Exploding the Myths of Scorecard Development

This paper discusses the misconceptions that exist in a model development of scorecard. The paper aims to:

• Explode these myths;
• Understand the problem that the scorecard is trying to solve.

The author discusses the following minutiae for developing a good scorecard model:

• Identify the business problem for which the model is needed;
• Provide performance definition that highlights what the model is going to predict;
• Hit upon an early measure of a write off stage so as to minimize bad debt;
• Use the concept of indeterminates to differentiate the good and bad patterns in the data;
• Realize that changes in the measures of predictive model can be done by changing the design of the model;
• Select characteristics so that it can be implemented in the life system.

The paper:

• Explains need for inclusion of bureau information stating that although using them provides certain uplift, they must not dominate the scorecard.
• Details out the concept of reject inference whereby declined applications are inferred to ensure that the future scorecard caters to all risk profiles on which the organization makes a decision.
• Concludes that the success of any model will depend on the design of the actual business issue and the understanding applied by the developers and end users.