes with an investigation of new debates and studies in the Brazilian academic eld on corporate frauds in the country, since besides being few in number these have di erent scopes to those of this study.33 (1953), in which the absence of any one of the dimensions would prevent the violation of financial trust; that is, committing fraud.Therefore, because of the results obtained with factors 2, 4, and 5, hypothesis 1, which states that the three dimensions of the fraud triangle together condition the occurrence of corporate fraud in Brazilian banking institutions, cannot be rejected.

CONCLUSIONS
This article investigated the occurrence of corporate fraud in Brazilian banking institutions in the period between January 2001 and December 2012 using detection variables extracted from agency theory and the economics of crime, grouped according to the dimensions of the Cressey fraud triangle: pressure, opportunity, and rationalization.From agency theory, variables were identified that enabled the measurement of the pressure and opportunity dimensions pertaining to the instruments for monitoring manager actions, such as remuneration incentives and corporate governance.From the economics of crime, technical and empirical studies were used that enabled the identification of variables for measuring the pressure and rationalization dimensions, such as indicators of pressure for company performance and manager demographic characteristics, with age, educational level, training in the area of business, and gender standing out.
The study confirmed the general hypothesis of the Cressey fraud triangle, in which breaking financial trust is conditioned by the simultaneous existence of the three dimensions of the fraud triangle: pressure, opportunity, and rationalization.This article is relevant in that it fills a gap in the literature in the area by carrying out a differentiated analysis of frauds, contemplating all those that occur in the context of an institution and not being limited to only those of an accounting nature.Likewise, this study The Cressey hypothesis (1953) and an investigation into the occurrence of corporate fraud: an empirical analysis conducted in Brazilian banking institutions ere were limitations for carrying out the study, especially during the data gathering process, due to the existence of few institutions that presented the totality of their data series for the period analyzed, mainly impacting on the obtainment of information on manager demographic characteristics.Another limitation was the temporal range of the data, covering January 2001 to December 2012, due to the fact that via Circular Letter n. 3,630 of 2013(CB, 2013) the CB relieved banking institutions of presenting QFI from January 2013 onwards.
e decision made obtaining the quarterly data from 2013 onwards unviable.is is added to the limitations of the research from using punitive administrative proceedings as proxies for the occurrence of corporate fraud; in future studies, judicial proceedings could be used for this purpose.
It bears mentioning that the models that use accounting variables can cause the problem of variable endogeneity, which requires speci c treatment.In linear panel data models, the endogeneity of the regressors can be controlled using the generalized moments method (GMM) technique.However, controlling the endogeneity of the regressors in non-linear panel data models, especially in multinomial logit and probit cases, is something that is still being developed and is not yet available in the literature.For this reason, this problem was not considered in this article, representing a technical-scienti c limitation to be addressed at some point the future.
For future studies, we propose an in-depth analysis of other measures of manager remuneration incentives, such as share bonuses, and their e ect on the probability of corporate fraud occurring.We also suggest replicating this study for other activity sectors, such as non-nancial publicly-traded companies.
, as can be observed in the rules of the American Institute of Certi ed Public Accountants (AICPA, 2002) -Statement on Auditing Standards n. 99 (SAS n. 99) -, of the Federal Accounting Council (CFC, 2009) -Resolution n. 1.207 of 2009 -, and of the International Accounting Standards Board (IASB) -ISA 240 of 2009.

Figure 3
Figure 3 Flowchart to determine the most appropriate econometric modeling Source: Elaborated by the authors Crutchley et al. (2007),Erickson et al. (2006),Troy et al. (2011),Wang et al. (2010) 9 D_DEV_REM Dummy for remuneration received below the average remuneration paid by the market of publicly-traded banks.Elaborated by the authors.RFs = ReferenceForms; AIFs = Annual Information Forms; QFI = quarterly nancial information.

Table 1 -
Variables used -pressure dimension

Table 3
Variables used -rationalization dimensionTheCressey hypothesis (1953)and an investigation into the occurrence of corporate fraud: an empirical analysis conducted in Brazilian banking institutions

Table 4
Factor analysis: hypothesis 1TheCressey hypothesis (1953)and an investigation into the occurrence of corporate fraud: an empirical analysis conducted in Brazilian banking institutions Source: Elaborated by the authors.