How credit scores can improve risk assessment

Credit scoring is used for determining the probability that a borrower will repay their loan on time.

It is based on attributing points to customer characteristics reflecting both quantitative (mainly financial data) and qualitative elements (e.g. marital status, education level, etc.). It applies both to individual customers and enterprises (especially in the micro and SME segments).

The history of using scoring models in credit processes dates back to the Second World War. David Durand can be regarded as a precursor here. In 1941, in a project conducted for the US National Bureau of Economic Research, he studied the characteristics of "good" and "bad" borrowers. In the following decades, this method became more and more popular, but the real breakthrough was the introduction of the US Equal Credit Opportunity Act in 1975, which prohibits discrimination against borrowers. The risk assessment based on scoring models was ideally suited to the guidelines contained in the Act, ensuring full objectivity in the granting of increasingly popular credit products. Although at first scoring was used for the simplest products (credit cards), over time banks started to use this method also for cash loans, and later for mortgage loans. Currently, credit scoring is the most common method used for assessing the credit risk of individuals.

Comarch Loan Origination

Credit scoring explained

The scoring of a customer's credit score in credit processes is usually based on a scorecard. It takes the form of a set of customer characteristics and corresponding attributes with assigned points. For example, for the "education" feature we define three attributes (basic, secondary, higher) and assign points to each of these attributes (in our example it could be 3, 5 and 8 respectively). After analyzing all the characteristics, the individual points are summed up and constitute the final score. Within each scoring model, a cut-off point is also defined, which illustrates the border between "good" and "bad" borrowers. However, the scale of points possible to obtain is often divided into more than 2 categories, which makes it possible to maintain a separate approach to the customer depending on the level of estimated risk.

However, the question is why specific characteristics are taken into account in a given scoring model and on what basis specific points are assigned to particular attributes. Building a model or a scorecard is not a trivial activity, as it requires expert knowledge of mathematics or statistics. Each bank has data on loans granted and their repayment. This data is divided into groups of "good" and "bad" loans, and then within these groups sets of data are determined. Those sets will be used to build a scoring model, and to verify its effectiveness.

The scope of analyzed data depends on the type of information previously collected and archived by a given institution and the type of scoring used (application- or behavior-based). Using various statistical techniques or data mining, risk departments try to determine which parameters have a significant impact on the probability of a default. For years, logistic regression has been the most popular method used for building scoring models, but this area also uses decision trees, genetic algorithms, neural networks and other AI or machine learning methods. Therefore, specialist statistical software or data mining is most often used for creating scoring models.

Creating a scoring model and building a scorecard is the first stage in the full life cycle of this risk assessment method. Before it is fully utilized in credit processes, it is verified, most often under the Champion-Challenger approach, where the new model works alongside the current one and the metrics of effectiveness of both variants are checked. If the new model is more efficient, it becomes a Champion and is used for determining new customer risk. In order to streamline the scoring calculation process, credit processes most often use scoring engines, i.e. IT solutions based on business rules processing. Such systems also provide the ability to monitor the effectiveness and efficiency of scoring models implemented. This is important as the risk assessment should be adjusted to the current internal procedures of the bank and the general economic and demographic situation on a given market.

Credit scoring software

Banks need to use efficient tools to best assess the creditworthiness of their customers and to reduce the risk of potential insolvency. Therefore, more and more often they use credit scoring software, which allows for determining the risk of an undesirable event such as a defaulted loan. Such solutions are most often one of the key elements of the entire system infrastructure supporting the bank's employees in the credit granting processes.

The most important advantages of scoring tools are as follows:

  • significant shortening of risk assessment processes and improvement of calculation efficiency. Creditworthiness assessment processes may be accelerated by as much as 75 per cent by selecting the appropriate software in the bank and ensure a multiple increase in loan sales.
  • faster and more effective assessment of creditworthiness, minimizing the number of risky loans granted
  • better efficiency, with the same employment - credit scoring software allows to streamline credit processes without additional human resources, i.e. the need to hire additional employees
  • automation of the credit assessment process, which also reduces the risk of human error and potential fraud
  • integration with other systems provides the possibility of using a wide range of data but also allows to determine the scoring within different systems or business processes
  • minimizing the risk of a wrong decision or granting a loan on conditions unfavorable to the bank
  • comprehensive credit policy management and its constant monitoring – all data and scoring calculations are performed and stored in one place.
Want to learn more?

Tell us about your business needs. We will find the perfect solution.