Credit Scorecards: Weight of Evidence, Information Value & Points Scaling

What you'll build

What you will be able to do

Each lab ends with a concrete result that you build yourself and the checker confirms, cell by cell.

• Group ten fine classes of debt-to-income into five coarse classes with an approximate-match lookup, count goods and bads per class and check bad rates and class sizes.
• Turn good and bad counts per class into distributions, calculate the Weight of Evidence of each debt-to-income class and chart the WoE pattern.
• Calculate the WoE and the Information Value of the credit-enquiries characteristic and rate a shortlist of eight characteristics with the usual IV bands.
• Check the bad-rate trend of "years with current employer", flag the reversal, merge the two affected classes and recalculate goods, bads and WoE.
• Scale a WoE logistic model to 600 points at 30:1 odds with 20 points to double the odds and turn every attribute into rounded scorecard points.
• Score six new applicants with approximate-match lookups in Larkspur's three-characteristic scorecard and convert each score back into odds and a probability of default.
• Build Larkspur's gains table by cut-off score, chart approval rate against the bad rate of approved loans and pick the lowest cut-off that meets a 3% bad-rate target.
• Compare the score distribution of last quarter's 2,920 applications with the development sample, calculate the PSI and its verdict, and measure how much the debt-to-income mix moved the average score.