Verifying Borrowers Characteristics Stability
One of the main conditions for successful use of the scorecard is the absence of significant changes in borrower characteristics over the course of time.
If the development of the scorecard lasts for a long time, changes in the credit portfolio may occur. Since such changes are not reflected in the data that were used to develop the scorecard, then prior to the implementation of the scorecard into the process of evaluating a borrowers’ creditworthiness, we need to re-check that the quality of the scorecard remains the same and to remove any deviations, if any.
To determine the presence of changes in borrower characteristic, we use the Population Stability Index (PSI), that allows finding differences between the expected and actual borrower characteristics in all the selected risk ranges:
Scorecard Correction (if required)
In case of significant changes in borrower characteristics (PSI>0.25), the scorecard must be corrected.
The first step in defining the required corrections is the search for changes in particular borrower characteristics. For that purpose we compare the expected and actual distributions for all borrower characteristics that participate in the scorecard.
To evaluate the changes that have occurred (in points), we use the following index: Index=∑(%Actual-%Expected)Points
This example illustrates a decrease in share of borrowers of the older age group, who typically have lower credit risk. At the same time, we see a considerable increase in share of borrowers in the following categories: from 18 to 24 years of age and from 25 to 29 years of age.
Possible reasons for these changes:
Marketing campaigns that got traction with younger borrowers.
In this case, we can use penalties (lowering of credit rating) for younger borrowers.
Older borrowers are less interested in this particular credit product.
To attract desirable customers, we can use both bonus points (awarding a higher credit rating) and offers of complimentary financial services. More younger people reside in the area (for example, due to numerous educational facilities).
In this case, we need to adjust the scorecard so that it reflects regional peculiarities.
The cut-off point must provide:
- Acceptable level of credit risk
- Required level of accepted applications
Policy Rules Definition
The score is the natural criterion for formulating risk-based working rules for borrowers. Simpler rules can mandate the level of detail with which the institution analyses the borrower’s application depending on his/her score; more sophisticated rules can involve an individual selection of the interest rate (risk premium) or of the loan duration.
The standard file format for storing scoring models is XML-based PMML (predictive modeling markup language). This format is supported by many software applications and is highly recommended.
The scorecard can also be represented as C++, C#, SQL or any other source code that can be directly integrated into the decision-making system. This option can cause certain difficulties in maintaining the scorecard.
Sometimes scores are manually transferred to the existing application processing or customer management system or workflow management system via a special interface provided in such systems that allows manual entry of point-based scorecards.
Champion-Challenger Deployment Approach
The Champion-Challenger approach allows using a new scorecard alongside the existing rules for credit risk assessment.
In the first stage of implementing a new scorecard, we need to check that it is superior to the existing methods of risk evaluation (or to the existing scorecard).
For that purpose, we process the majority of incoming loan applications using our existing methods (that are called Champion), while remaining loan applications are processed using the new scorecard (that is called Challenger).
Processed results are compared; if the new scorecard demonstrates a better assessment of borrowers, it will be used instead of old methods (and becomes the Champion method). If needed, the new scorecard is improved while it is in the Challenger mode, until we see an acceptable improvement in risk assessment.
Credit Scoring Software is the most easy-to-use and the fastest to integrate scoring system.