Part 08 — Credit Risk Deep Dive
Portfolio Analytics
Portfolio analytics is where lending beliefs meet repayment evidence. Underwriting can approve a file with elegant ratios; portfolio analytics asks whether that kind of file repaid across thousands of borrowers. For SME lending, this is especially important because losses do not emerge evenly. A new sourcing channel, digital campaign, GST-surrogate rule, DSA cluster or industry pocket can look clean for three months and then roll sharply into 30, 60 and 90 days past due (DPD).
The core artefacts are vintage curves, roll-rate matrices, bounce dashboards, collection-efficiency reports, concentration limits and exception cohorts. They should feed back into scorecards and models, monitoring EWS and collections operations.
Vintage Curves
Section titled “Vintage Curves”A vintage is a cohort of loans originated in the same period, usually month or quarter. Vintage analysis tracks how that cohort performs as it seasons. Do not compare a three-month-old cohort with a twelve-month-old cohort using cumulative 90+ DPD; the older cohort had more time to fail.
Common vintage metrics:
- cumulative first-payment default (FPD);
- ever 30+ DPD by month on book (MOB);
- ever 90+ DPD within 6, 9 or 12 months;
- current 30+ DPD by MOB;
- cumulative write-off or net credit loss;
- prepayment rate and closure rate.
Example unsecured business-loan cohort:
| Origination cohort | Booked amount | Loans | MOB 3 ever 30+ | MOB 6 ever 30+ | MOB 9 ever 90+ | MOB 12 ever 90+ |
|---|---|---|---|---|---|---|
| Apr-Jun 2025 | ₹42 crore | 1,850 | 3.2% | 6.8% | 2.4% | 3.1% |
| Jul-Sep 2025 | ₹55 crore | 2,420 | 4.4% | 9.1% | 3.8% | Not seasoned |
| Oct-Dec 2025 | ₹61 crore | 2,760 | 5.9% | 12.6% | Not seasoned | Not seasoned |
| Jan-Mar 2026 | ₹49 crore | 2,100 | 7.8% | Not seasoned | Not seasoned | Not seasoned |
This table says the problem started before 90+ DPD appeared. The Jan-Mar 2026 cohort has 7.8% ever 30+ by MOB 3 versus 3.2% for Apr-Jun 2025. Risk should not wait for NPA. Segment the cohort by channel, state, score band, loan amount, assessment method and DSA. The bad curve may be hiding in one source.
SIDBI and TransUnion CIBIL’s July 2026 MSME Pulse is a current market reminder: the report said outstanding commercial balances including individual business-oriented loans were ₹65.8 lakh crore as of March 2026, overall commercial portfolio delinquency was broadly stable, but unsecured business-loan accounts originated in March 2025 showed 2.9 times higher ever-90+ within 12 months than the overall level, and ₹2-10 lakh exposure entities showed 2.1 times higher delinquency (SIDBI MSME Pulse, July 2026, TransUnion CIBIL MSME Pulse July 2026). That is exactly the kind of pocket stress vintage analysis should detect internally.
Roll-Rate Math
Section titled “Roll-Rate Math”Roll rate measures movement from one delinquency bucket to the next. A simple one-month roll:
roll from 30 to 60 = amount that was 30 DPD last month and is 60 DPD this month / amount that was 30 DPD last month
Example:
| Opening bucket at 31 May | Amount | Current in June | Same bucket | Rolled forward | Closed/write-off |
|---|---|---|---|---|---|
| Current | ₹100.0 crore | ₹94.0 crore | Not applicable | ₹5.0 crore to 1-30 | ₹1.0 crore |
| 1-30 DPD | ₹6.0 crore | ₹2.8 crore | ₹1.4 crore | ₹1.5 crore to 31-60 | ₹0.3 crore |
| 31-60 DPD | ₹2.0 crore | ₹0.6 crore | ₹0.5 crore | ₹0.8 crore to 61-90 | ₹0.1 crore |
| 61-90 DPD | ₹1.0 crore | ₹0.2 crore | ₹0.2 crore | ₹0.5 crore to 90+ | ₹0.1 crore |
Rolls:
- Current to 1-30 = ₹5.0 crore / ₹100.0 crore = 5.0%.
- 1-30 to 31-60 = ₹1.5 crore / ₹6.0 crore = 25.0%.
- 31-60 to 61-90 = ₹0.8 crore / ₹2.0 crore = 40.0%.
- 61-90 to 90+ = ₹0.5 crore / ₹1.0 crore = 50.0%.
The cure rate is just as important. From 1-30 DPD, ₹2.8 crore cured to current, so cure rate is 46.7%. If cure rate falls while roll rate rises, collections capacity, borrower stress or sourcing quality has changed. In an EMI product, bucket definitions are straightforward. In cash credit and overdraft, “out of order” status, drawing power breaches and interest servicing complicate the bucket; see delinquency fundamentals.
Bounce Rates
Section titled “Bounce Rates”Bounce rate is usually measured at presentation level and customer level.
presentation bounce rate = bounced presentations / total presentations
amount bounce rate = bounced amount / total presented amount
unique borrower bounce rate = borrowers with at least one bounce / borrowers presented
For SME loans, the first-bounce cohort matters. A borrower who bounces first EMI has a much higher risk than a borrower who bounces EMI 18 after a bank holiday mismatch, but the reason code matters. Insufficient funds, account closed, mandate cancelled and technical failure are not equivalent. A good dashboard separates:
- NACH or eNACH failure reason;
- cheque bounce;
- UPI AutoPay or card mandate failure;
- first EMI bounce versus repeat bounce;
- bounce cured within 3 days, 7 days, 30 days;
- bounce plus bureau new enquiry or new loan.
Collection Efficiency
Section titled “Collection Efficiency”Collection efficiency has several definitions. Use them consistently.
billing collection efficiency = amount collected against current month demand / current month demand
overall collection efficiency = total collections including arrears / total demand including arrears
current bucket collection efficiency = current dues collected from non-delinquent borrowers / current dues billed
arrear collection efficiency = arrear collected / opening arrear demand
Example for July:
| Metric input | Amount |
|---|---|
| Current month EMI demand | ₹8.0 crore |
| Opening arrear demand | ₹1.5 crore |
| Collections against current demand | ₹7.3 crore |
| Collections against arrears | ₹0.6 crore |
| Total collection | ₹7.9 crore |
Billing collection efficiency = ₹7.3 crore / ₹8.0 crore = 91.25%. Overall collection efficiency = ₹7.9 crore / ₹9.5 crore = 83.16%. Arrear collection efficiency = ₹0.6 crore / ₹1.5 crore = 40%. A portfolio can show healthy billing efficiency while arrears rot; that means delinquency stock is not curing.
Concentration Limits
Section titled “Concentration Limits”SME portfolios fail through concentration as much as through individual default. Concentrations to monitor:
- geography: state, district, branch, pin code, industrial cluster;
- channel: DSA, connector, anchor, marketplace, branch, digital campaign;
- industry: textile, gems and jewellery, construction contractors, transport, restaurants;
- product: unsecured BL, LAP, machinery, invoice finance;
- ticket band and tenor;
- assessment method: banking surrogate, GST surrogate, financials-based;
- bureau band and new-to-credit share;
- top borrower group, top supplier/anchor, top employer in a cluster.
Set both exposure limits and performance triggers. For example: no DSA above 8% of monthly disbursal without national credit approval; no district above 12% of unsecured book; pause if MOB 3 ever 30+ exceeds cohort benchmark by 1.5 times; reduce maximum ticket for a GST-surrogate segment if first EMI bounce exceeds 8%.
The Portfolio Meeting
Section titled “The Portfolio Meeting”A useful monthly risk meeting does not review 40 charts. It answers:
- Which new vintages are worse than benchmark?
- Which variables explain the deterioration?
- Are collections curing early buckets?
- Are scorecards separating risk?
- Which policy changes are needed now?
Portfolio analytics must have authority. If risk only reports and business ignores, the book will keep growing into a known weak segment until provisioning arrives. Strong lenders close the loop: source, approve, monitor, collect, learn, and change policy.
For growth planning, Evaluating Growth: The AUM Math and Rational CAGR shows how runoff, vintage seasoning and capital constraints can make headline AUM growth misleading.