Delinquency on three successive vintages of two-wheeler loans is rising while disbursements grow 40% and the approval rate climbs. Separate seasoning from underwriting drift and say what the credit committee should do.

The rise is underwriting, not seasoning: at the same six months on book, each newer vintage is worse. If the loans the old policy would have approved still run at 2.1%, the 13 extra approvals per 100 applicants in the Q3 vintage imply delinquency of about 11%. They are 19% of approvals and about 55% of the bad loans. Tighten the cut-off back for that marginal segment, rather than slowing the whole book.

A newly planted orchard has no rotten apples, because nothing has had time to ripen. Loans are the same: a borrower cannot be 30 days late in the first month. A fast-growing book is full of young loans, so its overall delinquency rate is diluted even when every new vintage is worse than the last. That is why the board's number is flat. The fix is a {term('vintage analysis', 'Tracking each batch of loans disbursed in the same period separately, and comparing batches at the same age since disbursement.')}: line the vintages up by months on book, so every batch is compared at the same age.

Assume the applicants did not change and the old policy's 55 approvals per 100 still behave at 2.1%. Then the Q3 vintage's 68 approvals contain 55 old-policy loans and 13 extra ones. The 13 extra must carry the rest of the delinquency: 68 x 3.8 less 55 x 2.1, divided by 13, is about 11.0%, roughly five times the old book. The Q2 vintage tells the same story: its 6 extra approvals imply about 10.2%. Two independent readings agreeing is what makes the finding credible.

Target the marginal segment. Identify the score bands and sourcing channels that the looser policy let in and restore the old cut-off there, rather than slowing all lending. The old-policy book is performing, and cutting it would punish good business. Then ask whether the marginal loans could be priced for their risk: at around 11% early delinquency the eventual loss is likely to be several per cent of the loan, more than the rate difference a competitive market allows. Add a trigger to the policy: any vintage whose six-month delinquency exceeds the previous one by a set margin goes back to the committee automatically.

Name the limits of the inference. The calculation assumes the applicant pool stayed the same; if a new dealer channel brought weaker applicants, part of the rise belongs to the channel, not the cut-off. Seasonality matters too: two-wheeler loans disbursed in a festival quarter can behave differently, so compare each vintage with the same quarter a year earlier before concluding.