Jonathan Crook

Credit Scoring and Its Applications

From the Publisher: About the Author Lyn C. Thomas is a Professor of Management Science at the University of Southampton. Jonathan N. Crook is Reader in Business Economics at the University of Edinburgh. David B. Edelman is Credit Director of Royal Bank of Scotland, Edinburgh.

A comparison of neural networks and linear scoring models in the credit union environment

Abstract The purpose of the present paper is to explore the ability of neural networks such as multilayer perceptrons and modular neural networks, and traditional techniques such as linear discriminant analysis and logistic regression, in building credit scoring models in the credit union environment. Also, since funding and small sample size often preclude the use of customized credit scoring models at small credit unions, we investigate the performance of generic models and compare them with customized models. Our results indicate that customized neural networks offer a very promising avenue if the measure of performance is percentage of bad loans correctly classified. However, if the measure of performance is percentage of good and bad loans correctly classified, logistic regression models are comparable to the neural networks approach. The performance of generic models was not as good as the customized models, particularly when it came to correctly classifying bad loans. Although we found significant differences in the results for the three credit unions, our modular neural network could not accommodate these differences, indicating that more innovative architectures might be necessary for building effective generic models.

Recent developments in consumer credit risk assessment

Consumer credit risk assessment involves the use of risk assessment tools to manage a borrower’s account from the time of pre-screening a potential application through to the management of the account during its life and possible write-off. The riskiness of lending to a credit applicant is usually estimated using a logistic regression model though researchers have considered many other types of classifier and whilst preliminary evidence suggest support vector machines seem to be the most accurate, data quality issues may prevent these laboratory based results from being achieved in practice. The training of a classifier on a sample of accepted applicants rather than on a sample representative of the applicant population seems not to result in bias though it does result in difficulties in setting the cut off. Profit scoring is a promising line of research and the Basel 2 accord has had profound implications for the way in which credit applicants are assessed and bank policies adopted.

Support vector machines for credit scoring and discovery of significant features

The assessment of risk of default on credit is important for financial institutions. Logistic regression and discriminant analysis are techniques traditionally used in credit scoring for determining likelihood to default based on consumer application and credit reference agency data. We test support vector machines against these traditional methods on a large credit card database. We find that they are competitive and can be used as the basis of a feature selection method to discover those features that are most significant in determining risk of default.

Credit scoring with macroeconomic variables using survival analysis

Survival analysis can be applied to build models for time to default on debt. In this paper, we report an application of survival analysis to model default on a large data set of credit card accounts. We explore the hypothesis that probability of default (PD) is affected by general conditions in the economy over time. These macroeconomic variables (MVs) cannot readily be included in logistic regression models. However, survival analysis provides a framework for their inclusion as time-varying covariates. Various MVs, such as interest rate and unemployment rate, are included in the analysis. We show that inclusion of these indicators improves model fit and affects PD yielding a modest improvement in predictions of default on an independent test set.

The demand for household debt in the USA: evidence from the 1995 Survey of Consumer Finance

This paper investigates first the factors which determine whether a household is likely to be rejected or discouraged from applying for credit and second, which factors explain the amount of debt which a household demands. All of the published papers which have addressed the first question have used data relating to the period 1978–1983 or, in one case only, 1984–1989. All the papers which have investigated the second question have used data for the earlier period only. In this paper data for 1990–1995 from the latest version of the Survey of Consumer Finance are used. A univariate probit model with standard errors corrected for sampling weights is used to shed light on the first question and a bivariate probit model followed by a two stage least squares selection model to estimate the demand for debt. Results are found which are similar to those for the earlier years and some new ones. In common with earlier results it is found that a household demands less debt when the head of the household is aged over 55 years and when the head is relatively risk averse. A household demands more debt when its income is higher, when it owns its own home, when the family size is larger and the head is working. It was also found that the result of being black increases the probability of being credit constrained but it does not increase a households demand for debt. This is therefore a result found consistently for the late 1980s through to the early 1990s. In addition to these results which are in common with earlier papers for earlier periods it was also found that if a household has a large expected expenditure in the next few years it demands a larger amount of debt now, that the higher the net worth of a household the less debt it desires and that a households expectations concerning future interest rates has no effect on its demand for debt.

Sample selection bias in credit scoring models

One of the aims of credit scoring models is to predict the probability of repayment of any applicant and yet such models are usually parameterised using a sample of accepted applicants only. This may lead to biased estimates of the parameters. In this paper we examine two issues. First, we compare the classification accuracy of a model based only on accepted applicants, relative to one based on a sample of all applicants. We find only a minimal difference, given the cutoff scores for the old model used by the data supplier. Using a simulated model we examine the predictive performance of models estimated from bands of applicants, ranked by predicted creditworthiness. We find that the lower the risk band of the training sample, the less accurate the predictions for all applicants. We also find that the lower the risk band of the training sample, the greater the overestimate of the true performance of the model, when tested on a sample of applicants within the same risk band — as a financial institution would do. The overestimation may be very large. Second, we examine the predictive accuracy of a bivariate probit model with selection (BVP). This parameterises the accept–reject model allowing for (unknown) omitted variables to be correlated with those of the original good–bad model. The BVP model may improve accuracy if the loan officer has overridden a scoring rule. We find that a small improvement when using the BVP model is sometimes possible.

Credit constraints and US households

Empirical tests of the permanent income/life cycle hypothesis of consumption often find that consumption is excessively sensitive to current income. Some evidence suggests that this finding is due to the existence of credit constraints on consumers. Using data relating to a period before 1983, Jappelli has tried to identify the characteristics of those who are credit constrained. However, conditions affecting the supply of credit have changed since then. Jappellis work is replicated for a period before 1989. Using direct questions from the Survey of Consumer Finance, the paper identifies characteristics of households who have been rejected or discouraged from applying for credit, using untransformed household characteristics then using their principal components. The results show that being old and not working for pay, being White or having a high income and high wealth reduces the probability that a household will be rejected or discouraged from applying for credit.

Modelling profitability using survival combination scores

Abstract The paper presents the first empirical investigation of the relationship between present value of net revenue from a revolving credit account and times to default and to second purchase. The analysis is based on the data for a store card which is used to buy ‘white’ durable goods in Germany. It is demonstrated that there exists a relationship between the above given measures. It appears that there is a scope for improving profit if an application for a store card is assessed by using a model which estimates the revenue and includes the survival probability of default and the survival probability of second purchase (a survival combination model) rather than merely a static probability of default predicted by a logistic regression.

Reject inference, augmentation, and sample selection

Many researchers see the need for reject inference in credit scoring models to come from a sample selection problem whereby a missing variable results in omitted variable bias. Alternatively, practitioners often see the problem as one of missing data where the relationship in the new model is biased because the behaviour of the omitted cases differs from that of those who make up the sample for a new model. To attempt to correct for this, differential weights are applied to the new cases. The aim of this paper is to see if the use of both a Heckman style sample selection model and the use of sampling weights, together, will improve predictive performance compared with either technique used alone. This paper will use a sample of applicants in which virtually every applicant was accepted. This allows us to compare the actual performance of each model with the performance of models which are based only on accepted cases.