This study follows a three–pronged approach to investigate the effects of the global financial crisis on the South African retail credit market (using Woolworths as subject). These three prongs, or areas, include a literature study, step–by–step credit scoring guide and an application of this guide in an empirical study. To achieve this goal, credit scoring was selected as the quantitative tool to illustrate these effects. Two different periods were chosen to supply a snapshot of the retail credit industry, namely the retail credit situation before and during the global financial crisis. To correctly define and understand the mechanics affecting South Africa's retail credit industry, a literature review was conducted to investigate the global financial crisis, the South African retail credit market and credit scoring itself. The literature investigation explains the global financial crisis and identifies some of the primary drivers behind it. These drivers included the US housing bubble, the introduction of subprime loans and the securitisation of these loans (mortgage backed securities). The study found that these drivers, especially the securitisation of subprime loans, were the vehicle used to enable the crisis to spread globally. The ultimate goal of the study was to provide the individual, and companies, with an understanding of the global financial crisis' effects on the consumer specifically through their credit worthiness and retail credit behaviour. Through the use of credit scoring, the study found that at least one retailer (Woolworths) in the retail industry was affected. Woolworths placed a stronger emphasis on reducing their credit exposure whilst consumers were steadily increasing their facility utilisation. / Thesis (M.Com. (Risk management))--North-West University, Potchefstroom Campus, 2012.

Submitted by Aelson Maciera (aelsoncm@terra.com.br) on 2017-08-16T20:00:07Z No. of bitstreams: 1 DissDMBS.pdf: 821487 bytes, checksum: 497fc9b102478d03042a1c3d10a45c19 (MD5) / Approved for entry into archive by Ronildo Prado (bco.producao.intelectual@gmail.com) on 2018-01-29T18:10:09Z (GMT) No. of bitstreams: 1 DissDMBS.pdf: 821487 bytes, checksum: 497fc9b102478d03042a1c3d10a45c19 (MD5) / Approved for entry into archive by Ronildo Prado (bco.producao.intelectual@gmail.com) on 2018-01-29T18:10:17Z (GMT) No. of bitstreams: 1 DissDMBS.pdf: 821487 bytes, checksum: 497fc9b102478d03042a1c3d10a45c19 (MD5) / Made available in DSpace on 2018-01-29T18:14:39Z (GMT). No. of bitstreams: 1 DissDMBS.pdf: 821487 bytes, checksum: 497fc9b102478d03042a1c3d10a45c19 (MD5) Previous issue date: 2017-06-13 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) / Regression models for binary response variables are very common in several areas of knowledge. The most used model in these situations is the logistic regression model, which assumes that the logit of the probability of a certain event is a linear function of the predictors variables. When this assumption is not reasonable, it is common to make some changes in the model, such as: transformation of predictor variables and/or add quadratic or cubic terms to the model. The problem with this approach is that it hinders parameter interpretation, and in some areas it is fundamental to interpret the parameters. Thus, a common approach is to categorize the quantitative covariates. This work aims to propose two new classes of categorization methods for continuous variables in binary regression models. The first class of methods is univariate and seeks to maximize the association between the response variable and the categorized covariate using measures of association for qualitative variables. The second class of methods is multivariate and incorporates the predictor variables correlation structure through the joint categorization of all covariates. To evaluate the performance, we applied the proposed methods and four existing categorization methods in 3 credit scoring databases and in two simulated cenarios. The results in the real databases suggest that the proposed univariate class of categorization methods performs better than the existing methods when we compare the predictive power of the logistic regression model. The results in the simulated databases suggest that both proposed classes perform better than the existing methods. Regarding computational performance, the multivariate method is inferior and the univariate method is superior to the existing methods. / Modelos de regressão para variáveis resposta binárias são muito comuns em diversas áreas do conhecimento. O modelo mais utilizado nessas situações é o modelo de regressão logística, que assume que o logito da probabilidade de ocorrência de um dos valores da variável resposta é uma função linear das variáveis preditoras. Quando essa suposição não é razoável, algumas possíveis alternativas são: realizar transformação das variáveis preditoras e/ou inserir termos quadráticos ou cúbicos no modelo. O problema dessa abordagem é que ela dificulta bastante a interpretação dos parâmetros do modelo e, em algumas áreas, é fundamental que eles sejam interpretáveis. Assim, uma abordagem muitas vezes utilizada é a categorização das variáveis preditoras quantitativas do modelo. Sendo assim, este trabalho tem como objetivo propor duas novas classes de métodos de categorização de variáveis contínuas em modelos de regressão para variáveis resposta binárias. A primeira classe de métodos é univariada e busca maximizar a associação entre a variável resposta e a covariável categorizada utilizando medidas de associação para variáveis qualitativas. Já a classe de métodos multivariada tenta incorporar a estrutura de dependência entre as covariáveis do modelo através da categorização conjunta de todas as variáveis preditoras. Para avaliar o desempenho, aplicamos as classes de métodos propostas e quatro métodos de categorização existentes em 3 bases de dados relacionadas à área de risco de crédito e a dois cenários de dados simulados. Os resultados nas bases reais sugerem que a classe univariada proposta têm um desempenho superior aos métodos existentes quando comparamos o poder preditivo do modelo de regressão logística. Já os resultados nas bases de dados simuladas sugerem que ambas as classes propostas possuem um desempenho superior aos métodos existentes. Em relação ao desempenho computacional, o método multivariado mostrou-se inferior e o univariado é superior aos métodos existentes.

Le credit scoring est généralement considéré comme une méthode d’évaluation du niveau du risque associé à un dossier de crédit potentiel. Cette méthode implique l'utilisation de différentes techniques statistiques pour aboutir à un modèle de scoring basé sur les caractéristiques du client.Le modèle de scoring estime le risque de crédit en prévoyant la solvabilité du demandeur de crédit. Les institutions financières utilisent ce modèle pour estimer la probabilité de défaut qui va être utilisée pour affecter chaque client à la catégorie qui lui correspond le mieux: bon payeur ou mauvais payeur. Les seules données disponibles pour construire le modèle de scoring sont les dossiers acceptés dont la variable à prédire est connue. Ce modèle ne tient pas compte des demandeurs de crédit rejetés dès le départ ce qui implique qu'on ne pourra pas estimer leurs probabilités de défaut, ce qui engendre un biais de sélection causé par la non-représentativité de l'échantillon. Nous essayons dans ce travail en utilisant l'inférence des refusés de remédier à ce biais, par la réintégration des dossiers refusés dans le processus d'octroi de crédit. Nous utilisons et comparons différentes méthodes de traitement des refusés classiques et semi supervisées, nous adaptons certaines à notre problème et montrons sur un jeu de données réel, en utilisant les courbes ROC confirmé par simulation, que les méthodes semi-supervisé donnent de bons résultats qui sont meilleurs que ceux des méthodes classiques. / Credit scoring is generally considered as a method of evaluation of a risk associated with a potential loan applicant. This method involves the use of different statistical techniques to determine a scoring model. Like any statistical model, scoring model is based on historical data to help predict the creditworthiness of applicants. Financial institutions use this model to assign each applicant to the appropriate category : Good payer or Bad payer. The only data used to build the scoring model are related to the accepted applicants in which the predicted variable is known. The method has the drawback of not estimating the probability of default for refused applicants which means that the results are biased when the model is build on only the accepted data set. We try, in this work using the reject inference, to solve the problem of selection bias, by reintegrate reject applicants in the process of granting credit. We use and compare different methods of reject inference, classical methods and semi supervised methods, we adapt some of them to our problem and show, on a real dataset, using ROC curves, that the semi-supervised methods give good results and are better than classical methods. We confirmed our results by simulation.

Made available in DSpace on 2016-06-02T20:06:10Z (GMT). No. of bitstreams: 1 6034.pdf: 941825 bytes, checksum: 6d06b85571d5cab86cee2ed1c1d699da (MD5) Previous issue date: 2014-03-28 / Universidade Federal de Sao Carlos / Granting credit to an applicant is a decision made in a context of uncertainty. At the moment the lender decides to grant a loan or credit sale there is always the possibility of loss, and, if it is associated with a probability, the decision to grant or not credit will be more reliable. In order to aid the decision to accept or not the request for applicants are used the credit scoring models, which estimate the probability of loss associated with granting credit. But one of the problems involving these models is that only information about the applicants accepted are used, which causes a sampling bias, because the rejected applicants are discarded. With the aim to solve this problem it can use rejected inference, which are considered individuals who have had credit application rejected. However, only considering rejected inference and one method of modeling data, usually, is not sufficient to get satisfactory predictive measures, and thus, were used combined results of three methods, logistic regression, analysis probit and decision tree. The purpose of this combination were to increase the predictive perfomance and the metrics used were sensitivity, specificity , positive predictive value, negative predictive value and accuracy. Through the application in data sets we concluded that the use of the combined results increased the predictive performance, specially regarding to sensitivity. / A concessão de crédito e uma decisão a ser tomada num contexto de incertezas. No momento em que o credor decide conceder um empréstimo, realizar um financiamento ou venda a prazo sempre existe a possibilidade de perda, e, se for atribuída uma probabilidade a esta perda, a decisão de conceder ou não credito será mais confiável. Com o objetivo de auxiliar a tomada de decisão em relação ao pedido de credito dos solicitantes são utilizados os modelos credit scoring, os quais estimam a probabilidade de perda associada a concessão de credito. Um dos problemas envolvendo estes modelos e que somente informações a respeito dos proponentes aceitos são utilizadas, o que causa um viés amostral, pois, os solicitantes recusados são descartados no processo de modelagem. Com intuito de solucionar este problema tem-se a inferência dos rejeitados, em que são considerados os indívíduos que tiveram pedido de credito rejeitado. No entanto, considerar a inferência dos rejeitados e o uso de somente um método de modelagem de dados, muitas vezes, não e suficiente para que se tenha medidas preditivas satisfatórias. Desta forma, foram utilizados resultados combinados de três metodologias, regressão logística, probit e árvore de decisão/classificação concomitantemente a utilização dos métodos de inferência dos rejeitados que incluem o uso de variável latente, reclassificação, parcelamento e ponderação. O objetivo dessa combinação foi aumentar a capacidade preditiva e as métricas utilizadas foram a sensibilidade, especificidade, valor preditivo positivo, valor preditivo negativo e acurácia. Através da aplicação em conjuntos de dados concluiu-se que a utilização dos resultados combinados aumentou a capacidade preditiva, principalmente, em relação a sensibilidade.

A presente tese objetivou investigar a contribuição de variáveis e escalas psicológicas sugeridas pela literatura de Psicologia Econômica, a fim de predizer o risco de crédito de pessoas físicas. Nesse sentido, através das técnicas de regressão logística, e seguindo todas as etapas para desenvolvimento de modelos de credit scoring, foram construídos modelos de application scoring para pessoas físicas com variáveis sociodemográficas e situacionais, comumente utilizadas nos modelos tradicionais, mais a inclusão de variáveis comportamentais e escalas psicológicas, tais como: variáveis de comparação social, variáveis relacionadas com educação financeira, variáveis de comportamento de consumo, proxies de autocontrole e horizonte temporal, escala do significado do dinheiro (ESD), escala de autoeficácia, escala de lócus de controle, escala de otimismo, escala de autoestima e escala de comprador compulsivo. Os resultados foram contundentes e direcionaram para uma significativa contribuição de algumas dessas variáveis em predizer o risco de crédito dos indivíduos. As variáveis oriundas da ESD mostraram que as dimensões negativas relacionadas com o dinheiro estão mais associadas a indivíduos com problemas com dívidas. Também foi possível constatar que indivíduos com altos escores na escala de autoeficácia, provavelmente indicando um maior grau de otimismo e excesso de confiança, estão mais associados ao grupo de mau pagador. Notou-se ainda que compradores classificados como compulsivos possui maior probabilidade de se encontrar no grupo de mau crédito. Indivíduos que consideram presentear crianças e amigos em datas comemorativas como uma necessidade, mesmo que muitas pessoas considerem um luxo, possuem maior chance de se encontrarem no grupo de mau crédito. Problemas de autocontrole identificados por indivíduos que bebem em média mais de quatro copos de bebida alcoólica no dia ou são fumantes, mostraram-se importantes para identificar tendências ao endividamento. A partir desses achados acredita-se que a presente tese avançou no entendimento do risco de crédito das pessoas físicas, de forma a suscitar variáveis que podem aumentar a precisão da previsão dos modelos de credit scoring, tendo como uma das implicações imediatas a consideração de algumas das variáveis significativas como uma pergunta no formulário cadastral para novos clientes, tais como: Você acha que presentear amigos em datas comemorativas é uma necessidade ou luxo? Você acha que presentear crianças em datas comemorativas é uma necessidade ou luxo? Na média, você bebe mais de 4 copos de bebida alcoólica no dia? Você fuma cigarros? As implicações dos resultados também podem ser discutidas no âmbito dos modelos de behavioral scoring e modelos de credit scoring para pessoas jurídicas. / This works aimed to investigate the contribution of variables and psychological scales, suggested by the literature of Economic Psychology, in order to predict the credit risk of individuals. Accordingly, through the techniques of logistic regression, and following all the steps for developing credit scoring models, application scoring models were built for individuals with socio demographic and situational variables, commonly used in traditional models, further the inclusion of behavioral variables and psychological scales, such as: variables of social comparison, variables related to financial education, variables in consumption behavior, proxies of self-control and temporal horizon, meaning of money scale (MMS), scale of self efficacy, locus of control scale, scale of optimism, scale of self-esteem and scale of compulsive buyer. The results were blunt, and directed a significant contribution to some of these variables in predicting the credit risk of individuals. The variables derived from the MMS showed that the negative dimensions related to money are more associated to individuals with debt problems. It was also noted that individuals with high scores on selfefficacy scale, probably indicating a higher degree of optimism and overconfidence, are the group most associated with bad credit. It was noted also that buyers classified as compulsive ones are more likely to find in the group of bad credit. Individuals who consider gifting children and friends on commemorative dates as a necessity, even though many people consider a luxury, have more chance in being found in the group of bad credit. Self-control problems, identified by individuals who drink more than four glasses of alcohol a day, or are smokers, were important to identify indebtedness trends. From these findings it is believed that this works has advanced the understanding of the credit risk of individuals, giving rise to variables that may increase the forecast accuracy of credit scoring models, having as one of the immediate implications, considering of some of the significant variables as one of the questions about the individual when he fills the new application form, such as: Do you think gifting friends in commemorative dates is a necessity or luxury? Do you think gifting children in commemorative dates is a necessity or luxury? On average, you drink more than four glasses of alcohol a day? Do you smoke cigarettes? The implications of these results can also be discussed in the context of behavioral scoring models and credit scoring models for corporations.

The Bayesian approach to logistic regression modelling for credit scoring is useful when there are data quantity issues. Data quantity issues might occur when a bank is opening in a new location or there is change in the scoring procedure. Making use of prior information (available from the coefficients estimated on other data sets, or expert knowledge about the coefficients) a Bayesian approach is proposed to improve the credit scoring models. To achieve this, a data set is split into two sets, “old” data and “new” data. Priors are obtained from a model fitted on the “old” data. This model is assumed to be a scoring model used by a financial institution in the current location. The financial institution is then assumed to expand into a new economic location where there is limited data. The priors from the model on the “old” data are then combined in a Bayesian model with the “new” data to obtain a model which represents all the available information. The predictive performance of this Bayesian model is compared to a model which does not make use of any prior information. It is found that the use of relevant prior information improves the predictive performance when the size of the “new” data is small. As the size of the “new” data increases, the importance of including prior information decreases

This thesis provides a comprehensive ranking of 11 Czech statistical and 4 foreign credit scoring models. The ranking is based on the predictive performance of individual models, as measured by the area under curve, evaluated on a randomly sampled set of 250 training and validation samples. After establishing a baseline comparison, 3 avenues of estimation setup optimization are explored, namely missing value treatment, estimation method and the use of additional non-financial variables. After being optimized, the models are once again ranked based on their predictive performance. Statistical inference is drawn using ANOVA and the Friedman test, along with the corresponding Tukey and Nemeyi pos-hoc tests. In their baseline form, the Czech credit scoring models are found to be outperformed by the foreign benchmark model. Treating the missing values by OLS imputation and estimating the models by probit, significantly is found to significantly improve their predictive performance. In their optimized form, the difference in predictive performance between Czech and foreign credit scoring model is found to be only marginal. JEL Classification G28, G32, G33, G38 Keywords credit scoring, multiple discriminant analysis, logit analysis, probit analysis Author's e-mail 71247263@fsv.cuni.cz Supervisor's e-mail...

Machine learning is becoming a part of everyday life and has an indisputable impact across large array of industries. In the financial industry, this impact lies particularly in predictive modelling. The goal of this thesis is to describe the basic principles of artificial intelligence and its subset, machine learning. The most widely used machine learning techniques are outlined both in a theoretical and a practical way. As a result, four models were assembled within the thesis. Results and limitations of each model were discussed and these models were also mutually compared based on their individual per- formance. The evaluation was executed on a real world dataset, provided by Home Credit company. Final performance of machine learning methods, measured by the KS and GINI metrics, was either very comparable or even worse than the performance of a traditional logistic regression. Still, the problem may lie in an insu cient dataset, in the improper data prepara- tion, or in inappropriately used algorithms, not necessarily in the models themselves.

Thesis submitted in fulfillment of the requirements for the degree of Master of Management in Finance & Investment In the Faculty of Commerce and Law Management Wits Business School at the University of the Witwatersrand, 2015 / Risk based pricing has been a topic of discussion since the 2008 financial crisis as a result of the on-selling of packaged sub-prime assets. This paper will highlight the importance of correctly assessing risk within the framework of consumer credit provision. We will begin with a brief overview of the South African unsecured lending market, look into the definition of risk based pricing and the impact it has had in the market and conclude the paper by using a model by Robert Phillips to calculate the interest rate to be offered to a customer. / AC2016

Credit scoring is critical for banks to evaluate new loan applications and monitor existing customers. Machine learning has been extensively researched for this case; however, the adoption of machine learning methods is minimal in financial risk management. The primary reason is that algorithms are viewed as ‘black box models’ and cannot satisfy regulatory requirements. While deep learning methods such as LSTM have been evaluated for behavioural credit scoring based on performance, research has not holistically evaluated these models on performance and explainability. To answer the research question, How can traditional machine learning and deep learning methods conform with regulatory guidelines for explainable artificial intelligence (XAI), and are they preferable to benchmark methods? this thesis used a public customer credit card dataset to compare the performance and explainability of machine learning and deep learning models against the benchmark statistical model linear regression. Model performance was evaluated using ROC-AUC, accuracy, Brier scores, F1 scores and the G-mean. The McNemar test evaluated whether, through pairwise comparison, the model performances were statistically different. The models were then evaluated on whether local and global explanations could be ascertained using feature/permutation importance and SHAP. The results found that neither the machine learning model, XGBoost, nor the deep learning model, LSTM, produced a statistically superior performance from the benchmark model. While there were performance improvements, only the machine learning model using post-hoc methods could produce local and global explanations. Given the strict regulatory environment, it is understandable that banks are hesitant to implement machine learning or deep learning models that lack the adequate levels of explainability regulators require.