Case 041Credit analysis and lendingCore
Anantya Credit gives you default counts for three loan vintages of different ages. Estimate the lifetime default rate and expected loss on the youngest vintage, and judge the provision held against it.
1The situation
Anantya Credit lends three-year personal loans. Its risk team sends you the default history by vintage, the year the loans were made. The 2021 vintage of 4,000 loans has had 120 defaults in three years: 40 by the end of year 1 and 80 by the end of year 2. The 2022 vintage of 5,000 loans has had 110 defaults in two years, 55 of them in year 1. The 2023 vintage of 6,000 loans is one year old and has had 60 defaults.
Average exposure at default is Rs 5 lakh and the loss given default is 55%. The company holds a provision of Rs 3 crore against the 2023 vintage.
2Your task
Estimate the lifetime probability of default and the expected loss on the 2023 vintage, and say whether Rs 3 crore is enough.
Quick check
Pooling everything, 290 defaults on 15,000 loans is about 1.9%. Is that the right default rate to apply to the 2023 vintage?
Worked solution
Try it on paper, then open one step at a time.
30-second answerThe answer to give first
The 2023 vintage is heading for a lifetime default rate of about 3.0%, an expected loss of about Rs 4.95 crore, so the Rs 3 crore provision is about Rs 1.95 crore short. Compare vintages at the same age: year-1 rates are 1.0%, 1.1% and 1.0%, so 2023 is tracking the older books. Both older vintages double between year 1 and year 2, and 2021 adds half again by year 3. Project 2023 along that curve: 1.0%, 2.0%, 3.0%, which is 180 defaults.
Step 1Why can you not just pool the defaults?
Think of judging three batches of saplings planted in different years by counting the dead ones. The batch planted last year has had one winter; the oldest has had three. Counting all the dead across all the batches tells you nothing about how many of the young batch will eventually die. A loan vintageAll the loans made in one period, tracked together over their life so that performance can be compared with other periods at the same age. must be read at a given age, and the young vintage projected along the shape the old ones have already traced. The pooled rate here is about 1.9%; the lifetime rate is nearer 3%.
Step 2How do you build the curve and project it?
Turn counts into cumulative rates. 2021: 40 of 4,000 is 1.0% at year 1, 80 is 2.0% at year 2, 120 is 3.0% at year 3. 2022: 1.1% then 2.2%. 2023: 1.0%. Now read the shape. Both older vintages double between year 1 and year 2, and the only vintage old enough to show year 3 adds half again, so the multipliers are 2.0x and then 1.5x. Apply them to 2023's 1.0%: 2.0% at year 2 and 3.0% at year 3. Because the loans run three years, that 3.0% is the lifetime rate, 180 defaults on 6,000 loans.
| Vintage | Loans | Year 1 | Year 2 | Year 3 |
|---|---|---|---|---|
| 2021 | 4,000 | 1.00% | 2.00% | 3.00% |
| 2022 | 5,000 | 1.10% | 2.20% | not yet |
| 2023 | 6,000 | 1.00% | 2.00% projected | 3.00% projected |
Step 3What is the expected loss, and is the provision enough?
| N | loans in the 2023 vintage |
| PD | lifetime probability of default, projected at 3.0% |
| EAD | exposure at default, Rs 5 lakh a loan |
| LGD | loss given default, the 55% not recovered |
Rs 3 crore covers 109 defaults, a lifetime rate of 1.8%, which the vintage will pass during its second year with a full year still to run. It looks as if the provision was set on observed defaults so far, not on the lifetime expectation. Say that plainly: the provision should rise by about Rs 1.95 crore, and if the 2023 book was written with looser standards than 2021, the curve could steepen; a 20% steeper curve gives 3.6% and about Rs 5.94 crore.
State the limits. Three vintages is a thin sample; the year-2-to-3 multiplier rests on one vintage. Recoveries can take years, so the 55% LGD is itself an estimate. And a book growing 20% a year is a reason to ask whether the lender has been chasing volume, which shows up in vintages before it shows up in the total. The method is still right: same age, same curve, then project.
Where candidates lose it
The common loss is pooling all defaults over all loans and quoting about 1.9%. That averages a one-year-old book with a three-year-old one and understates the lifetime rate by a third.
The second is treating the 2023 vintage's 1.0% as its final rate because it looks like the others' year-1 rate. It is the starting point of a curve, not the end of one.
What the interviewer asks next
- The 2023 vintage was 50% larger than 2022. What would you check before trusting the 2021 curve for it?
- How would you treat loans that are 90 days overdue but not yet written off?
- If the 2024 vintage shows 1.5% at year 1, what provision would you set?
Asked at Jane Street, Credit Risk, London, 2025 (Wall Street Oasis): credit risk data that I had to analyse using Excel or Python
Asked at Jane Street, Credit Risk, London, 2025 (Wall Street Oasis): Asked to analyse the data using advanced and dynamic probability metrics
Company names and figures are illustrative.

