Scoring is a decision system, not a number. Classifying customers on a statistically derived scorecard gives an objective, repeatable basis for credit decisions — and a measurable one, so both the model and the strategy built on it can be improved over time.
- Outputs are objective, consistent and repeatable — the same case always gives the same result
- Strategies can be implemented directly in the decision process
- Scalable to handle peaks in application volumes, and therefore cost effective
- Performance of the scorecard, of each characteristic within it, and of the credit decision is measurable against benchmarks
- Monitoring creates a feedback loop: the model can be adjusted and the decision strategy controlled
- A continuous scale of risk lets the approval rate be moved to take the next least risky applicants, or decline the next highest risk
Engagement
A credit scoring project
StatDec's scoring services can cover the whole project cycle — project design and data quality analysis, development, implementation, independent model validation and ongoing maintenance — or support only the parts you need. Scoring projects generally take one of four forms.
Application scores support the credit initiation process — the point at which a customer applies. In their most basic form they are built on information collected from the application form; usually they are augmented by credit bureau reports and by information from any existing relationship with the applicant. They predict the probability of the applicant defaulting and are used to underwrite a credit proposal.
With a cut-off table, a strategy curve can be produced to guide the credit process and answer questions such as:
- With a 10% increase in the approval rate, what is the effect on the loss rate?
- To manage to a given loss rate, what should the approval rate be?
- If the loss rate is low and we can approve more applicants, how many — and which of those previously rejected — should now be approved?
- If the loss rate is too high and we must reject more, how many — and which of those currently approved — should now be rejected?
Behaviour scores are developed from the account activity of existing customers. Where a customer holds more than one account, an overall customer score can be developed using information from all of them. Segmentation based on behaviour scores gives a clear, objective and statistically optimal classification.
Their predictive power, range of uses and ease of management place them at the centre of most financial institutions' decision-making functions. Common areas of application include:
- Key driver for risk segmentation in collections strategies
- Provisioning under IFRS 9
- Basel capital requirements — model design for risk components (PD, LGD)
- Limit management
- Renewals and cross-selling
- Credit assessment for existing clients
- Marketing campaigns
- Portfolio quality tracking
Customised bureau scores are built on raw credit bureau data rather than bought as a finished product. Where the raw data are available, they are considered best practice compared with externally provided generic scores, because they:
- are customised to the specific business or portfolio, and can therefore outperform external generic scores
- can be validated down to component level and adjusted when needed
- provide competitive advantage — you are not using the same information as everybody else
- are cost effective, as raw data are usually less expensive than bureau score services
- are optimal to the decision point of reference, for example the application point
StatDec's approach to optimal decisions in multi-scoring environments, in a simplified layout. A MetaScore is a final layer in which the risk information carried by several different models is incorporated into one final model.
MetaScores have significant advantages over the approaches usually followed where several scores are in play:
- A single measure of risk level
- Quantifies off-setting risk factors by considering information sources together rather than individually
- The approval process can be simplified by removing multiple stages
- Manageable output, with one final score distribution
- Unlike a matrix, using a MetaScore is a process that can be validated and benchmarked against any other practice
Our MetaScore approach has been implemented successfully in multiple organisations, with considerable gains in the efficiency of the decision process and in effectiveness in terms of the approval rate.
FAQ
Frequently asked questions
An application scorecard supports the credit initiation point. It is built on application form data, usually augmented by credit bureau reports and any existing relationship with the applicant, and predicts the probability of default. A behaviour scorecard is developed from the account activity of existing customers and drives limit management, renewals, cross-selling, collections segmentation and provisioning.
A scorecard is developed on the historical performance of your own portfolio, so the starting point is application or account data covering a period long enough for the outcome to be observed. Every project begins with data quality analysis, because the definition of the target event and the reliability of the inputs determine what can realistically be built.
A customised bureau score is built on raw credit bureau data rather than bought as a finished product. Because it is fitted to your portfolio and decision point it can outperform an external generic score, it can be validated down to component level and adjusted when needed, and raw data are usually less expensive than a bureau score service.
A MetaScore is a final modelling layer in which the risk information carried by several different models is combined into one score. It gives a single measure of risk, quantifies off-setting risk factors by considering the sources together rather than one at a time, simplifies multi-stage approval processes and, unlike a matrix, can be validated and benchmarked.
A cut-off table allows a strategy curve to be produced, which quantifies the trade-off between the two. It answers questions such as what a ten per cent increase in the approval rate does to the loss rate, or what approval rate is consistent with a target loss rate, so the accept and reject boundary becomes a business choice.
Yes. The operating environment of a scorecard is not stable: systems, policies, processes, customer profiles and economic conditions all change. StatDec supports implementation, validation and ongoing maintenance as part of the project cycle, and provides independent external validation aligned with the ECB and EBA framework where a separate review layer is required.
Discuss a scorecard project
Tell us about the portfolio, the decision point and the data you hold, and we will set out what can be developed and how it would be validated.
Get in touch