Behavioural scorecard for a leading US Bank to predict.
Delinquency likelihood of the accounts.
Assess expected losses on the delinquent accounts (DQ).
Rank the accounts by severity in order to prevent severe losses.
Variable screening using both statistics and business acumen. Iteration with different subset of variables is supported by Smart tool.
Determination of optimal probability cut off to obtain low error. In the validation sample, the model achieved very low error rate minimizing the cost arising from both type of errors – an attractive feature of ScoreBuilder, to make the decision.
False positives: Non-DQ accounts misclassified as DQ – Collection costs - 5%
False Negatives: DQ accounts misclassified as Non DQ – Loss incurred as no action was taken - 8%
High discriminatory scorecard as evident from the KS Statistic value of 0.63.