Repayment Track Record Analysis: Turning 36 Months of History into a Credit Judgement

Repayment track record analysis is the reading of a borrower's month-by-month payment history — usually 36 months of days-past-due (DPD) values per facility — to form a view on whether they will service the facility you are about to sanction. It is not a score. It is a judgement about pattern, severity, recency and direction.

The raw material is three lines of numbers per tradeline. The output is a paragraph in the credit memo that a sanctioning authority signs against. YuSight's Repayment Tracker is built on a platform that has processed 10 Mn credit journeys, and it reads those three lines the same way an experienced analyst does: frequency first, severity second, recency third, direction last.

Key facts

• Bureau DPD grids in India, the US and the UAE almost universally carry 36 months of history per tradeline. Anything older is either dropped or aggregated into a summary line, so 36 months is the practical horizon for track record work.
• The Reserve Bank of India classifies a standard account as SMA-0 at up to 30 days overdue, SMA-1 at more than 30 and up to 60 days, and SMA-2 at more than 60 and up to 90 days, with NPA classification beyond 90 — and requires that flagging be done as part of the day-end process for the relevant date (RBI/2021-2022/125, 12 November 2021).
• Indian credit institutions report to bureaus fortnightly — as on the 15th and the last day of the month — with 7 calendar days to submit and 5 calendar days for the bureau to ingest (RBI Master Direction – Credit Information Reporting Directions, 2025, 6 January 2025). The most recent month of any DPD grid can therefore be up to roughly 27 days stale.
• The CBUAE Credit Risk Management Standards require licensed financial institutions to run regular reviews of credit exposures with "suitable metrics and early warning indicators" for detecting deteriorating exposures (Article 3.12, CBUAE).
• YuSight's Repayment Tracker reads the DPD grid and the bank statement together, so every repayment claim in the memo is 100% cited back to the month and the source line it came from.

What is repayment track record analysis, and what is it not?

It is the structured reading of payment history across every facility a borrower holds, over the longest window the data supports, to answer three questions:

1. Has this borrower paid on time, and if not, how badly and how often?
2. Is the pattern getting better or worse?
3. Does the pattern have an explanation that survives contact with the rest of the file?

It is not the bureau score. A score compresses 36 months of behaviour into one number optimised for portfolio-level rank ordering. You are underwriting one borrower, not a portfolio, and the compression throws away exactly the information you need — whether the delinquency was one event or a habit, whether it was two years ago or last quarter, and whether it moved in one direction.

It is also not a pass/fail gate. Plenty of good credits have a 60-day event in month 29. Plenty of accounts that have never breached 30 days are deteriorating in front of you.

How do you read a 36-month DPD string?

A DPD grid is a row of monthly values, most recent first, per tradeline. Each cell is the number of days the account was past due as at that reporting month-end.

Read it in four passes, in this order.

Pass 1 — frequency. Count the months with any DPD greater than zero, and divide by the months actually reported. Do not divide by 36 unless all 36 months carry data. XXX or blank means not reported, not paid on time.

Pass 2 — severity. Find the maximum DPD in the window, and note which month it fell in. Then find the maximum in the last 12 months separately. A max of 94 in month 30 and a max of 94 in month 2 are entirely different facts.

Pass 3 — recency. How many months since the last non-zero value? This is the single most predictive element of the grid and the one most often lost when analysts summarise the row as "some past dues".

Pass 4 — direction. Split the window into the recent 12 and the prior 24. Compare frequency and severity across the two halves. A borrower whose delinquency rate has gone from 12% to 42% is a different proposition from one who went from 42% to 12%, even if both average 25% across the window.

Only after those four passes do you go looking for a story.

What do the ageing buckets actually mean?

Buckets are reporting conventions, not risk categories, and they differ by market. Know which convention you are reading before you interpret a number.

Two traps sit in that table.

The first is the off-by-one-bucket problem in the US: a 2 on a US tradeline means 30–59 days late, not 1–30. Reading US and Indian grids in the same session and applying the same mental model will misgrade half the file.

The second is that Indian bureau DPD is a days value, while the SMA classification is a regulatory flag applied by the lender at day-end. They usually agree. When they do not — a 045 sitting next to an SMA-0 flag, say — the disagreement is itself a finding: one of the two was reported from a stale system.

Worked example: two borrowers, same "past due" summary

Both borrowers are mid-market manufacturers. Both hold one term loan of ₹1.20 crore with an EMI of ₹3,45,000. Both bureau reports summarise as "delinquencies present in last 36 months". A summary line treats them identically. They are not remotely alike.

Borrower A — the chronic 30-day payer. Fourteen months in 36 carry a non-zero DPD, none above 28 days.

• Months reported: 36
• Months with DPD > 0: 14 → frequency = 14 ÷ 36 = 38.9%
• Maximum DPD in window: 28 days — never breached SMA-0
• Months since last DPD > 0: 0 (current month is 019)
• Recent 12 months with DPD > 0: 7 → 7 ÷ 12 = 58.3%
• Prior 24 months with DPD > 0: 7 → 7 ÷ 24 = 29.2%
• Direction: deteriorating (58.3% vs 29.2%)

Borrower B — the single 90-day event. Four consecutive months carry DPD, peaking at 94 days, 16 to 19 months ago.

• Months reported: 36
• Months with DPD > 0: 4 → frequency = 4 ÷ 36 = 11.1%
• Maximum DPD in window: 94 days — crossed into NPA territory
• Months since last DPD > 0: 16
• Recent 12 months with DPD > 0: 0 → 0.0%
• Prior 24 months with DPD > 0: 4 → 4 ÷ 24 = 16.7%
• Direction: cured and stable

Now weight for recency. Take the sum of DPD days in each 12-month block and weight the most recent block ×3, the middle ×2, the oldest ×1. (These weights are a house convention for illustration, not a regulatory or industry standard.)

Borrower A:

• Months 1–12: 7 delinquent months averaging 18 days → 7 × 18 = 126 days → 126 × 3 = 378
• Months 13–24: 4 months averaging 14 days → 4 × 14 = 56 days → 56 × 2 = 112
• Months 25–36: 3 months averaging 9 days → 3 × 9 = 27 days → 27 × 1 = 27
• Weighted score = 378 + 112 + 27 = 517

Borrower B:

• Months 1–12: 0 days → 0 × 3 = 0
• Months 13–24: DPD values of 021, 094, 067, 035 → 021 + 094 + 067 + 035 = 217 days → 217 × 2 = 434
• Months 25–36: 0 days → 0
• Weighted score = 0 + 434 + 0 = 434

The borrower who never breached 30 days scores worse — 517 against 434 — because his delinquency is frequent, current and getting more frequent. Borrower B had a bad year, cured it, and has paid 16 consecutive months clean.

Add the cash arithmetic. At a 2% late-payment charge on the EMI, Borrower A has paid 14 × (₹3,45,000 × 2%) = 14 × ₹6,900 = ₹96,600 in penalties over three years — a running cost that tells you the delinquency is structural, not incidental.

The judgement: Borrower A needs a working-capital cycle conversation and probably a covenant on the operating account, because his pattern says the cash is arriving 15 to 20 days after the due date every month. Borrower B needs one question answered — what happened in that window — and if the answer is documented and non-recurring, the event is priced, not declined.

How do you tell seasonality from distress?

Both look like clustered delinquency. The difference is whether the clusters repeat on a calendar.

Plot the delinquent months by calendar month, not by position in the string. If a borrower's non-zero DPD falls in June–August in each of three consecutive years, that is a seasonal working-capital trough, and it should be corroborated elsewhere in the file — a monsoon-exposed agri-input distributor, a school-uniform manufacturer whose collections land in April, a Gulf contractor whose certification cycle stalls over Ramadan and summer.

Distress does not respect the calendar. It shows up as a run of consecutive months that starts once and does not recover, or as clusters whose gaps are shortening.

Three corroborations to demand before you accept a seasonality story:

1. The bank statement shows the same trough in the same months — credits falling away, balance dipping, then recovering. If the statement shows steady inflows in the months the borrower was late, seasonality is not the explanation.
2. The GST or VAT returns show the same revenue shape. See GST return analysis for lending for how to line the two up.
3. The pattern is older than the current facility. Seasonality that only appeared after the last enhancement is leverage, not weather.

How does SMA progression work in India, and why does it matter to a track record?

RBI's clarifications circular sets out the classification steps precisely: SMA-0 up to 30 days overdue, SMA-1 above 30 and up to 60, SMA-2 above 60 and up to 90, and the account moves to NPA beyond 90 days. Classification runs as part of the day-end process, and the classification date is the calendar date for which day-end was run (RBI/2021-2022/125).

Two consequences for track record analysis.

Progression is the signal, not the level. An account that touches SMA-0 and returns to standard within the month is noise. An account that walks SMA-0 → SMA-1 → SMA-2 across three consecutive reporting dates is on a trajectory, and the trajectory is visible a full quarter before the NPA date. If you see that walk anywhere in the 36 months, find out where it stopped and why.

Upgradation is stricter than analysts assume. The same circular clarifies that an NPA may be upgraded to standard only if the entire arrears of interest and principal are paid. A tradeline that shows 091 followed by 000 therefore usually means full clearance, not a partial payment — which makes the cure more meaningful than a naive reading of the grid suggests.

How do you weight old delinquencies against recent ones?

Use an explicit rule and write it into the memo, rather than doing it by feel.

• Months 1–6: treat any non-zero DPD as a current condition. Ask about it directly and get a documented answer.
• Months 7–12: relevant to the recommendation. A 60-day event here is a covenant conversation.
• Months 13–24: relevant to pricing and structure, not usually to the accept/decline. Requires a cure of at least six clean months to be treated as historical.
• Months 25–36: context. Useful for establishing whether the borrower has a habit, and for testing management's account of their own history.

The exception that overrides recency: any 90+ event, at any point in the window, that has not been explained in writing. Age does not launder an unexplained NPA-band event, because you cannot tell whether it cured or was restructured.

What does a clean 36-month record not prove?

More than most memos admit. A run of 36 zeroes is consistent with all of the following:

• The facility never carried real repayment stress. A ₹5 lakh term loan serviced cleanly says nothing about capacity on a ₹5 crore facility.
• The borrower has been paying from borrowings. Clean DPD funded by fresh unsecured debt is a deteriorating credit with a perfect grid. The tell is in the bureau versus bank statement reconciliation — new tradelines opening while operating inflows fall.
• The lender is not reporting. Long runs of XXX are absence of data, not evidence of payment.
• The debt is not on the bureau at all. Informal borrowing, promoter loans, supplier credit and unregulated lenders leave no tradeline and therefore no DPD.
• The account was restructured. A restructuring resets the schedule; the grid goes clean because the terms changed, not because behaviour did. Look for a RES/restructured flag and a change in the EMI value.

This is why track record analysis has to be read alongside the bureau report as a whole and the spread financials, not on its own. Repayment behaviour tells you what the borrower did; the spread tells you whether they could afford it.

How does the analysis differ across India, the US and the UAE?

The practical implication: in India you can build a repayment narrative largely from the bureau, because reporting is dense and regulated. In the US and UAE, thinner commercial files mean the bank statement carries proportionally more of the evidentiary load.

How does YuSight's Repayment Tracker fit into this?

The Repayment Tracker reads the DPD grid for every tradeline and the borrower's bank statements in the same pass, then lines them up month by month. It surfaces the frequency, severity, recency and direction figures automatically, flags SMA-band progressions, and marks every month where a bureau tradeline and the statement disagree about whether a payment happened.

Every figure it produces is traced to its source — the tradeline and reporting month in the bureau file, or the debit line and page in the statement — so the repayment paragraph in the credit assessment memo is verifiable rather than asserted. Analysts edit the judgement; the platform holds the evidence.

FAQ

What is repayment track record analysis?

It is reading a borrower's month-by-month payment history across all their facilities — usually 36 months of days-past-due values — to judge whether they will service a new facility. You are looking at four things: how often they were late, how badly, how recently, and whether it is getting worse.

How many months of repayment history is enough?

Thirty-six months is the standard because that is what bureaus carry, and it is genuinely enough to see a pattern repeat. Twelve months tells you the current state but cannot distinguish a habit from a bad year. Below six months you are not doing track record analysis at all.

How do you weigh old delinquencies against recent ones?

Recency dominates. A 30-day late payment last month matters more than a 60-day event two years ago, because the recent one is a current condition and the old one has had time to cure. The exception is any 90-plus-day event that nobody has explained in writing — age does not fix that.

Is a chronic 30-day payer worse than a borrower with one 90-day event?

Often, yes. Someone who is 20 days late every month has a structural cash timing problem that will not fix itself; someone who missed a quarter two years ago and has paid cleanly since had an event. Score for frequency and recency, not just for the worst number in the row.

What does XXX mean in a DPD grid?

It means no data was reported for that month. It is not a zero. If you count XXX as on-time payment you will overstate the borrower's record, and long recent runs of it usually mean the lender's file is late or the account was sold or written off.

What is the difference between SMA-1 and SMA-2?

They are RBI stress flags on a standard account. SMA-1 means principal or interest is overdue by more than 30 and up to 60 days; SMA-2 means more than 60 and up to 90 days. Past 90 days the account becomes an NPA.

Can a clean repayment record hide a deteriorating borrower?

Yes, and this is the most common failure in the file. A borrower can service everything on time out of fresh borrowings, promoter infusions or a drawn-down overdraft. Check whether new tradelines opened while operating inflows fell.

How stale can a DPD grid be?

In India, up to about 27 days for a fully compliant lender — the event falls in a fortnight ending on the 15th or month-end, the lender has seven calendar days to submit, and the bureau five to ingest. Treat the most recent month as provisional.

Should a restructured account be read as a clean record after the restructure?

No. Restructuring changes the schedule, so the grid goes clean because the terms moved, not because the borrower's behaviour did. Look for the restructure flag and compare the pre- and post-restructure instalment amount before you credit the clean run.

What corroborates a seasonality explanation?

Three things: the bank statement shows the same trough in the same calendar months, the GST or VAT filings show the same revenue shape, and the pattern predates the current facility. If seasonality only appeared after the last enhancement, it is leverage, not weather.

Key takeaways

• Read a DPD grid in four passes — frequency, severity, recency, direction — before you go looking for an explanation.
• Recency-weight the history. A frequent 20-day payer can score worse than a cured 90-day event, and usually should.
• SMA-0 → SMA-1 → SMA-2 progression is visible a quarter before the NPA date. Find where the walk stopped and why.
• XXX is missing data, not a clean month. Divide by months reported, not by 36.
• A clean grid proves the reported facilities were serviced. It does not prove the borrower is solvent, or that all the debt is reported.

Run one borrower through the analyzer — bring a real bureau file and 12 months of statements, and see the repayment narrative built month by month with every figure cited.

Next: how to reconcile bureau obligations against bank statement debits, and how to read a commercial bureau report end to end.