Operations

Bucket movement and flow rates: reading a delinquency matrix instead of a bucket chart

CS Sushil Choudhary · 2026-10-09 · 7 min read (estimated)

Almost every lending MIS carries a bucket chart: so many accounts in 0–30, so many in 31–60, so many at 90+. It is a stock statement, and it has a specific blind spot — two books with identical bucket charts can be heading in opposite directions, and the chart cannot tell them apart.

What distinguishes them is flow: not how many accounts are in 31–60, but how many arrived there from 0–30 versus how many fell back from 61–90, and how many left for current. That is a movement matrix, and it is the only form of this analysis that forecasts.

Why the stock view cannot work

Two books, same month-end chart:

Bucket	Book A	Book B
Current	900	900
1–30	60	60
31–60	25	25
61–90	10	10
90+	5	5

Identical. Now the movement:

Book A: of last month's 1–30, 45 cured to current and 15 worsened. The 31–60 bucket filled from

a smaller inflow than last month.

Book B: of last month's 1–30, 10 cured and 50 worsened. The bucket is the same size because a

fresh wave of current accounts fell into it.

Book A is improving. Book B is in the early part of a deterioration that will reach 90+ in three months. The chart is the same. The books are not.

Building the matrix

You need each account's bucket at two consecutive month ends and nothing else. Rows are the starting bucket, columns the ending bucket, cells the count or value that moved.

From ↓ / To →	Current	1–30	31–60	61–90	90+	Closed
Current	880	55	—	—	—	15
1–30	45	8	15	—	—	2
31–60	6	4	7	12	—	1
61–90	1	—	2	3	8	—
90+	—	—	—	—	14	3

Four properties this has that the chart does not:

Everything above the diagonal is deterioration, everything below is cure, the diagonal is

stasis.

Rows sum to last month's bucket, columns sum to this month's. If they do not, something left

the book without being recorded as closed, and that is a data question not a credit one.

Skips are visible. An account cannot go from Current to 31–60 in one month if the buckets are

monthly — if a cell above the first off-diagonal is populated, your bucketing and your month-end spacing disagree.

Closure is a column, not a disappearance. Foreclosed, written-off and settled accounts leave,

and lumping them with cures is the most common way a matrix flatters.

Roll rate, cure rate, and the forward estimate

Roll rate — the share of a bucket that worsened:

roll rate (1–30) = moved to 31–60 or beyond ÷ opening 1–30

Here: 15 ÷ 70 = 21.4%

Cure rate — the share that improved:

cure rate (1–30) = moved to Current ÷ opening 1–30

Here: 45 ÷ 70 = 64.3%

Net flow into 90+ — the number that actually matters, because 90+ is where classification bites:

inflow to 90+ − outflow from 90+

Here 8 in, 0 cured out, 3 closed = net +5.

And the forward estimate. Chain the roll rates and you get an expectation of 90+ three months out:

1–30 → 31–60 → 61–90 → 90+

With roll rates of 21.4%, then (12 ÷ 29) = 41.4%, then (8 ÷ 11) = 72.7%, today's 1–30 population of 63 projects roughly 63 × 0.214 × 0.414 × 0.727 ≈ 4 accounts reaching 90+ in three months, before any new inflow.

That is a forecast a bucket chart cannot produce at all, and it is built from two month-end snapshots.

The three ways a matrix misleads

Rates computed on a tiny bucket. A 61–90 bucket with 11 accounts gives a roll rate quantised in 9-point steps. Report the denominator next to every rate or the precision is fictional.

Value and count disagreeing. Run both. If the count roll rate is 20% and the value roll rate is 45%, the deteriorating accounts are the large ones — a different problem with a different response, and the count-only matrix hides it entirely.

Restructuring and re-ageing. An account restructured in the month may move to Current without a rupee arriving, and it will sit in the cure column looking like a collections success. Tag them and show them as their own column. A matrix where restructuring is invisible overstates cure by exactly the restructured population.

The retention requirement, which is the actual obstacle

A movement matrix needs each account's bucket as it stood at each month end, retained. Not recomputed.

Asset classification and overdue status are computed in the day-end process for the relevant date (RBI/2021-2022/125, 12 November 2021), and an NPA upgrade requires the entire arrears to be cleared. So a bucket re-derived today from current balances will not reproduce the bucket that existed then — particularly on any account that took a backdated receipt, where the position was computed on a principal the receipt has since changed.

The practical test: re-run last month's matrix. If the row sums no longer match the bucket chart you circulated last month, your positions are being derived and not retained, and every rate above is being computed against a denominator that moves. Why that happens.

Where it sits among the other three

Collection efficiency is a period cash measure — [four definitions, four

answers](/blog/collection-efficiency-how-to-compute/).

Static pool is a cohort credit measure — [how to build

one](/blog/static-pool-analysis-how-to-build/), and the fields a system must freeze first.

DCB ties the borrower's arrears to the ledger — [why it stops

tying](/blog/dcb-reconciliation-why-it-stops-tying/).

Bucket movement is the only one of the four that is predictive, and the only one that

answers "what happens next quarter if nothing changes".

Diligence asks for all four because each defeats a different flattery — what each request is designed to catch.

Frequently asked questions

What does a movement matrix show that a bucket chart does not?

Direction. Two books with identical bucket charts can be moving opposite ways: one where most of the 1-30 population cured and one where it was replaced by a fresh wave falling out of current. The stock view cannot distinguish them; the matrix shows cure below the diagonal and deterioration above it.

How is a roll rate computed?

The share of a bucket's opening population that worsened — accounts that moved to the next bucket or beyond, divided by the opening population of the bucket. Report the denominator alongside it: a roll rate on a bucket of eleven accounts is quantised in nine-point steps, and the decimal places are fictional.

Can roll rates be used to forecast NPA?

Chained across consecutive buckets, they give an expectation of how much of today's early delinquency reaches 90+ in three months, before any new inflow. That is the one genuinely predictive output of the four standard portfolio analyses.

Why must restructured accounts be shown separately?

A restructured account can move to current without a rupee arriving, and it then sits in the cure column looking like a collections success. A matrix where restructuring is invisible overstates cure by exactly the restructured population.

Run your lending on Lenviq

Seeing it run on your own book is faster than reading about it — a demo works through your products, your schemes and your classification rules, not a generic tour.

Lenviq is loan origination, servicing and accounting for NBFCs, built by FastLegal Technologies.

Book a demo What the platform covers

Or just ask+91 96641 46595 WhatsApp hello@lenviq.in What it implements, and the direction each one comes from