Big Data, Credit Risk Strategies, and Analytics
Big Data, Credit Risk Strategies, and Analytics
Prescriptive analytics gives a recipe for business success in form of actions by answering the key question, How can we make it happen? In the credit risk arena, the answer to this question is found in credit risk strategy.
Join the DZone community and get the full member experience.Join For Free
Hortonworks Sandbox for HDP and HDF is your chance to get started on learning, developing, testing and trying out new features. Each download comes preconfigured with interactive tutorials, sample data and developments from the Apache community.
Gartner's analytic value escalator identifies four different types of analytics — descriptive, diagnostic, predictive and prescriptive — ordered by level of difficulty and business value. Prescriptive analytics, at the most complex level but offering the greatest value, is at the top of that escalator. Prescriptive analytics gives a recipe for business success in form of actions by answering the key question, "How can we make it happen?" In the credit risk arena, the answer to this question is found in credit risk strategy.
Credit risk strategy is the process that follows after scorecard development and before its implementation. It tells us how to interpret the customer score and what would be an adequate actionable treatment corresponding to that score. The winning strategy is the one which, if implemented, increases the customer base, reduces credit risk, and maximizes profit.
Before beginning strategy analysis and running many what-if scenario iterations, it is important to identify the clear business objective and understand the business processes that consequently shape the analysis. The simplest and most common form of credit risk strategy is based on a one-dimensional cut-off for an accept or reject decision. The cut-off level — the minimum score for credit approval — can be a hard cut-off with a single fixed value or can have adjustable values with multiple treatments such as unconditional accept, conditional accept, or reject.
Often, lenders use a segmentation strategy to identify different cut-off levels across customer segments. Segmentation may be carried out by many factors including region, demographics, channel distributions or previous declined customers. For example, strategy segmentation can be based on the same bad rate across customer segments, benefiting from higher approval rates for better segments or the same approval rate can be preserved across all segments, resulting in lower bad debt across better segments.
Cut-off levels depend on the overall business objectives. For example, if the objective is based on retaining an 80% acceptance rate, the retrospective analysis may specify the cut-off value to be 320. If, however, the objective is based on a maximum default rate of 6%, the strategy could be even more restrictive and the cut-off level increase to 360. If the strategy is based on a pure profit/loss measurement, this would require setting the cut-off score to 440, as per the example in Figure 1.
Figure 1: Different cut-off strategies
Table 1 illustrates how different key performance indicators (KPIs) such as acceptance rate, default rate, or profit amount inform scorecard cut-off levels. Different departments within a company may have different, often conflicting, objectives. For example, credit risk departments aim to reduce default rate and cut the debt amount, while a marketing department may request lifting up the cut-off level in order to expand their customer base. A compromise solution could be to design a new scorecard that, for the same bad rate, increases the number of accepts, or for the same approval rate, leads to a decrease in the bad rate. Increasing the number of accept decisions is better for growing market share or having greater overall profitability. Aiming to decrease the bad rate is more appropriate during economic drifts.
Table 1: Scorecard cut-off levels Informed by different KPIs
More sophisticated credit risk strategies have multiple cut-off levels or combine two or more credit scores; for example, internal application score and bureaux scores. Often, strategies include other predictive models such as customer retention or response rate or customer lifetime value. These behavioral scores, combined with policy and regulatory rules and business KPIs, can make the best advantage of predictive analytics and business rules.
Figure 2: Multiple cut-off levels for multiple treatments
Scores may be further used for risk-based pricing to adjust product offers such as interest rates, credit limits, repayment terms, and so on. Risk-based pricing takes many forms from one-dimensional multiple cut-off treatments based on profit/loss analysis (for example, accept with lower limit) to a matrix approach combining two dimensions (for example, behavioral score and outstanding balance to identify credit limits or interest rates). The matrix approach can also be adopted for a simple optimization in order to control operational cost. For example, combining two predictive models — scores and response rate — may enable marketing departments to focus on customers with low risk and high probability to respond to an offer.
Figure 3: Risk-based pricing using matrix approach
Figure 4: Retention and risk segmentation strategy
There is danger in using over-simplistic strategies; the strategy may, for example, reject risky customers that would have been loyal or highly profitable. A customer lifetime value (CLV) model helps to identify valuable segments; lenders, however, can be reluctant to use CLV as it can be extremely difficult and complex to determine. In such situations, a thorough insight analysis may help identify valuable segments and adjust the strategy accordingly.
Published at DZone with permission of Natasha Mashanovich , DZone MVB. See the original article here.
Opinions expressed by DZone contributors are their own.