Scoring
In recent years, computer-assisted procedures that allow statements about the creditworthiness of a potential borrower on the basis of mathematical-statistical methods have increasingly been used in the consumer credit business, such as scoring.
Techniques like scoring are used by banks, online providers and merchants in the context of credit-financed consumer purchases by their customers. They enable lenders to make efficient and more objective credit decisions without having to examine the personal income and asset situation of borrowers in detail.
Scoring models: Predicting creditworthiness
Every lender faces the challenge of estimating the likelihood of repayment when making a credit decision. This is largely determined by the borrower's future ability to pay.
Traditionally, lenders use information about their customers and general experience from credit business when assessing creditworthiness. In scoring (from English score = point value), mathematical-statistical methods are used to try to generate a statement about creditworthiness based on observable and as discriminatory as possible borrower characteristics.
Scoring procedures are prediction models in this sense: they use historical data to make as reliable a prediction for the future as possible. Scoring aims for better accuracy in prediction than would be achieved by the subjective assessments of loan officers. The typical application area for scoring procedures is the consumer credit business, which is primarily characterised by a large number of installment loans with relatively small loan amounts.
Normally, such installment loans are granted without specific collateral; repayment of the loan is therefore determined solely by the borrower's personal ability to pay. Because a detailed examination of assets and income would be too time-consuming for the comparatively small loan amounts, but a meaningful basis for credit decisions is still needed, scoring is used.
Methodologically, it therefore contrasts with so-called credit rating, which is mainly used in corporate lending. While scoring focuses on the standardised use of a few, highly predictive borrower data, credit rating aims at as comprehensive and systematic a consideration and individual assessment of customer data as possible according to a rating scheme.
Scoring: Necessary data, methods and
For use in a scoring model, both personal characteristics of the borrower and data about their economic circumstances are considered. Personal characteristics include, for example, information on occupation, employment relationship, marital status, the duration and quality of the business relationship, and payment behaviour. Economic circumstances include information on disposable income, assets and spending behaviour.
The lender generally obtains this data from the loan application, experience from the existing business relationship or — especially for new customers — via credit reference agencies. In consumer lending, obtaining a SCHUFA credit report is common.
Methods in scoring
A score function — also called a scorecard — weights borrower characteristics suitable for prediction and maps them in a calculation algorithm with which the individual score value for each borrower is determined.
Usually, the weightings are distributed so that no single characteristic dominates the others. The calculation of the score function is carried out on the basis of a large amount of historical borrower data using appropriate methods. Discriminant analysis, logistic regression, artificial neural networks and other data mining techniques are often used.
The goal is a scorecard that is as discriminating as possible, minimising Type I and Type II errors. A Type I error is a positive creditworthiness prediction despite poor creditworthiness of the borrower, while a Type II error is a negative creditworthiness prediction despite good creditworthiness. The credit score is the specific numerical or point value that results for a potential borrower from applying the scorecard based on their characteristics. There is a critical score value above which sufficient creditworthiness is no longer predicted.
Scoring - The informative value
The determined score value is an important factor in the credit decision. It not only provides indications of the probability of default on a loan, but can also be used to determine loan conditions and to calculate risk costs.
Scoring can thus also be used as an important control basis for consumer lending. However, scoring should not be the sole decisive factor for granting credit. Debt service capacity, assets and any collateral are further factors that should also be taken into account in credit decisions.
In consumer lending, the so-called household budget calculation plays an important role as another component in the creditworthiness assessment. Regular household income is compared with regular expenses.
The balance of income and expenses indicates how much money the borrower has available for interest and principal payments. If the financing balance is insufficient, granting the loan — despite a possibly positive scoring result — is not advisable.
Advantages of scoring models
The use of scoring should be evaluated in a differentiated way. Advantages include:
- the credit decision is overall accelerated thanks to automation. This means cost savings and more efficient processes for the lender, and typically faster credit decisions and potentially more favourable loan conditions for the borrower.
- Especially for the now frequently online consumer lending, scoring is practically indispensable. With just a few entries the user can often trigger an almost immediate processing of their loan request.
- credit granting becomes more objective due to the application of mathematical-statistical methods, as the influence of subjective assessments by loan officers is reduced. Credit decisions are transparent and easy to follow thanks to the calculation algorithm.
Disadvantages of scoring models
The following disadvantages should be taken into account:
- Standardisation can lead to the loss of informational advantages from a long-standing business relationship; qualitative data normally do not flow into the scoring.
- The informative value of scoring depends crucially on the timeliness and accuracy of the entered data. Incomplete or incorrect data lead to false results.
- Scoring models are generally not stable over time. They therefore require continuous and often costly review and, if necessary, further development with regard to their predictive power.
Scoring and data protection
From a data protection perspective, scoring procedures have been controversial in the past. The amendment to the Federal Data Protection Act (BDSG), which came into force on 1 April 2010, brought clarification here. According to this, scoring is permissible under certain conditions (see § 28 BDSG).
Essential for permissibility is, among other things, the use of scientifically recognised mathematical-statistical procedures. Companies that use scoring procedures are obliged to provide data subjects, upon request, with information about the scoring values determined in the last six months, the data used, the methodology and the findings obtained (see § 34 BDSG).
Credit reference agencies have a corresponding obligation to disclose information even for a twelve-month period. SCHUFA, as the most important credit reference agency in consumer lending, has been offering scores based on the data stored with it in addition to the classic SCHUFA report since 1997. Scores are available as a basic score or differentiated as an industry score for a total of seven industries.
Consumers, however, have the option to prohibit SCHUFA from disclosing score values relating to them. To what extent a prohibition on data disclosure affects the credit decision cannot be conclusively answered at present.