Credit Scoring Models: an Updated Review Guilherme Barreto Fernandes Nº 93 December 2015

The G-Sifi Concept and its Implications for Local Subsidiaries

The Importance of Image as a Distinctive Feature in the Financial Market

Capital Planning in Good and Bad Times

Marcelo Petroni Caldas Frederico Turolla 20

Eduardo Borges da Silva Benjamin Miranda Tabak

Lourenço Miranda 

31

9 3>

9 772177 60311 7

Positive Prepare sua empresa para extrair o máximo de benefícios



Saiba como explorar todo o potencial dessa nova realidade



Conceda mais crédito com menos riscos

 

Aperfeiçoe sua gestão de clientes



Minimize prejuízos causados por fraudes e inadimplência

dos dados positivos com a ajuda da Serasa Experian. Positive-se! Contrate agora mesmo a Consultoria Serasa Experian em Cadastro Positivo e rentabilize suas decisões de negócios.

Para saber mais, acesse serasaexperian.com.br/cadastropositivo ou ligue para 3004 7728

Aumente o retorno das suas ações de cobrança

+

++

3

06 Credit Scoring Models: an Updated Review

Guilherme Barreto Fernandes

The article discusses the evolution of Credit Scoring techniques over the past 60 years, beginning with traditional regression models all the way through to elaborate algorithm ensembles, enabling more accurate credit risk measurement. The paper also emphasizes the importance of new information sources.

20 The G-Sifi Concept and its Implications for Local Subsidiaries

Marcelo Petroni Caldas Frederico Turolla

After the 2008 crisis, developments took place in the regulation of financial institutions and their subsidiaries as a result of concerns associated with systemic risk. The regulatory response included the definition of G-SIFI - Global Systemically Important Financial Institutions, whose discontinuation would impact the global economy.

31 The importance of image as a distinctive feature in the financial market

Eduardo Borges da Silva Benjamin Miranda Tabak

The study innovates on the concept of how the image effect relates with aspects of financial decision making. The purpose of the study was to investigate how decision-making relates with the image individuals have of a situation as they make a risky financial investment. The study also relates biological factors, in particular the androgenic hormone testosterone.

 Capital Planning in Good and Bad Times Lourenço Miranda

Professor Lourenço Miranda’s article emphasizes the historical truth that it is far more difficult to manage risk in good times than in adverse ones. He warns that one does not manage a business as if a storm would abate every month, but, when such a storm does come, one must be prepared for action and know how to act. The fact is that risk exposure rises or declines with the economic cycles and, therefore, the capital management strategy must address the cyclicality of risk.

4

From the Editors This last issue of 2015, like its predecessors, presents articles relevant to readers in search of news on means at the disposal of finance practitioners in support of credit-granting and risk-management decisions. Guilherme Barreto Fernandes, Manager in charge of Serasa Experian’s Fraud & Consulting in Analytics team, an expert with ample experience in credit risk modeling, offers an article in which he provides a brief history of the evolution of credit and credit-scoring models, and talks about the importance of their application to the credit cycle. The author points out that, although credit has been granted for more than 4,000 years, Credit Scoring as we know it has only been in existence for 60 years, and is intended to identify good and bad obligor profiles. Statistical and mathematical techniques with this in mind began in the 1930s with Fisher (1936) and traditional regression methods, through to elaborate model ensemble algorithms that enable more accurate measurement of each individual’s and each operation’s credit risk. The article also discusses some non-traditional data methods and emphasizes that, regardless of the technique, new sources of information must be included all the time, as the marginal gain is always greater on this dimension. The article by Professors Marcelo Petroni Caldas and Frederico Turolla intends to investigate the impact of new Basel Committee regulations on the required capital of financial institutions. It focuses on a theoretical essay based on documental analysis of the compendia published by the Basel Committee and risk-management reports disclosed by the relevant financial institutions. The topic is novel and current and, according to the authors, has limited literature, particularly from the International Business angle. The article therefore contributes to knowledge of the concepts of business-firm internationalization, home host, risk management, and internal controls, given its focus on the financial system. The article addresses secondary-data research that includes documental analysis to determine the impact of these rules on global financial institutions with subsidiaries locally active in Brazil. It finds that the majority of the relevant banks were compliant with the rules in question on the base date for the study. Other banks require some effort to buttress their higher-quality capital. This issue also includes an article by renowned Economics scholars Dr. Eduardo Borges da Silva and Dr. Benjamin Miranda Tabak, in which they emphasize that image effect is related with decision-making aspects and is an important factor to understanding the decision-making process. Their study innovates by adding a new factor to the financial decision making process. An experiment was designed in which volunteers were asked to decide how much to invest in a specific option for which a visual of a financial advisor was provided. Two different images were used, one with the financial consultant in a suit and tie and working

5

on a laptop computer while surrounded by books, while in another he wore shorts and looked casual. The experiment’s main contribution is that volunteers with a 2D:4D ratio greater than 1, that is, less exposed to testosterone, do not greatly change opinions regardless of the analyst’s image, and retain their risk-averse convictions. The purpose of the study was to investigate the relationship with the decision made and the view individuals have a situation as they make a risky financial investment. Professor Dr. Lourenço Miranda, AIG Executive Officer for Economic Capital Modeling and Stress Testing, offers an excellent article in which he discusses the subject of pro-cyclical capital management, based on the fact that it is easier to manage credit risk in difficult times than in easy ones. He emphasizes that companies maintain a capital cushion to absorb the rising exposure during economic downturns, or manage the capital available and their risk in harmony with the cycles. This is the mechanism known as procyclical capital management. The greater the agility balancing demand for capital (risk exposure) and available capital, the smaller the capital cushion needed to manage a company. In this sense, stress testing helps banks prepare for unexpected losses. Because available capital will inevitably be reduced when a major loss hits a bank, the quicker such a bank can rebalance its capital after the event, the smaller the capital cushion it will have to maintain without compensation (that is, unavailable for investment). Taking on excessive risk may have dire consequences – and, in modern history, the taxpayer has been footing the bill.

CREDIT TECHNOLOGY

YEAR XIII

Trimonthly published by Serasa Experian

Nº 93

ISSN 2177-6032

President - Brazil

Cover, Desktop Publishing and Ilustration

José Luiz Rossi

Gerson Lezak

Business Units Presidents/Superintendents

Translation

Júlio Leandro, Mariana Pinheiro, Steven Wagner e

Allan Hastings

Vander Nagata

Correspondence Serasa Experian - Comunicação & Branding Alameda dos Quinimuras, 187 - CEP 04068-900 - São Paulo - SP

Directors Amador Rodriguez, Guilherme Cavalieri, Lisias Lauretti,

www.serasaexperian.com.br Manzar Feres, Paulo Melo, Sergio Fernandes e Valdemir Bertolo [email protected] Responsible Editor Rosina I. M. D’Angina (MTb 8251) The concepts issued in the signed articles are the responsibility of the authors, which do not necessarily express the point of view of Serasa Experian and the Editorial Council. Total or partial reproduction of the articles hereby published is strictly forbidden.

6

Credit Scoring Models: an Updated Review

Guilherme Barreto Fernandes

7

Abstract Credit scoring models are currently used by creditors in every industry: banks, finance companies, retailers, and more. This article provides a brief historical survey of credit practices and the use of credit scoring models throughout the credit cycle. It then addresses the evolution of modeling techniques, starting with papers from 1941 and ending with recent papers from 2015. It concludes with a discussion of new informational techniques, as well as unorthodox sources such as Web data.

1. Introduction The concept of credit granting dates back from 2000 b.C.E. (Lewis, 1992), that is, from the Babylonian age. Evidence exists that farmers of that time already resorted to loans to regulate their cash flow relative to harvests. But it was only in the 1920s that the credit industry revolution took place, boosted by the demand for automobiles. Although credit has been granted for 4,000 years, the concept of Credit Scoring as we know it has only existed for six decades. By definition, the purpose of Credit Scoring models is to identify the profiles of good and bad creditors, whatever the meaning of “good” and “bad” may be. Statistical and mathematical techniques to this end emerged in the 1930s with Fisher (1936), who introduced discriminant analysis to identify two varieties of the iris flower. Durand (1941) was first to apply discriminant analysis to the identification of good and bad customers. But his project was ultimately limited to research and was never used in actual practice. After WWII, the United States experienced an age of growth and the use of automated credit-granting methods became a real necessity. In 1956 Bill Fair and Earl Isaac founded the first consultancy intended to help banks, finance companies and all other business firms that extended credit to develop and implement Credit Scoring models. Fair-Isaac, now renamed simply Fico, remains one of the main providers of credit support solutions in the United States, along with Experian, Equifax and TransUnion. Thomas et al. (2002) define Credit Scoring as the set of decision-making models that support credit granting, together with the techniques on which they are based. Credit Scoring models are often tasked with measuring a client’s cre-

8

ditworthiness. Credit Scoring models are expected to indicate what clients are appropriate risks given a bank’s or finance company’s policy. Based on a Credit Score, a bank must essentially make two decisions regarding a credit application: 1) whether or not to grant credit; and 2) the amount of money to grant. Although pricing can take place based on models the end price is defined by the marketplace, so that models can only suggest a price floor. Thomas et al. (2002) argue that the philosophical motivation behind the study of Credit Scoring lies in pragmatism and empiricism. Models are not intended to explain the risk of default but to predict it. Still according to Thomas et al. (2002), the essence of consumer Credit Scoring models must be based on the use of historical data and every variable that is relevant to risk prediction must be included. When it comes to measuring the risk of business firms, credit managers wield greater decision-making power (Thomas et al., 2002). Even so, automated decision-making methods, including Credit Scoring, are increasingly used in connection with business firms (Altman and Saunders, 1998). Altman (1968) developed the first Credit Scoring model intended to predict not defaulting, but a company’s insolvency based on its balance sheet. The product of his work was Altman’s z-score, which is often used in other studies (Orgler, 1970; Platt and Platt, 1990; Rosenberg and Gleit, 1994; Lennox, 1999; Shurmway, 2001; Keilhofer, 2003; Pesaran et al., 2003; Duffie and Singleton, 2003; Ravi Kumar and Ravi, 2007). His paper provided the groundwork for both academic studies on corporate credit risk and the risk rating systems banks use to this day. Evidence of this is that, even 30 years later, Altman (2000) revisited his z-score and redid the estimates with more recent data. Bear in mind that Credit Scoring is not synonymous with rating. The former is based purely on quantitative and empirical methods, while the latter provides a credit-analysis based rank, and may or may not take into account the output of a Credit Scoring model (Sicsú, 2010). According to Sicsú (2010), the benefits of using Credit Scoring models to grant credit include: • Consistent decision-making; • Quick decision-making; • Decisions in line with the bank’s or financial institution’s appetite for risk; • Remote decision-making; • Risk-based credit portfolio monitoring and management. The above benefits might also include more practical regulatory oversight. The bank becomes aware of the quality of the portfolio it is building, and the regulator may prescribe allocation and provisioning.

2. The Credit Cycle Figure 1 shows the credit cycle and the main phases where opportunities exist to develop models. Credit Scoring models have different names on di-

9

fferent portions of the cycle: Prescreening Scoring, Application Scoring, Behavior Scoring and Collection Scoring (Thomas et al., 2002).

Figure 1: The Credit Cycle.

Management

Assignment

Granting

Reactivation Reativação

Prospecting Prospecção

Collection Prospecção Concessão Gestão Cobrança Reativação Prescreening Cessão

Prospecting Granting Management Collection Reactivation Scoring Assignment

2.1 and Application Prescreening and Application Scoring models are used in the prospecting and granting phases. Two important differences distinguish these models: 1) target public and 2) available independent variables. An Application Scoring model assumes that there will be a well defined credit application, with information on the operation such as percentage payment up front, term, collateral and more. On the other hand, Prescreening models measure the risk of potential customers that the bank prospects for “in open sea”, and there is no need for a specific application to be involved. In this case, the explanatory variables are those available from credit bureaus and public sources. 2.2 Behavior Scoring Behavior Scoring models are crucial to compliance with the requirements of Basel II and III (BCBS, 2006). They attempt to measure the risk of default for customers to whom credit has already been granted. A key element in the construction of a Behavior Scoring model is the sample’s composition. In general, the statistical models used assume independence across observations in the database. However, when several snapshots of a single credit portfolio are used over several months, the same customer may be present more than once. This is particularly true of installment products, such as personal credit and automotive and real-estate financing. The independence across observations, in observation of the statistical concept, is easily refuted in this case. Behavior Scoring models are also marked by the abundance of predictive information: reference file information, operation features, operation payment history, credit bureau information, such as derogatory notes and credit usage

10

history. It is no coincidence that these models show far superior predictive capabilities than the others. From the customer-management viewpoint, Behavior Scoring models can be used in facility renewal processes and cross-selling actions. They can also be used to manage credit portfolios and in the credit allocation and provisioning for bad credits processes. 2.3 Collection and Recovery Scoring The models used in a credit portfolio’s collection phase are referred to as Collection Scoring models. In practical terms, these models are segmented according to the bank’s or finance company’s collection rule. This segmentation phase significantly leverages model performance in the identification of the customers most likely to pay their debts. These models are generally the basis for the collection strategy, defining the public of customers or operations whose rules are to be held back, accelerated, or even shifted straight to the end-stages. Recovery Scoring models are used in the final phase of the collection rule, over 180 days after payment is due, and support the identification of operations subject to in-court collection and those that may be traded by means of credit assignment deals.

3. Modeling Techniques The methods used to develop Credit Scoring models include all kinds of mathematical, statistical and computing techniques. The very first model (Durand, 1941) used discriminant analysis to arrive at the final formula. Hand and Henley (1997) provide a complete review of the methods: discriminant analysis (Durant, 1941), simple linear regression (Orgler, 1970 e 1971), linear programming (Hand, 1981; Kolesar and Showers, 1985), regression trees (Makowiski, 1985; Coffman, 1987), expert systems (Zocco, 1985; Davis, 1987), neural networks (Rosenberg e Gleit, 1994), nonparametric smoothing methods (Hand, 1986) and logit regression (Wiginton, 1980). Baesens et al. (2003) offer a more comprehensive review of methods. They compare the logit regression, linear discriminant analysis, linear programming, linear, support vector machine (4 algorithm variations), neural networks, Bayesian probabilistic networks, regression trees (algorithm C4.5 and 4 application variations) and KNN (k-nearest neighbor classifier – 10 and 100 clusters). The comparison of methods is based on application to 8 real databases. Graph 1 shows their results. The methods are ranked by ordering the techniques according to each database provided by the authors. Because they used 8 data banks, they were able to calculate average rank, maximum rank and minimum rank. Note that being ranked “1” indicates the best performing technique (AUROC: area under the ROC curve; Andrade and Oliveira, 2002) for a given database.

11

Graph 1: AUROC methodology ranking results (Table 4 Graphic 1: AUROC methodology ranking results (Table adapted from Baesens et al. (2003)) adapted from Baesens et al. 4(2003)) 1

1 1 3

7

5

9

7

11

9

13 15 17

2

3

4

5

6

7

8

9

10

11

12

13

14

14

15

15

16

17

Técnica de modelagem

5

Ranking

Ranking

3

Gráfico 1: Resultados do ranking de AUROC por Modelingde technique metodologia (Tabela 4 adaptada Baesens et al. (2003)) 1

2

3

4

5

6

7

8

9

10

11

12

13

16

17

11 13 15 17

Ranking médio Average RankingRanking médio

Ranking Médio Ranking Máximo 1: Neural networks Ranking Mínimo

Ranking Máximo Maximum Ranking Ranking Máximo

Ranking Mínimo Minimum Ranking Ranking Mínimo

Average Ranking Maximum Ranking 7: Naive Bayesian networks Minimum Ranking Key to techniques

13: Linear programming

2: SVM RBF LS

8: KNN100

14: C4.5 4

3: Logit regression

9: SVM RBF

15: C4.5 3

4: SVM LF LS

10: SVM LF

16: KNN10

5: Linear discriminant analysis

11: C4.5 1

17: C4.5 2

6: Bayesian probabilistic networks

12: Quadratic discriminant analysis

Please note that although the neural network and SVM RBF LS techniPlease note that although the neural network and SVM RBF LS techniques rank better ques rank better than the others on average, five other methods ranked first on than the others on average, five regression, other methods SVM rankedLF firstLS, on at least one database: logit Bayeat least one database: logit discriminant analysis, SVM LF LS, discriminant analysis, Bayesian probabilistic networks, and the susianregression, probabilistic networks, and SVM LF. The authors find that despite SVMresults LF. The for authors find that despite the superior neural andthe SVMdiscrimiperior neural networks and SVMresults RBFfor LS, thenetworks fact that nant analysis and logit regression techniques show good AUROC results RBF LS, the fact that the discriminant analysis and logit regression techniques show good indicates AUROC a weakresults non-linearity between covariables and theand output indicates a relationship weak non-linearity relationship between covariables the variable output (good/bad variablecredit). (good/bad credit). A little more than ten years after the publication of Baesens et al. (2003), A little more than(2015) ten years after the publication of Baesens et al.The (2003), Lessmann et al. Lessmann et al. published an updated version. latter article evaluates published an updatedsuch version. The latter articleand evaluates new modeling new(2015) modeling techniques, as Classifi cation Regression Tree (CART), Extreme Learning Algorithm J4.8, regularized and techniques, such asMachine, Classification and Regression Tree (CART),Logistic Extreme Regression Learning Voted Perceptron. also compare Homogeneous EnMachine, Algorithm They J4.8, regularized Logisticmethods Regression combining and Voted Perceptron. They sembles, which combine models using the same modeling technique. These tealso compare methods combining Homogeneous Ensembles, which combine models chniques are based on the Bagging, Boosting and Random Forest methods. Lessmann et al. (2015) also compare Heterogeneous Ensemble combination methods, divided into direct static, indirect static and dynamic methods. In all, they compare 1,141 methods on the 8 bases used previously in Baesens et al. (2003). Figure 2, below, indicates the modeling and model combination flow (source: Lessmann et al., 2015).

métodos de Bagging, Boosting e Random Forest. Lessmann et al. (2015) também comparam métodos de combinação heterogêneos (Heterogeneous Ensamble) divididos entre métodos estáticos direto, estáticos indiretos e dinâmicos. Ao todo são comparados 1.141 métodos nas 8 bases já utilizadas por Baesens et al. (2003). A figura 2 abaixo indica o fluxo de modelagem e combinação de modelos (fonte: Lessmann et al., 2015). Figure 2: Classifier development and evaluation process (source: Lessmann et al., 2015)

Figura 2: Fluxo de modelagem e combinação de modelos (fonte: Lessmann et al., 2015)

Lessmann et al. (2015) show that financial gain may be measured considering different cost functions for the incorrect classification of a good or bad credit. According to Thomas (2011), logit regression is by far the most frequen7 tly used Credit Scoring technique and, as such, Lessmann et al. (2015) show that the neural network, Hill-climbing Ensemble Selection combined with Bootstraping (heterogeneous ensemble of several models, and) and Random Forest (homogeneous ensemble) techniques may improve the cost function by up to 10% compared to logit regression. Naturally, there is no single best modeling technique, and the speed at which new modeling techniques or model ensembles are created make it impractical to run a comprehensive comparison for each Credit Scoring process. Thus, common sense must drive model development in practice. Factors such as systemic implementation feasibility, the need to objectively explain the variables that determine risk, systematic Credit Score tracking, and the development deadline must limit the techniques used in such a project at a bank or finance company.

4. Technical Advances in the Use of Information Modeling method advances may contribute relevant gains in Credit Score performance, but the information available is the main watershed. Traditionally, operation and customer data are used, but a customer’s credit risk may also be explained by three other dimensions: economic, relationship and digital (figure 3).

12

13

Figure 3: Traditional and emerging sources of information

Economic situation

Customer Operation Relationship among individuals

Internet and digital footprint data

4.1 Macroeconomic Information The macroeconomic scenario is a factor that may alter customers’ credit risk. This effect may be understood and modeled in two different ways: sensitivity to the economic situation may affect the credit risk of all customers homogeneously, or sensitivity may vary across customers (heterogeneous sensitivity). In brief, the former affects a portfolio’s overall rate of default, but retains the order of customers by risk. The latter is more flexible and allows risk reversals across groups of customers as a result of their sensitivity to economic factors. For example, a footwear company is highly sensitive to foreign exchange rate changes, but for a company in the services industry the rate of unemployment has greater impact in terms of credit risk. Banasik et al. (1999) explores survival analysis models relative to Credit Scoring and, almost ten years later, Bellotti and Crook (2008) introduce macroeconomic variables while using the same models. Belotti and e Crook’s (2008) models allow heterogeneous sensitivity. 4.2 Unstructured data The use of unstructured data lies strongly associated with the use of information obtained by means of web crawlers, algorithms that search the Internet for references based on previously provided keywords. No studies exist so far including this information in credit risk assessment, but the very discussion about the use of private information, however publicly available to anyone conducting a Google search, leaves the use of this information at a distance for Credit Scoring models. However, unstructured data may be obtained from sources already present at banks and finance companies. De Andrade (2014) uses the transcript of

14

a collection call to create variables and form a Collection Scoring model. In this case, Text Mining (Tan, 1999) and N-Grama (word sequence) creation techniques are used together with natural language processing (Hobbs et al., 1982). It is currently estimated that 85% of the information stored in databanks concerns unstructured text (Bess, 2003). De Andrade’s (2014) results show that the format holds as-yet unexplored value. 4.3 Individual Relationships Every credit operation may be connected to others in many ways: applicant, warrantor, credit analyst, credit provider, billing address, telephone number, to name a few. This, as Souza (2013) shows, this information is relevant within the fraud-risk context to identify operations connected with past frauds and thereby minimize the risk of loss. Souza (2013) uses the Social Network Analysis technique to create predictive variables such as number of connections and centrality level. However, credit risk may also be established based on connections between customers or operations. A recent article on The Fiscal Times (2015) mentions a Facebook patent to assign a credit score to an individual based on his or her network of friends. In the same article, Fico spokesman Goethe clarifies that this information does not supersede those available on payments history and debt taken. In fact, for publics such as young adults and recent immigrants, which have little credit bureau information in the United States, the analysis of Facebook friends may be a helpful tool as it provides a living-standards estimation. Another form of relationship among customers is geographic distance. Fernandes (2015) shows that, in order to evaluate the credit risk of small and medium sized businesses, one may estimate a risk measure called Spatial Risk that translates the effects of the local economic activity level. This metric uses defaulting information on neighboring business firms subject to the same local economic influence. Krigging (Matheron, 1963) is used to incorporate the spatial dependency factor, and the performance gain for a Credit Scoring model was important and statistically significant.

4. Modeling Techniques Vs. Information Variety In Economics, the Cobb-Douglas production function (Cobb and Douglas, 1929) describes the level of production given investments made in capital and labor. Investment in one of the two dimensions raises the productivity level, but as investment in one dimension increases, the marginal production gain decreases. Similarly, one may define the performance gain function of Credit Scoring models according to investment in technique and data. Figure 4 illustrates the notion. Note that we are offering that a greater investment in technique (t) than in information (i) is needed to generate a similar performance increase (level curves).

15

Figure 4: Credit Score performance gain function given investment in technique and information (based on the Cobb-Douglas production function)

Investment in technique

Credit Scoring performance curves

Direction of the gain

Investment in information Curvas de desempenho... Credit Scoring performance curves Investimento em técnica Investment in technique As Lessmann et al. (2015) point out, the various modeling techniques geInvestimento em informações Investment in information Direção de ganhogains relative to the usual Direction of the On gain the other hand, small performance metric.

nerate Belotti and Crook (2008), Souza (2013), De Andrade (2013), and Fernandes (2015) show that the performance gain from introducing new information can be quite significant. In these papers, technical innovation is a means to incorporate new information sources into Credit Scoring models.

5. Conclusion This article succinctly presents the historic evolution of credit and Credit Scoring models, as well as their use throughout the credit cycle. Credit Score modeling techniques have advanced significantly over the past 60 years since their creation. From traditional regression models to elaborate model combination algorithms, it has become possible to more accurately measure each individual’s and each operation’s credit risk. Finally, the article concludes by discussing some unorthodox methods to use data. However much techniques may evolve, new sources of information must always be included, as the marginal gain is greater on this dimension.

Author

16

Guilherme Barreto Fernandes Manager in charge of the Serasa Experian Fraud & Consulting in Analytics team, with close to 10 years’ experience in credit-risk modeling, fraud, and the development of analytical solutions for other businesses, such as telecommunications, insurance, utilities and large retail. Before joining Serasa Experian, Mr. Fernandes was active in credit-risk modeling areas. He has a bachelor’s degree in Statistics from UFSCar and a Master’s degree in Economics from Insper-SP. Guilherme may be contacted by e-

References

-mail: [email protected]

ALTMAN, E. (200): Predicting financial distress of companies: revisiting the Z-score and ZETA models. Working Paper Stern School of Business, New York University. ANDRADE, F.W. e OLIVEIRA, J.G. (2002): Comparação entre Medidas de Performance de Modelos de Credit Scoring. Tecnologia de Crédito, edição 33. BANASIK, J.; CROOK. J. e THOMAS, L. (1999): Not if but when loans default. Journal of Operations Research Society, Vol. 50, pp. 1185-1190. Banco Central do Brasil (2011): Edital de audiência pública nº 37. Acessado em https://www3.bcb. gov.br/audpub/edital/ExibeEdital.jsp?edt=50 acessado em 11/out/2012 às 15:30 (GMT+3). Banco Mundial (2012): Domestic credit to private sector (% of GDP). Acessado em 13/out/2012 em http://data.worldbank.org/indicator/FS.AST.PRVT.GD.ZS. BCBS: Basel Committee on Banking Supervision (2006): International Convergence of Capital Measurement and Capital Standards – A Revised Framework. Bank of International Settlements. BAESENS, B., VAN GESTEL, T., VIAENE, S., STEPANOVA, M., SUYKENS, J., & VANTHIENEN, J. (2003): Benchmarking state-of-the-art classification algorithms for Credit Scoring. Journal of the Operational Research Society, 54(6), 627-635. BELLOTTI, T. e CROOK, J. (2009): Credit Scoring with macroeconomic variables using survival analysis. Journal of the Operational Research Society, Vol. 60, pp. 1699 - 1707. BESS, C. et al. The grand challenges of information technology. In: Engineering Management Conference, 2003. IEMC’03. Managing technologically driven organizations: the human side of innovation and change; 2003, p. 610-5.

References

17

COBB, C. W.; DOUGLAS, P. H. (1928). "A Theory of Production". American Economic Review 18 (Supplement): 139–165 DAVIS, D. (1987): Artificial intelligence goes to work. High Tecnology, Vol. April, pp. 16-17. DE ANDRADE, V. L. (2014): Utilização de Técnicas de Dados Não Estruturados para Desenvolvimento de Modelos Aplicados ao Ciclo de Crédito Pontifícia Universidade Católica de São Paulo–PUC-SP. Acesso em 01/11/2015 em http://www.sapientia.pucsp.br/tde_arquivos/33/TDE-201409-17T06:43:58Z-15273/Publico/Valter%20Lacerda%20de%20Andrade%20Junior.pdf DURAND, D. (1941): Risk elements in consumer instalment financing. New York: National Bureau of Economics. FERNANDES, G. B., & Artes, R. (2015). Spatial dependence in credit risk and its improvement in Credit Scoring. European Journal of Operational Research. doi:10.1016/j.ejor.2015.07.013 GOPAL, R., MARSDEN, J. R., & VANTHIENEN, J. (2011). Information mining—Reflections on recent advancements and the road ahead in data, text, and media mining. Decision Support Systems, 51(4), 727-731. GRUNERT, J. e WEBER, M. (2009): Recovery Rates of Commercial Lending: Empirical Evidence for German Companies. Journal of Banking and Finance, Vol. 33, pp. 505-513. HAND, D. (1981): Discrimination and classification. Chichester, Wiley. HAND, D.; HENLEY, W (1997): Statistical Classification Methods in Consumer Credit Scoring: A Review. Journal of the Royal Statistical Society Series A, vol 160, Issue 3, pp. 523-541. HOBBS, Jerry R.; WALKER, Donald e AMSLER, Robert A. Natural language access to structured text. Proceedings… of the 9th conference on Computational linguistics 1. pp. 127–32, 1982. KEALHOFER, S. (2003): Quantifying Credit Risk I: Default Prediction. Financial Analysts Journal, Vol. 59, No. 1. Available at SSRN: http://ssrn.com/abstract=390141 KOLESAR, P. e SHOWERS, J.L. (1985): A robust credit screening model using categorical data. Management Science, Vol. 31, pp. 123-133. LENNOX, C. (1999): Identifying failing companies: a re-evaluation of the logit, probit and DA approaches. Journal of Economics and Business, Elsevier, vol. 51, issue 4, pp. 347-364. LESSMANN, S., BAESENS, B., SEOW, H. V., & THOMAS, L. C. (2015). Benchmarking stateof-the-art classification algorithms for Credit Scoring: A ten-year update. European Journal of Operational Research. 247(1), 124–136.

References

18

MAKOWISKI, P. (1985): Credit Scoring Branches. Credit World, Vol. 75, pp. 30-37. MATHERON, G. (1963): Principles of geostatistics, Economic Geology, Vol. 58, pp. 1246–1266. ORGLER, Y. (1970): Credit Scoring Model for Commercial Loans. Journal of Money, Credit and Banking. Vol. 2, No. 4, pp. 435-445. ORGLER, Y. (1971): Evolution of bank consumer loans with Credit Scoring models. Journal of Banking Residential. Vol. 2, pp. 31-37. PESARAN, M., SCHUERMANN, T., TREUTLER, B-J. e Weiner, S. (2003): Macroeconomic Dynamics and Credit Risk: A Global Perspective. Center for Financial Institutions Working Papers 03-13, Wharton School Center for Financial Institutions, University of Pennsylvania. PLATT, H.D. e PLATT, M.B. (1990): Development of a class of stable predictive variables: the case of bankruptcy prediction. Journal of Business Finance & Accounting, Vol. 17, Issue 1, pp. 31–51. doi: 10.1111/j.1468-5957.1990.tb00548.x RAVI KUMAR, P. e RAVI, V. (2007): Bankruptcy prediction in banks and firms via statistical and intelligent techniques–A review. European Journal of Operational Research, Vol. 180, Issue 1, pp. 1–28. ROSENBERG, E. e GLEIT, A. (1994): Quantitative Methods in Credit Management: A Survey. Operations research, vol. 42, no. 4, pp. 589-613. SICSÚ, A.L. (2010): Credit Scoring: desenvolvimento, implantação e acompanhamento. Editora Blucher, São Paulo, Brasil. SOUZA, Thaine Clemente. Fraude no E-Commerce: Uma abordagem com redes sociais. 2013. 51 f. Dissertação (Mestrado) – Insper Instituto de Ensino e Pesquisa, São Paulo, 2013. SHUMWAY, T. (2001): Forecasting Bankruptcy More Accurately: A Simple Hazard Model. The Journal of Business, Vol. 74, No. 1, pp. 101-124. TAN, Ah-Hwee Tan. Text mining: the state of the art and the challenges. 1999. Pacific-Asia Workshop on Knowledge Discovery from Advanced Databases – PAKDD’99. The Fiscal Times (2015): Can your Facebook profile really hurt your Credit Score? Acessado em 1/11/2015 em http://www.thefiscaltimes.com/2015/10/27/Can-Your-Facebook-Profile-Really-Hurt-YourCredit-Score THOMAS, L. (2011): Credit Scoring and the Edinburgh Conferences: Fringe or International Festival. Credit Scoring and Credit Control Conference 2011, Edimburgo, Reino Unido.

References

19

THOMAS, L., EDELMAN, D. e CROOK, J. (2002): Credit Scoring and its applications. SIAM Monographs on mathematical modeling and computation. SIMMEN, D. E., ALTINEL, M., MARKL, V., PADMANABHAN, S., & SINGH, A. (2008, June). Damia: data mashups for intranet applications. In Proceedings of the 2008 ACM SIGMOD international conference on Management of data (pp. 1171-1182). ACM. WIGINTON, J.C. (1980): A note on the comparison of logit and discriminant models of consumer credit behaviour. Journal of Financial quantitative, Vol. 15, pp. 757-770. ZOCCO, D.P. (1985): A framework for expert systems in bank loan management. Journal of commercial banking lending, Vol. 67, pp. 47-54.

20

The G-SIFI Concept and its Implications for Local Subsidiaries

Marcelo Petroni Caldas Frederico Turolla

21

Abstract The 2008 crisis triggered a movement towards the regulation of financial institutions and their subsidiaries based on certain concerns associated with systemic risk. The sweeping regulatory response included, among other initiatives, stipulating the existence of globally important financial institutions, or G-SIFI (Global Systemically Important Financial Institution) whose eventual discontinuation might impact the world economy. Literature on this recent subject is scarce, particularly from the angle of International Business, and addressing it contributes to improving the understanding of the concepts of business firm internationalization, home host, risk management, and internal controls, given the focus on the financial system. This paper uses secondary data, including documental analysis, to determine the impact on Brazil of such regulations affecting global financial institutions that have locally active subsidiaries. We find that the majority of the banks covered by the analysis match the potential application of the relevant rules on the base date for our research. The remaining banks require attention in the sense of some effort to reinforce higher-quality capital.

1. Introduction The 2008 crisis raised concerns among regulators in various jurisdictions about how to increment the supervision of risk management and controls for internationally active Financial Institutions (FIs). Another element that raised some concern was whether an actual risk-management governance framework was in place, or simply well-drafted flow charts that did not function properly in practice. All of these gaps, together with a good look into the rear-view mirror, led to the conclusion that these institutions had developed to such a point that they became financially powerful, with branches in many countries. When the crisis erupted, the term too big to fail came up, as the potential failure of any of these banks might compromise the entire financial system and have implications for the global macroeconomy, causing severe imbalances. Unfortunately, the word actually witnessed what these ex-post analyses found, evidencing the need for a new regulatory framework for the financial system to become safer. Until the emergence of the 2008 crisis, the conceptual framework for regulatory and risk management best practices found its basis on the Basel Committee1 and the documents it put forth, of which the most widely known are the Capital Accord and the New Accord, given their dissemination of important con-

22

cepts surrounding credit, market, liquidity and operational risks, and internal controls. The first accord to focus prominently on credit risk disseminated an indicator commonly known as the “Basel Index” (BIS, 1988 International Convergence of Capital Measurement and Capital Standards, July 1988), which represents a relevant and globally implemented solvency indicator for financial institutions that helps drive the decisions of regulators and the regulated. Starting from a reasonably simple formula where the institution’s regulatory capital (RC) must be greater than its credit, market and operational risk tranches (or required capital), one can generally determine when a bank is solvent, or does not necessarily have a quotient between these two variables equal to or greater than 8% (11% in Brazil2). The subject is extensive and setting its limits is determinant to developing a train of thought on the implications of new global rules set by supranational bodies for multinational financial institutions (FIs) with subsidiaries in Brazil. The subject has been covered in the literature with papers and studies by GOODHART, 2008 and ALLEN; CARLETTI; GU, 2015, respectively. Despite significantly contributing to the discussion on the financial system and the behavior o banks before invariably cyclic crises, we find that the studies do not address the impact of G-SIFIs on Brazilian subsidiaries of global FIs. This paper aims to fill this gap by investigating the potential impact of the new Basel Committee regulations issued on the required capital of the financial institutions at hand. Therefore, the purpose of this paper is to address an theoretical essay based on the documental analysis of compendia published by the Basel Committee and the risk-management reports published by the affected financial institutions.

2. Systemic Risk and the Development of Regulations The crisis and its aftermath revealed an urgent need for regulators to articulate in order to mitigate the risk of contagion followed by economic depression (GOODHART, 2008). Banks play a relevant role (ALLEN; CARLETTI; GU, 2015) because they are heavily interconnected in am increasingly globalized world where physical and psychic distances grow ever smaller. Sholud a certain bank run into solvency issues in a certain country, it may spread the symptom to other continents in which it operates, leading to the much-feared systemic risk. Bear in mind that topical and simple regulatory constraints, such as closing the bank and raising capital requirements without prudential measures may speed up the dissemination of a systemic problem (ACHARYA, 2009). Banking involves trust, and when a financial institution loses it, the chances of a bank run become relevant and, as a consequence, other banks – ultimately the central bank – must bail out the house in question to prevent contaminating the real economy. It is worth emphasizing the contribution from AHARONY; SWARY (1983), who analyze a sample of three banks that collapsed in the US financial market to conclude that the main cause of such events are operational risk events (asso-

23

ciated with internal and external fraud, process failures, etc.). Such events usually cause systemic, but low-impact and manageable, reflections. However, credit, market and liquidity risks tend to have an impact on other banks in the system, and this is quick to reflect on stock prices. Another study (FURFINE 2003) relies on an expanded sample using interbank payments and show that failures at this stage initiated by a single financial institution might immediately cause other credit, market and liquidity risks to crystallize, in a ripple effect capable of pervading the system. Give this context, the G20 group of the world’s 20 most relevant countries was created at the time. Its concern was to not allow an economic depression to set in worldwide and, on a different front, to add to the global financial system’s regulation. Thus came to be the FSB (Financial Stability Board), with the mission of coordinating global regulation and checking for implementation of its policies by means of a peer review process. Systemic risks and the so-called moral hazard are also part of its monitoring purview (FSB, 2012). Please note that the FSB is successor to the FSF (Financial Stability Forum), created in 1999 by the ministers and governors of the central banks of the G7 group of the seven most developed counties to watch over the stability of the international financial system (FSB, About the FSB). Given the above, it was found that some institutions had become too big to fail and that regulators were no longer willing to use the taxpayers’ funds to bail out banks that had made investments and/or extended credit in a risky manner or above their leverage limit (FSB, 2013). For an idea of the amounts involved, the Bank of America had 1.7 trillion US Dollars in assets in 2008 3, and JPMorgan had 1.8 trillion at the same time 4 Brazil’s GDP (Gross Domestic Product) in 2008 was 1.3 trillion US Dollars5. Therefore, the banks in question had become extraordinarily large and, as a consequence, had sufficient financial weight to trigger problems in any country in the event of failure of system lack of confidence. According to Stern and Feldman (2004), the issue surrounding the concept of “too big to fail” and the attitudes of US regulators in bailing out large banks clearly show that reforms are needed in the US and global financial system to prevent cyclic events or the formation of new bubbles capable of jeopardizing market stability. On the other hand, Basel had not yet completed its implementation of the second accord and had to quickly release a third accord whose focus was to drive regulatory attitudes in the face of the events that were under way, set liquidity risk management principles, and disseminate recommendations intended to reinforce financial institutions’ equity. The main goal was for banks to have higher quality capital, preferably in the form of paid-in equity capital. Although they lie beyond the scope of this article, it is worth noting that the 2008 crisis also led to the emergence of other initiatives. Both the Basel Committee and the FSB took measures to reinforce the world financial system’s soundness. It has become relatively commonplace to discuss the stress tests implemented by regulators worldwide. They have become best practice in measu-

24

ring the financial soundness of banks in the face of adverse scenarios. One may argue that, as important as a test of this magnitude, is the action plan to be deployed should an institution fail to pass the test. Such a plan is reasonably simple: a capital increase, but with higher quality capital, that is, the injection of new funds from shareholders. Depending on the circumstances, there may not be funds available to take on new risks. Another important study concerns shadow banking6 (GENNAIOLI, SHLEIFER; VISHNY, 2013). This survey focused on mapping and identifying credit providers that lacked the formalities of the traditional financial system and were therefore entities relatively removed from the traditional regulatory oversight model. Still, being active in credit granting, they offer facilities that the traditional system represented by banks often fails to provide.

3. Crisis and Regulation in International Business Literature Since the 2008 crisis it has been found that in periods of economic boom or prosperity financial institutions tend to alleviate credit granting and monitoring procedures and control. On the other hand, regulators understand that a stricter pattern sets in in challenging times as a natural form of selection for credit checks and collateral acceptance. Given this, regulators created buffers intended to cause banks to not be so restrictive in adverse economic periods, without compromising their capital base and remaining able to absorb losses, since their capital was reinforced during more prosperous times. Therefore, for the banks deemed to be systemically important, cyclic buffers were established, with implementation set to begin in 2016 and be complete by 2019. In Brazil, the increase in capital requirements is linked to two globally accepted concepts, as provided by the Central Bank of Brazil in Comunicado 20.165 of 2011, as follows: Capital conservation shall be an amount in addition to minimum regulatory requirements, made up of accepted elements for Core Capital. Its purpose is to increase the ability to absorb financial institutions’ losses beyond the minimum requirement in favorable periods of the economic cycle, so that the added capital can be used in times of stress. Countercyclical capital aims to ensure that the capital maintained by financial institutions addresses the risks arising from changes in the macroeconomic environment. Countercyclical capital must also be formed with elements accepted for Core Capital, and shall be required in the event of excessive credit expansion associated with potential systemic risk accumulation. (BANCO CENTRAL DO BRASIL, 2011)

25

It is worth emphasizing that, locally, the Central Bank of Brazil has established a time schedule in line with the Basel Committee’s global guidelines. As shown next, the new risk-weighted capital calculation procedures will enter into force in 2016, and so will their impacts7.

Regulatory Capital

2013

2014

2015

2016

2017

2018

2019

11%

11%

11%

10.5%11.125%

10.5%11.75%

10.5%12.375%

10.5%13%

4. Methodology It is relevant for the analysis at hand that the coining of the term G-SIFI - Global Systemically Important Financial Institution began with the attempt to identify the vectors that determine whether a bank is systemically important on a global level. In this sense, the Basel Committee implemented a data-gathering effort whose selected indicators attempted to reflect a range of FI attributes. These attributes will drive this study’s categories and include: (1) the size of banks, (2) their interconnections, (3) their complexity, (4) their global activities; and (5) the presence of substitutes in the event of discontinuity. A weight was assigned to each indicator for harmonization purposes (FSB, 2013). After depuration and practice of the method, by late 2013 the following list of 13 banks emerged: Bank of China, Group Crédit Agricole, Deutsche Bank, Unicredit Group, Mitsubishi UFJ FG, ING Bank, Santander, Nordea, UBS, Barclays, HSBC, Bank of America, JP Morgan Chase (FSB, 2013 Update of Group of Global Systemically Important Banks, November 2013) Based on the FSB list of the 13 globally important banks, a survey was conducted on the Central Bank of Brazil’s website, which includes a survey of the 50 biggest institutions by assets on the base date of December 2014. Our goal was to locate the respective banks and their representativeness in the system. Out of the 13 FSB-listed banks, 3 (23%) were not identified (Unicredit, Mitsubishi UFJ and Nordea). Therefore, out of the 13 initially mentioned in FSB’s list, only 10 (77%) were idetified as active among the top 50 banks by assets in the Brazilian financial system. Our analysis of the 10 banks with Brazilian subsidiaries reviewed their websites and found the public risk-management reports provided by the banks 8 . We located the number associated with the Basel index and compared it with the rules set for locally active G-SIFIs (Global Systemically Important Financial Institutions), should the implementation be valid for the information-collection base date.

26

5. Analysis of the Results This study simulated the capital reinforcement rule for the 10 globally important banks that are active in the Brazilian financial system. Should the rule be implemented on the study’s base date, the following would emerge:

Status

Share (%)

Financial Institutions

Compliant

50%

Santander, ING, Barclays, Bank of China, Bank of America

Barely compliant

30%

Deutsche Bank, Crédit Agricole, JPMorgan

Non-compliant

10%

HSBC

Information not available

10%

UBS

Source: developed by the authors

Concerning these indicators, the following observation were made on the March 2015 base date: • Santander Brasil operates as a universal bank, that is, in every segment of the Brazilian banking industry, and states that, based on simulations, it would have sufficient capital to meet the new requirements. Indeed it would, as its Basel index is at 16%, while the new rule (considering full implementation) provides a minimum of 13%. • Deutsche Bank is active in Brazil as an investment bank and makes no mention of compliance with the new rules, but, given that its Basel index is at 13%, we understand that, should the rules be in place, it would be barely compliant at the threshold of the indicator. • Crédit Agricole is active in Brazil as an investment bank and makes no mention of compliance with the new rules, but given that its Basel index is at 13.55%, we understand that, should the rules be in place, it would be barely compliant at the threshold of the indicator. • ING is active in Brazil as an investment bank and makes no mention of compliance with the new rules, but given that its Basel index is at 22% it would have sufficient capital to meet the new requirements.. Indeed it would, as the new rule (considering full implementation) provides a minimum of 13%. • HSBC Brasil operates as a universal bank, that is, in every segment of the Brazilian banking industry. It does not mention the steps it will take to comply with the new rules, but given that its Basel index is at 12,08%, we understand that, should the rules be currently in place, it would ne non-compliant and, therefore, some action plan would have to be implemented. It is worth pointing out that HSBC is in the process of divesting its Brazilian operations to Bradesco in 2015. • Barclays is active in Brazil as an investment bank and makes no mention of compliance with the new rules, but given that its Basel index is at 30.43%, it would have sufficient capital to meet the new requirements.. Indeed it would,

27

as the new rule (considering full implementation) provides a minimum of 13%. • Bank of China is active in Brazil as an investment bank and makes no mention of compliance with the new rules, but given that its Basel index is at 24.16%, it would have sufficient capital to mee the new requirements. Indeed it would, as the new rule (considering full implementation) provides a minimum of 13%. • JPMorgan is active in Brazil as an investment bank and makes no mention of compliance with the new rules, but given that its Basel index is at 13.76%, we understand that, should the rules be in place, it would be barely compliant, and slightly above threshold of the indicator. • Bank of America is active in Brazil as an investment bank and makes no mention of compliance with the new rules, but given that its Basel index is at 16.46% (considering full implementation) and the new rule provides a minimum of 13%, the bank is compliant. • UBS is active in Brazil as an investment bank and because we were unable to find the institution’s information on the Internet, the study was not viable. Our research with potential links leads to the bank’s overseas Website, which makes no mention of local reports.

6. Conclusions The study aims to explore the new classification for internationally active banks and shows that prospects are safe for half the financial institutions we covered, as they have sufficient capital to meet the new demand and still maintain operations growth. This simulated anticipation of the requirement is interesting, particularly when analyzing the banks that are at (and below) the indicator’s threshold, showing that capital adequacy plan must be drawn while the deadline for implementation of the regulatory demands is still favorable. Therefore, considering 2019 as the date of implementation, the exercise this article carries out applies to the visualization of certain actions to be emphasized in these financial institutions’ capital plans, not only locally, but globally as well. This paper is constrained by the fact that the methodology is recent and still being implemented by regulators around the world. As potential developments from this study, we recommend treating the same database with a methodology focused on exploring how much high-quality capital banks will need to meet the approved requirements.

28

Notes: 1 The Basel committee is a committee of Banking Oversight authorities formed by the governors of the G-10 central banks in 1975. The committee includes senior representatives from the central banks of Belgium, Canada, France, Germany, Italy, Japan, Luxemburg, the Netherlands, Spain, Sweden, Switzerland, United Kingdom and United States. It usually meets at the Bank for International Settlements in Basel, where its permanent Secretariat stands. 2 Central Bank of Brazil, Circular 2784/97 3 Annual Report http://media.corporate-ir.net/media_files/irol/71/71595/reports/2008_AR.pdf 4 Investor Relations http://files.shareholder.com/downloads/ONE/380315057x0x264148/ D882827A-99AB-44D8-9C82-3702CCAF5FD0/4Q08-Earnings-Supplement.pdf 5 Portal Brasil 6 Strengthening Oversight and Regulation of Shadow Banking 7 Central Bank of Brazil, Comunicado 20.615/11

Authors

8 Central Bank of Brazil, Circular 3.678/13

Marcelo Petroni Caldas Bachelor of Business Administration and Accounting from Universidade Presbiteriana Mackenzie, Graduate degree in Financial Accounting Management from Faap, MBA in Financial Management and Risks from Fipecafi –USP, and master’s candidate in International Business at ESPM. E-mail: [email protected]

Frederico Turolla Bachelor of Economics from Universidade Federal de Juiz de Fora, master of Economics from Brandeis International Business School, and Doctor of Economics from Fundação Getúlio Vargas (FGV). E-mail: [email protected]

References

29

Thematic Review on Supervisory Frameworks and Approaches for SIBs http://www.financialstabilityboard.org/wp-content/uploads/Thematic-Review-on-SupervisoryApproaches-to-SIBs.pdf Global systemically important banks updated assessment methodology and the higher loss absorbency requirement http://www.bis.org/publ/bcbs255.pdf Financial Stability Board www.financialstabilityboard.org/publications/r_091107c.pdf BANK OF AMERICA Annual Report http://media.corporate-ir.net/media_files/irol/71/71595/reports/2008_AR.pdf JPMORGAN Investor Relations http://files.shareholder.com/downloads/ONE/380315057x0x264148/D882827A-99AB-44D8-9C823702CCAF5FD0/4Q08-Earnings-Supplement.pdf BANCO CENTRAL DO BRASIL - 50 maiores bancos e o consolidado do Sistema Financeiro Nacional http://www4.bcb.gov.br/fis/TOP50/port/Top50P.asp BANCO CENTRAL DO BRASIL. Comunicado 20165, de 2011. FSB Progress and Next Steps Towards Ending “Too-Big-To-Fail” (TBTF) (September 2013) http://www.financialstabilityboard.org/wp-content/uploads/r_130902.pdf BANK FOR INTERNATIONAL SETTLEMENTS. Basel Committee on Banking Supervision. International Convergence of Capital Measurement and Capital Standards: a Revised Framework. Suíça, 2005. BANK FOR INTERNATIONAL SETTLEMENTS. Sound Practices for the Management and Supervision of Operational Risk. Basiléia, 2003. Blundell-Wingnall, Adrian and Paul Atkinson (2012), “Deleveraging, Traditional versus Capital Markets Banking and the Urgent Need to Separate and Recapitalise G-SIFI Banks”, OECD Journal: Financial Market Trends, Vol. 2012/1. ALLEN, F.; CARLETI, E.; GU, X. The role of baking in financial systems. In BERGER, Al.L.; MOLYNEUX, P. WILSON, J.O.S. (eds) The Oxford of Banking. Second edition. Oxford: Oxford University Press, p.27-46, 2015. AHARONY, J; SWARY I. Contagion effects of Bank Failures: Evidence from Capital Markets. The Jounal of Business. Vol. 56, No. 3(Jul., 1983), pp. 305-322.

References

30

FURFINE, C. Interbank Exposures: Quantifying the Risk of Contagion. Journal of Money, Credit and Banking. Vol. 35, No. 1 (Feb. 2003), pp. 111-128. ACHARYA, V. A Theory of Systemic Risk and Design of Prudential Bank. Journal of Financial Stability. Vol. 5, No. 3, (Sep 2009), pp. 224–255. STERN, G. H.; FELDMAN, R. J. Too big to fail: The hazards of banks Bailouts (2004). GENNAIOLI, N., SHLEIFER S.; VISHNY V. A Model of Shadow banking. The Journal of Finance. July 2013. LASTRA, R. M. Systemic risk, SIFIs and financial stability. Capital Markets: Law Journal. (2011).

31

The Importance of Image as a Distinctive Feature in the Financial Market

Eduardo Borges da Silva Benjamin Miranda Tabak

32

Abstract This paper supports the understanding that image effect is related with decision-making aspects and is an important factor to understanding the decision-making process. It innovates, however, by adding a new factor to the financial decision-making process. We designed an experiment in which we ask volunteers to choose how much money they would invest in a specific opportunity for which we presented the image of a financial consultant who provides some advice prior to the choice. We used two different images, one with the financial advisor wearing a suit, working on a laptop computer and surrounded by books, and another with the same advisor in shorts and looking casual. The experiment’s main contribution is that volunteers with a 2D:4D ratio greater than one, that is, less exposed to testosterone, do not greatly vary their decisions, regardless of the analyst’s image, and persist in their risk-aversion convictions. The change occurs mainly in participants who were more exposed to the hormone. Keywords: Risk. Experiment. Image. 2D:4D. Testosterone.

1. Introduction The decision-making process is a very important and often researched subject. The literature on it is comprehensive, but the topic remains controversial. Decisions are made without effective safety because several aspects must be taken into consideration. Making a decision is a choice that will involve consequences or developments at a future point in time. As a consequence, the outcome of choices do not always meet the decision makers’ initial objectives. Our choices are based on only on rationality, but also by behavioral biases. (TVERSKY & KAHNEMAN,1974). In our everyday lives we constantly ace situations, such as purchasing home insurance, what financial institution to select to manage our investments, whether or not to purchase a car, investing in a certain stock, etc. Because the future is not known in advance, decisions are always made in an atmosphere of uncertainty. Neoclassical economics adopts three assumptions in connection with decision making: first, that there is such a thing as the Homo-economicus, second that self-interest motivations exist, and, third, the ability to make rational decisions. These are the foundations of the expected utility theory. (VON NEUMANN & MORGENSTERN 1944). The theory became prevalent among theoreti-

33

cians. It is therefore believed that individuals are aware of the odds of every possible outcome, which are in turn associated with several states of nature, and can therefore calculate expected utility and, based on this calculation, make the decision most favorable to them. On the other hand, several later studies used experiments to confirm that people systematically make mistakes as a result of cognitive biases that violate several axioms from which the expected utility theory draws its support. See, for example, the Allais paradox (1953), Simon (1955), and the Ellsberg paradox (1961). This led to several experimental studies intended to identify situations in which agents deviate from the expected utility theory. (CAMERER & WEBER 1992). Kahneman and Tversky (1979) published ”Prospect Theory: An Analysis of Decision under Risk”. Prospect theory is an alternative to the expected utility theory. The essence of this theory is that what really matters to people, what really influences their behavior, is not the expected outcome of a choice, but their different reactions to gains and losses. It shows an asymmetry between the sense of loss and the sense of gain. People show greater propensity to risk when they are losing than when they are winning. The sense of realizing a loss is more intense than that of a gain, which is why people delay their losses the most they can and realize their gains as quickly as possible. Gains make them feel better. As a result, people are risk-averse when they are in the gains sector and less averse when they are in the losses sector (KAHNEMAN & TVERSKY,1979). The purpose of this study’s experiment is to investigate the relationship between decision-making with the image individuals have of the situation at the time of making a risky financial investment. In parallel, we also related biological factors, in particular the androgenic hormone testosterone, with the financial decision-making process in an atmosphere of risk and uncertainty. In this sense, we used a non-invasive method called 2D:4D to check for testosterone levels. The method measures the ratio of the index (2D) to the ring (4D) fingers. The 2D:4D ratio is sexually differentiated, so that men tend to have a lower 2D:4D ratio than women (Manning, et al., 1998; Manning, 2002). A ratio lower than 1, that is, an index finger shorter than the ring finger, indicates greater exposure to the hormone. The lower this ratio, the greater the prenatal exposure. Similarly, a ratio equal to or greater than one, where the index finger is equal in length or to or lengthier than the ring finger, indicates lower exposure to testosterone. Therefore, the greater the ratio, the lower the prenatal exposure to testosterone (Manning et al. 1998). This study aims to show that individuals do not act completely rationally in their investment decisions, and are also influenced by other behavioral effects. This study adds to the literature the use of two factors that jointly influence risky behavior, bringing into the discussion the image effect in association with exposure to testosterone as measured by the 2D:4D ratio. Baker, J. et al. (2004) and Paulins, A. V. (2005) show that more smartly dressed people generate higher expectations for the quality of the services they provide and that significant difference exists in service to customers depending

34

on what they are wearing. Rehman, S.U. et al. (2005) and Maruani A. et al. (2012) show that hospital patients have greater respect for doctors in white lab coats and that this image conveyed to patients a higher level of confidence in their practitioners. Rosenberg et al. (1986) created election pamphlets for fictional candidates to the US Congress, using male models for the pictures. The image effect was proven to be very strong, as “better looking” candidates got 59% of votes on average. In a second study, Rosenberg et al. (1991) only recruited female models for the pamphlets. They were considered to be ordinary women, but the researchers used makeup artists to create two photographic versions of each candidate: one that looked more competent and another that looked less so. A woman’s “competent” version was compared with the other’s “incompetent” version. The results showed that a more competent appearance produced a 15% polling advantage. The response to the experiment was very strong, as when a candidate was presented in her more “competent” version she would win the election, and lose it when presented in the “incompetent” version, by an average 15% margin in each case. Druckman (2003) simulated the US presidential race between Democrat John F. Kennedy and Republican Richard Nixon. The researcher’s impression was the same as the voters’ at the time. The image of the two candidates in the last debate broadcast on live TV, when Nixon appeared weakened by a stay in hospital, while Kennedy looked stronger and healthier, heavily influenced the election. These studies was heavily criticized, however, with claims that they were conducted in simulated environments and might not depict reality. To check for this, Todorov et al. (2005) gathered pictures of every winning candidate in US congressional elections of 2000, 2002 and 2004, including 95 running for Senate and 600 for the House of Representatives. They then asked the volunteers to evaluate the candidates’ competence based only on a quick look at the pictures, eliminating information on any face that a volunteer might recognize. The results showed that the candidates that volunteers regarded as more competent had won 72% of Senate elections and 67% of House elections, showing a success rate even higher than seen in laboratory conditions. After these findings, the researchers evaluated faces prior to the relevant elections and predicted the winners based only on the candidates’ looks. The result was that candidates that looked more competent won in 69% of gubernatorial elections and 72% of Senate elections. To relate individual risky behavior factors with biological variables, we list, next, several papers that studied the effects of testosterone on human behavior. Wingfield et al. (1990) and Dixson (1998) state that testosterone is a hormone that does not just influence maleness, but also plays an important role in explaining certain male behaviors. The 2D:4D ratio is a widely studied prenatal testosterone marker. Manning et al. (1998) and Manning (2002) confirm the existence of sexual dimorphism, indicating that men show a lower 2D:4D ratio than women. Peters et al. (2002) and Voracek et al. (2008) show studies proving that

35

this small difference in human hand anatomy was known since the late 19th century. Brown et al. (2002.a e 2202.b), Ökten et al. (2002) and Lutchmaya et al. (2004) found evidence for the hypothesis that the 2D:4D is related with prenatal androgens. Their studies show that a smaller 2D:4D ratio is associated with greater exposure to prenatal testosterone. Sapienza et al. (2009) and Garbarino et al. (2011) found that an individual subject to greater exposure shows greater propensity to risk, showing that men are therefore more inclined to take risks than women. Coates et al. (2009) carried out a study comparing exposure to testosterone and the rate of return on an investment portfolio. Apicella et al. (2008) adapted a game proposed by Gneezy and Potters (1997) and found that men with higher testosterone levels invested almost 12% more than men with average levels of testosterone. Da Silva et al. (2014) tested 141 children for a bias called endowment effect and found an inverse relation between the 2D:4D and delayed collection of a reward. Manning & Fink (2011) found a positive and significant correlation between the 2D:4D and adversity to uncertainty behavior in men and women. Millet (2011) provides an overview of the investigation into the association between the 2D:4D ratio and different types of economic behavior, suggesting an interaction between the 2D:4D ratio and human behavior.

2. Methodology Drawing inspiration from Rosenberg (1986), this experiment showed participants a picture of a volunteer introduced as an analyst. Said analyst provided financial information on a fictional company, recommending, by means of a text, that participants invest their money in a certain company because, in his view, the company’s stock would show a bigger return than market average. For the purposes of the experiment, we recruited undergraduate students from Brasília, in the Federal District. Several sessions were held in different days and locations, so that one group would not influence others. The sample was made up of 219 volunteers and included 92 men and 127 women (Average Age = 27.5 y.o. Standard Deviation = 7.6 years). Participants had an initial endowment of 1,000 monetary units (M.U.) and had to decide how much of it to invest in the company that the analyst recommended. Participants were instructed to select a fraction of this endowment, that is, to choose an amount between 0.00 and 1,000.00 monetary units to invest. They were split into two teams, taken into different rooms, and the experiment was divided into three phases. In phase one, the information from the financial analyst was provided and a picture was projected onto a screen, as people were told that the person portrayed was the analyst. At that point, each volunteer had to decide how much of his or her initial financial endowment they were willing to invest. In phase two, participants completed a questionnaire about their biological and social traits. Finally, in phase three, all volunteers were asked to take photocopy pictures of their right and left hands. The experiment was conduc-

36

ted using two different treatments. Treatment 1 showed the volunteer’s picture while wearing casual clothes (shorts and a t-shirt), while treatment 2 showed the same volunteer’s picture in a suit and tie. Each session lasted 60 minutes on average. A piece of software developed by Debruine (2004) was used to determine finger length. The experiment was submitted to and approved by the research ethics committee of the Catholic University of Brasília, which is affiliated with the Ministry of Health via Plataforma Brasil, registered before CAAE under No.: 23282613.7.0000.0029. To confirm the relevance of the link between 2D:4D ratio and invested value, we estimated an econometric model (1). We also ran regressions for the other variables provided by participants, such as sex, experiment treatment, age, weight, skin color, height and laterality (whether left- or right-handed). For the purposes of evaluation, the equation below was estimated via Ordinary Least Squares (OLS): Vlr = α0 + β1X1 + β2X2 + β3X3 + β4X4 + β5X5 + β6X6 + β7X7+ β8X8 + μi

(1)

Where: Vlr is the value invested by the individual; X1 = is the right-hand

Where: Vlr is the value invested by the individual; X1 = is the right-hand 2D:4D ratio; X2 = is the experiment treatment (picture 1 or picture 2); X3 = dummy dummy for sex, 1 for men, 0 for women; X4 = age; X5 = weight; X6 = skin color, for sex, 1 for men, 0 for women; X4 = age; X5 = weight; X6 = skin color, 1 for whi1 for white and 0 for all others (non-white); X7 = dummy for laterality, 1 for right te and 0 for all others (non-white); X7 = dummy for laterality, 1 for right handed, 0 handed, 0 for left handed; X8 = height, and μi for an error term. The quantitative for left handed; X8 = height, and μi for an error term. The quantitative variables variables underwent logarithmic transformation, which allows treating the data underwent logarithmic transformation, which allows treating the data as elastias elasticity.city. Because the invested value variable included 0.00 M.U. values, we Because the invested value variable included 0.00 M.U. values, we added a added a constant 1.00 M.U. all invested amounts. amounts. constant 1.00 toM.U. to all invested

2D:4D ratio; X2 = is the experiment treatment (picture 1 or picture 2); X3 =

Table 1: OLS Estimations. Dependent Variable: Table 1 – OLS Estimations. DependentAmount Variable: Log Invested Amount Log Invested (A)

(B)

Constant

5.567466 *** 5.856012 *** (0.188291) (1.641913)

Ln_2D:4D Right Hand

-5.068199 *** -5.217447 ** (1.948204) (2.201367)

TREATMENT 1 or 2

0.567047 *** (0.149502)

Dummy - SEX

-0.066439 (0.192395)

Ln_AGE

-0.48829 (0.420407)

Ln_WEIGHT

0.642241 (0.429023)

Dummy - RACE

-0.359434 (0.256074)

Dummy - LEFT/RIGHT Left/Right Handed HANDED

0.112645 (0.219303)

Ln_HEIGHT

-3.260601 (2.506251)

N L

219 -370.8487

1

Standard Deviation in parentheses

2

** Statistically significant at 5%

3

*** Statistically significant at 1%

219 -362.3675

Column A shows the inferences obtained between the 2D:4D ratio and the invested amount. As expected, the sign for parameter 2D:4D was negative (t statistic = -2.60147, p = 0.0099), as in Garbarino et al. (2001) and Coates et al. (2009). We therefore infer that the lower the 2D:4D ratio, the higher the invested amount. In this case, for every 1% decrease in the 2D:4D ratio, the amount invested by volunteers rose by 5.06%. Column B shows the correlation between the invested amount and the other variables estimated in the model. Variables “sex”, “treatment”, “race” and “laterality” were included as dummies. Out of the analyzed variables, only “2D:4D” and “treatment”

37

were significant. For the “treatment” variable (t statistic = 3.7929, p = 0.0002) the result was positive and significant at 1%, showing that the invested amount does correlate with the picture provided of the analyst. Again, the signal of the 2D:4D was negative (t statistic = -2.37009, p = 0.0187), but now for each 1% decrease in the 2D:4D the amount invested by participants rose by 5.21%.

3. Results We ran the Mann-Whitney (U) test for the right- and left-hand measurements of all participants, and were unable to reject the hypothesis that they are statistically the same (p=0.1093). Because no significant difference existed, although both hands were measured, this study only concerns itself with the right hand’s results. The results are as in Manning et al. (1998) and McIntyre (2006), as women showed a higher 2D:4D ratio than men. Average right-hand measurement for women was 0.945, as opposed to 0.930 for men. To test for statistically significant difference between men’s and women’s measurements, we ran the Mann-Whitney (U) test and the t-test. The Mann-Whitney (U) test showed a significant difference (p=0.0251) and so did the t-test (p= 0,0055). This statistically proves that men’s 2D:4D ratio is smaller than women’s, confirming the results provided in the literature (Manning et al., 1998 and Manning, 2002). Table 2 shows the investment results separately for each group of participants, as well as t-test results. Table 2: Average Invested Values by Variable with t-test Results. Average invested amounts Total (Men and Women Men Women 2D:4D < 1 2D:4D > 1 Men 2D:4D < 1 Men 2D:4D > 1 Women 2D:4D < 1 Women 2D:4D >1

Treatment 1 (SD) 456.41 (286.31) 464.89 (305.04) 449.66 (272.94) 452.88 (285.24) 445.00 (312.69) 453.48 (303.26) 470.20 (395.93) 452.40 (272.93) 420.00 (303.31)

Treatment 2 (SD) 577.16 (240.39) 603.77 (241.12) 559.55 (240.06) 591.37 (234.91) 494.44 (262.15) 608.60 (236.83) 495.00 (417.19) 578.81 (234.74) 433.33 (250.00)

t -3.38765 -2.41574 -2.41427 -3.74621 0.381471 -2.64346 -0.07405 -2.64568 -0.08888

p 0.0008 *** 0.0177 *** 0.0172 *** 0.0002 *** 0.7073 0.0098 *** 0.9438 0.0093 *** 0.9306

Note: SD = Standard Deviation *** p < 0.05

The average value invested by all participants subjected to treatment 1 was 456.41 monetary units (M.U.), as compared to 577.16 M.U. for treatment 2 (tThe average value invested by all participants subjected to treatment 1 -test, T = -3.38765, p = 0.0008). Treatment-2 volunteers invested 10% more on avewas 456.41 monetary units (M.U.), as compared to 577.16 M.U. for treatment 2 rage than those who underwent treatment 1. The results were the same for both (t-test, T =T-3.38765, p= 0.0008). men (t-test, = -2.415739 p= 0.0177)Treatment-2 and women volunteers (t-test, T =invested -2.41426810% p = more 0.0172). on average than those who underwent treatment 1. The results were the same for both men (t-test, T = -2.415739 p = 0.0177) and women (t-test, T = -

38

The image of the analyst in treatment 2 increased the amount invested by participants, corroborating the findings of Rosenberg et al. (1986; 1991). Based on the literature, we infer that participants with 2D:4D ratios greater than 1 are risk averse and those with 2D:4D ratios lower than 1 are less averse (Sapienza et al., 2009; Garbarino et al., 2011; Coates et al., 2009 and Apicella et. al., 2008). Therefore, we find that 100% of risk-averse participants, that is, with 2D:4D ratios greater than 1 invested less than 50% of their initial endowment, regardless of the treatment. In addition, the average invested value was not significantly different between the treatments, at 445.00 M.U. for treatment 1 and 494.44 M.U. for treatment 2 (t-test, T = 0.381471, p = 0.7073), indicating that, for individuals with 2D:4D ratios greater than 1, that is, less exposed to testosterone, aversion to risk prevails regardless of the analyst’s appearance. Volunteers with 2D:4D ratios lower than 1, that is, less risk-averse, that underwent treatment 2 invested 23.5% more on average than those in treatment 1, which shows increased risk appetite (t-test, T = -3.76621, p = 0.0002). To check for different behaviors in men and women, we ran the t-test for men with 2D:4D ratios lower than 1 and women with ratios greater than 1 for treatments 1 and 2. In treatment 1, less risk-averse men invested almost 8% more than risk-averse women, but the difference was not statistically significant (t-test, T = 0.233701, p = 0.8163). For treatment 2, however, where higher investments were expected as a result of the analyst’s portrayal, less risk-averse men invested 40% more than risk-averse men, at a significance level of 5% (t-test, T = 2.000747, p = 0.050), a finding similar to that from the experiment in Apicella et al. (2008).

4. Discussion The average value invested by all participants under treatment 1 (analyst in t-shirt and shorts) was 456.42 M.U., as opposed to 577.17 M.U. for treatment 2 (analyst in a suit and tie), with significant difference (t-test, T = -3.38765, p = 0.0008) and suggesting that the image viewed in each of the experiment’s treatments as determinant to the values invested by volunteers. It was soon found that the image of a person that conveys greater trustworthiness increases individuals’ credibility, as Rosenberg (1986; 1991), Druckman (2003) and Todorov et al. (2005) point out. Volunteers with a 2D:4D ratio lower than 1, that is, more exposed to testosterone, that underwent treatment 2 invested on average 23.5% more than participants in treatment 1. This suggests that a greater exposure to prenatal testosterone causes individuals to show riskier behavior (Apicella et al. 2008). However, men who were more exposed to testosterone invested 40% more than less exposed women in treatment 2. In this case, the effect of greater exposure to testosterone appears to have magnified the difference in men’s risky behavior. We therefore confirm the findings of Apicella et al. (2008), Sapienza et

39

al. (2009), Coates et al. (2009) and Garbarino et al. (2011), according to which men show riskier behavior than women. The experiment’s main contribution is that participants with 2D:4D ratios greater than 1, that is, less exposed to testosterone, do not change opinions widely, regardless of the image provided, and retain their risk-averse convictions. The average amount of their investments was not significantly different between the two treatments, at 445.00 M.U. under treatment 1 and 494.44 M.U. in treatment 2 (t-test, T = 0.381471, p = 0.7073), indicating that aversion to risk stands regardless of the image provided. We can therefore infer that, for people less exposed to testosterone, the image is not interpreted in the same way as for those more exposed to the hormone, and does not interfere with their risky financial decisions. Translating the effect into the world of organizations, according to Baldissera (2008), it is the concept-image that matters. A concept-image’s distinguishing trait is the “appreciation” effect, a concept that a group assigns to a certain organization. It is therefore crucially important for business firms to concern themselves with the image they may be conveying to the market. However, as our conclusions suggest that not all individuals respond to this effect alone, we are working on additional experiments to identify other variables that may be used as proxies to identify lesser or greater exposure to testosterone. Organizations may soon be able to include into their customer identification questionnaires certain key questions to enable identifying greater or lesser exposure to testosterone and thereby offer a distinctive portfolio according to their risky behavior, securing an extremely important competitive advantage. Given our results, we may infer that exposure to the hormone testosterone as measured by the 2D:4D ratio is an important factor in people’s risky behavior. Therefore, we find that exposure to the hormone testosterone as measured by the 2D:4D ratio is an important predictor for the risky behavior of financial investors.

Authors

40

Eduardo Borges da Silva Bachelor of Economics, Graduate of Mathematics and Statistics, Master and Doctor of Economics degrees from the Universidade Católica de Brasília (UCB). Is a Matrix Consultant at Caixa Econômica Federal, active in the National Risk Models Monitoring Area (GEMOR/SUCOI). Academically, he is a Risk Management Professor at IBMEC – DF with the CBA – Business Administration Program. The author recognizes financial support from CNPq. Email: [email protected]

Benjamin Miranda Tabak Bachelor, Master and Doctor of Economics. A Legislative Consultant with the Brazilian Federal Senate in the Economics area (Economic Policy and Financial System). Academically, he is a lecturing researcher with the Master of Law program of Universidade Católica de Brasília, in the Economic Analysis of Law area, and with the same university’s Doctor of Economics program. CNPq scholar rated with 1C for productivity. The author recognizes financial support from CNPq.

References

Email: [email protected]

ALLAIS, M.(1953) "Le comportement de l’homme rationnel devant le risque: critique des postulats et axiomes de l’école Américaine". Econometrica, 21, n. 4 p.503–546. APICELLA C, DREBER A, CAMPBELL B, GRAY P, HOFFMAN M, LITTLE A (2008) Testosterone and financial risk preferences. Evolution and Human Behavior 29: 384–390. Baker J, SHAO Chris Y, WAGNER Judy A (2004) The effects of appropriateness of service contact personnel dress on customer expectations of service quality and purchase intention: The moderating influences of involvement and gender. Journal of Business Research 57: 1164 – 1176. BALDISSERA, R. (2008) Significação e comunicação na construção da imagem-conceito. Revista Fronteiras – Estudos Midiáticos. São Leopoldo: Ed. da Unisinos, p. 193-200. BROWN WM, FINN CJ, BREEDLOVE SM (2002) Sexual dimorphism in digit-length ratios of laboratory mice. The Anatomical Record 267: 231–234. BROWN WM, HINES M, FANE BA, BREEDLOVE SM (2002) Masculinized finger length patterns in human males and females with congenital adrenal hyperplasia. Hormones and Behavior 42: 380–386.

References

41

COATES JM, GURNELL M, RUSTICHINI A. (2009) Second-to-forth digit ratio predicts success among high-frequency financial traders. PNAS 106: 623–628. DA SILVA S, MOREIRA B, DA COSTA JR N (2014) 2D:4D Digit Ratio Predicts Delay of Gratification in preschoolers. PLOS ONE 9: e114394. DEBRUINE LM (2004) Autometric software for measurement of 2d4d ratios Disponível em: . Acesso em 10 dez. 2013. DIXSON AF (1998) Primate sexuality: Comparative studies of the prosimians, monkeys, apes and human beings. New York: Oxford University Press. DRUCKMAN NJ (2003) The power of television images: The first Kennedy-Nixon debate revisited. Journal of Politics 65: 559-571. ELLSBERG, D. (1961)Risk, ambiguity and the Savage axioms. Quarterly Journal of Economics, v.75, p. 643–669. GARBARINO E, SLONIM R, SYDNOR J (2011) Digit Ratios (2D:4D) as predictors of risky decision making for both sexes. Journal of Risk and Uncertainty 42: 1-26. GNEEZY U, POTTERS Jan (1997) An experiment on risk taking and evaluation periods. The Quarterly Journal of Economics 112: 631-645. KAHNEMAN D, TVERSKY A. (1979) Prospect Theory: an analysis of decision under risk. Econometrica 47: 263-292. LUTCHMAYA S, BARON-COHEN S, RAGGATT P, KNICKMEYER R, MANNING JT (2004) 2nd to 4th digit ratios, fetal testosterone and estradiol. Early Human Development 77: 23–28. MANNING JT, SCUTT D, WILSON J, LEWIS-JONES DI (1998) The ratio of 2nd to 4th digit length: A predictor of sperm numbers and concentrations of testosterone, luteinizing hormone and oestrogen. Human Reproduction 13: 3000–3004. MANNING JM (2002) Digit Ratio: A Pointer to Fertility, Behavior and Health. New Jersey: Rutgers University Press. MANNING, J.T., FINK, B.(2011). Digit ratio (2D:4D) and aggregate personality scores across nations: Data from the BBC internet study. Personality and Individual Differences, 51: 387–391. MARUANI A, LÉGER J, GIRAUDEAU B, NAOURI M, LE BIDRE E, SAMIMI M, DELAGE M (2012) Effect of physician dress style on patient confidence. Journal of the European Academy of Dermatology and Venereology 27: 333–337.

References

42

MCINTYRE MH (2006) The use of digit ratios as markers for perinatal androgen action. Reproductive Biology and Endocrinology 4: 10. MILLET K., (2011) An interactionist perspective on the relation between 2D:4D and behavior: An overview of (moderated) relationships between 2D:4D and economic decision making. Personality and Individual Differences 51: 397–401. ÖKTEN A, KALYONCU M, YARIS N (2002) The ratio of second- and fourth-digit lengths and congenital adrenal hyperplasia due to 21-hydroxylase deficiency. Early Human Development 70: 47–54. PAULINS Ann V (2005) An analysis of customer service quality to college students as influenced by customer appearance through dress during the in-store shopping process. Journal of Retailing and Consumer Services 12: 345–355. PETERS M, MACKENZIE K, BRYDEN P (2002) Finger length and distal finger extent patterns in humans. American Journal of Physical Anthropology 117: 209–217. REHMAN SU, NIETERT PJ, COPE DW, KILPATRICK Anne O (2005) What to wear today? Effect of doctor’s attire on the trust and confidence of patients. The American Journal of Medicine 118: 1279-1286. ROSENBERG SW, BOHAN Lisa, MCCARFFERTY Patrick, HARRIS Kevin (1986) The image and the vote, the effect of candidate presentation on voter preference. American Journal of political Science 30: 108-127. ROSENBERG SW, KAHN Shulamit, TRAN Thuy (1991) Creating a political image, shaping appearance and manipulation the vote. Political Behaviour 13: 345-367. SAPIENZA P, ZINGALES L, MAESTRIPIERI D (2009) Gender differences in financial risk aversion and career choices are affected by testosterone. PNAS 106: 15268–15273. SIMON, H. A. (1955) A behavioral model of rational choice. The quarterly journal of economics, 69, n.1, p. 99-118. THALER RH, KAHNEMAN D, KNETSCH JL (1991) anomalies: the Endowment Effect, Loss Aversion, and Status Quo Bias. Journal of Economic Perspectives 5: 193-206. TODOROV A, MANDISODZA AN, GOREN A, HALL CC (2005) Inferences of competence from faces predict election outcomes. Science 308: 1623-1626. VON NEUMANN J, MORGENSTERN O (1994) theory of games and economic behavior. Princeton: Princeton University Press.

References

43

VORACEK M, DRESSLER SG, AND LOIBL LM (2008) The contributions of Hans-Dieter Rösler: Pioneer of digit ratio (2D:4D) research. Psychological Reports 103: 899–916. WINGFIELD JC, HEGNER RE, DUFTY AM, BALL GF (1990) The challenge hypothesis: Theoretical implications for patterns of testosterone secretion, mating systems, breeding strategies. American Naturalist 136: 829 - 846.

44

Capital Planning in Good and Bad Times

Lourenço Miranda

45

Abstrat The history has demonstrated that is much more difficult to manage risks in good times than bad times. In good times, it is much more difficult to convince the Board and the shareholders that you should keep enough capital to survive a downturn (e.g. Lehman Brothers). Of course, you will not keep capital levels as if you were managing the Bank to face a Great Storm at any time, it would be too costly; the firm would become non-competitive and out-of-market. You do not manage a business as if it would have to face a storm every month but when it hits you must be prepared and know what to do. The fact is that risk exposure increases or decreases following economic cycles, therefore capital management strategy as stated in the Bank’s Capital Plan should address cyclicality of risk. Keywords: Pro-cyclicality, Stress Testing, Subprime.

1. Introduction Usually firms hold a capital buffer to absorb the increasing exposure during an economic downturn or manage available capital (define) and risk in a dynamic way according to and in harmony with the cycles – this mechanism is called pro-cyclicality of capital management. The more agile the ability to balance capital demand (risk exposure) and available capital the smaller the buffer of capital needed to manage the firm. On the same token, stress testing will help the Banks to be prepared to unexpected losses. As available capital will always decline after one big loss hits the Bank, the faster the Bank is able to rebalance capital after the event the lower the capital buffer it will need to retain unrewarded (not at the disposal of the business to be compensated).Taking on too much risk can have serious consequences—and in modern history, it’s often the taxpayers who pay the bills.

2. Washington Mutual Banking Let’s talk about the sad fate of Washington Mutual Bank, which at one time was the largest savings and loan association in the United States. Its origins were noble. Following a fire that devastated much of Seattle, Washington in 1889, the bank was founded as the Washington National Building Loan and Investment Association. It was dedicated to helping Seattle rebuild. Renamed Washington Mutual Savings Bank in 1917, as the years passed it grew, and during the Great Depres-

46

sion the tough little bank survived two runs by frantic depositors. The bank then became known for a number of innovative banking ideas, including a penny-deposit program for children, cash machines, and telephone banking. During the 1960s, it expanded throughout the state, and, after a series of mergers and acquisitions, throughout the West and the rest of the nation in the 1990s. An aggressive move into the sub-prime mortgage market in the 2000s brought unprecedented growth as well as enormous risk. At its peak in the summer of 2008, Washington Mutual Bank, known colloquially as WaMu, had total assets of $307 billion, with 2,239 retail branch offices operating in fifteen states, and 43,198 employees. On September 15, 2008, the holding company, Washington Mutual Inc., received a credit rating agency downgrade. From that date through September 24, 2008, WaMu experienced a bank run, whereby nervous customers withdrew $16.7 billion in deposits over nine days. This amounted to nine percent of the deposits it held. On Thursday, September 25, 2008, the United States Office of Thrift Supervision, a division of the US Treasury Department, closed the bank. The next day, Washington Mutual, Inc. filed for Chapter 11 voluntary bankruptcy in Delaware, where it was incorporated. With respect to total assets under management, Washington Mutual Bank’s closure and receivership is the largest bank failure in American financial history. Before the receivership action, it was the sixth-largest bank in the United States. The story of the bank’s failure is long and complex. It occurred during a tumultuous period in American financial history. In just a few weeks in the late summer of 2008, Merrill Lynch was sold, Lehman Bros. failed, Freddie Mac and Fannie Mae were placed in receivership, and AIG received a huge bailout to prevent a collapse of global markets. WaMu faced a growing mass of failed mortgage loans, creating an exposure that was impossible to quantify at the time, but which analysts believed ran to billions of dollars. Some WaMu supporters insist that if federal regulators had waited for just six business days before shuttering the bank, WaMu could have been saved by two actions: the government’s $700-billion Troubled Asset Relief Program, and by an increase in bank deposit insurance limits from $100,000 to $250,000, a measure that helped quell panic withdrawals at many other banks. Those changes may have cooled the bank run that struck WaMu. However, it is all past now. The storm came and washed away Washington Mutual and many other financial institutions that had inadequate capital plans.

3. Royal Bank of Scotland Another case that aggressive risk taking without measuring potential future consequences can be devastating. In the autumn of 2007, the Royal Bank of Scotland amazed investors by announcing an operating profit of £10.3 billion (about

47

$16 billion), the biggest ever for a Scottish company. The achievement was hailed as Scotland’s own economic miracle and the inspiring story of a regional bank that became the fifth largest in the world by borrowing billions to voraciously gobble up no fewer than twenty-six other companies in the space of seven years. Just a year later, the The Guardian reported that commentators were concerned that RBS did not have sufficiently robust capital reserves to cope with the effects of the looming credit crunch. The bank reported that its core tier one capital ratio—the key measure of a bank’s strength—was 5.7 percent. David Buik of BGC Partners said this was “unacceptable,” and predicted that RBS may need to raise another £4 billion in fresh capital. By the spring of 2009, RBS set a new record, this time by posting a loss of £28 billion (about $43 billion), the biggest in British corporate history. It was a truly mind-boggling reversal of fortunes. Analysts charged that the demise of RBS could be summed up in just one word: greed. Under the stewardship of its chief executive, Sir Fred Goodwin, RBS had bought every bank and financial institution it could get its hands on, regardless of whether it made economic sense. As The The Telegraph reported, Sir Fred (as he was known in the cheeky British press) became obsessed with his quest to make RBS a titan of world banking, and had such an overbearing personality that none of his staff had the backbone to stop him, even when some of them began to have concerns that the bank was over-stretching itself.

4. ABN Amro The tipping point in Sir Fred’s reckless pursuit of risk came in October 2007, when the RBS took over ABN Amro, the biggest bank in the Netherlands, in a deal costing £49 billion. This was at a time when other bankers were sensing a heightened risk environment. Northern Rock, a UK bank formed in 1965, had suffered the first run on a British bank for more than a century, exposing the vulnerability of UK banks to the worldwide credit crunch. Heedless of the gathering storm clouds, and after joining forces with Spain’s Banco Santander and Belgium’s Fortis, Sir Fred went ahead with what proved to be a grossly inflated offer for ABN Amro. The deal—described by then-prime minister Gordon Brown as “irresponsible”—left RBS dangerously overcommitted, and as the worldwide banking crisis widened, it became clear that many of the companies RBS had bought up, including several US-based firms, were seriously exposed to the subprime mortgage crisis, meaning they were worth only a fraction of what RBS had paid for them. As quoted in The The Telegraph at the time, David Buik, partner at BGC

48

Partners, said the demise of RBS was due to “a degree of arrogance the like of which you will never see again in your lifetime. Fred Goodwin is a megalomaniac. RBS never had a chance to digest anything they bought and so they’ve never delivered shareholder value. It’s a combination of relentless greed and an inability to deliver shareholder value. “They were buying companies when their share price was at its peak, rather than when shares were at rock bottom, and they clearly got involved with things they just didn’t understand. “RBS took over ABN Amro because the mindset was they had to stop Barclays getting their hands on it at any cost, and consequently they paid way over the odds. “There were people in that boardroom during the ABN Amro takeover who must have thought ‘this is madness,’ but no-one was prepared to stand up to Sir Fred. I know people who worked for him, and it was a case of ‘yes Sir, no Sir, three bags full, Sir’.” Because for years it had seemed that he could do no wrong, Sir Fred had wielded total power in the boardroom. In 2000, Sir Fred had been named businessman of the year by Forbes magazine for what it described as a “brilliantly strategized hostile takeover” of National Westminster Bank (known popularly as NatWest). Following the acquisition of NatWest, another bank might have taken a breather, prudently paying off some of their debts before launching any other major acquisitions. For Sir Fred, the ultimate risk addict, it was only the beginning. Acquisitions of Royal Insurance, Churchill Insurance and Charter One were among the major deals that followed as RBS steadily climbed into the ranks of the biggest banks. But even before the takeover of ABN Amro, the true health of RBS had been the subject of debate. In 2005, John-Paul Crutchley, an analyst at Merrill Lynch, had published research entitled “What Went Wrong With RBS?” For several years its shares had underperformed when compared with the rest of banking sector. Still the expansion continued. By the end of 2006 RBS shareholders were becoming so concerned about the relentless acquisitions of other companies that Sir Fred announced there would be no more major deals. Then he heard that Barclays was negotiating to buy ABN Amro, and, despite clear signs that the worldwide money markets were beginning to dry up, Sir Fred went head-to-head with Barclays to snatch away the glittering prize. To shareholders nervous that RBS would pay too much, he said not to worry: “The key to good deal-making is knowing when to walk away.” He was either lying to himself or to his shareholders. Only six months after the ABN Amro deal, RBS announced it needed to shore up its finances with a £12 billion infusion from its shareholders. The bank had set yet another record, this time for the biggest rights issue in Europe’s history.

49

5. Banking Crisis

Author

Then came the banking crisis in October 2008, triggered by the fall of Lehman Brothers in the United States. Confidence in British banks slipped because of their exposure to the US subprime mortgage crisis. As it became clear that many of the assets against which RBS and others had borrowed money were now worth only a fraction of their previous value, share prices fell. Weakened by its profligate borrowing, RBS had no option but to accept a British government bailout. As RBS threatened to drag down the entire British banking system, with banking shares plummeting across the board, prime minister Gordon Brown attacked the “irresponsible risks” taken by RBS, adding that he was “angry” at the reckless dealings of the Scottish bank. On October 11, 2008, Sir Fred Goodwin announced his resignation as chief executive of RBS and took an early retirement, effective after January 31, 2009. On February 26, 2009 RBS announced that its 2008 loss totaled £24.1 billion, the largest annual loss in UK corporate history. By that time, Sir Fred had left the building, taking with him his pension worth £16 million. By the way, in case you happen meet Fred Goodwin—say, if your yacht passes by his in the sunny Mediterranean—you should not address him as “Sir Goodwin.” His knighthood, awarded in 2004 for “services to banking,” was annulled on February 1, 2012. He’s back to being plain old Fred Goodwin. It is indeed way more difficult to manage risks in good times than it is in bad times. At least in bad senior managers will listen.

Lourenço Miranda Prof. Dr. Lourenço Miranda is currently Managing Director of AIG for Economic Capital and Stress Test modeling. Before that, he was with US Bancorp in Minneapolis as head of the Quant Unit. Previously he was Senior Risk Officer of the Advisory Access to Finance team of the International Finance Corp of the World Bank Group. Before that he was with ABN Amro and Santander Banks. He has over 20 years of professional experience. Also, Dr. Miranda is a published author with various papers in peer-reviewed journal. In the academic arena, he is was visiting Professor of the University of Minnesota and many other academic centers in the world.

Revista Tecnologia de Crédito Online Acesse informação de ponta, onde você estiver.

Sua revista Tecnologia de Crédito é publicada exclusivamente na web. É mais um passo da Serasa Experian para a redução do impacto de suas atividades sobre o meio ambiente, além de um alinhamento à tendência global da universalização de conteúdos. Acesse e mantenha-se bem informado sobre o mercado de crédito. serasaexperian.com.br/revista-tecnologia-de-credito

Proteja A Serasa Experian oferece soluções sob medida para o seu negócio. São produtos e serviços exclusivos, que vão desde consultas de crédito até ferramentas e softwares para gerir melhor sua empresa. Tudo muito prático, rápido e rentável. E o melhor: cabe no seu orçamento. Entre em contato e prepare-se para se destacar da concorrência.

Para saber mais, acesse serasaexperian.com.br ou ligue para 3004 7728

SoluçõeS eSPeCífiCaS Para Cada etaPa do Seu negóCio

 

Busque novos clientes

 

Venda com segurança

Monitore empresas, clientes e sócios

Cobre devedores

Compre só os dados de que precisa