Gravitywell
Research
CREDIT RESEARCH
GWR-2026-IN-004
AUGUST 2026
India Non-Bank Credit Series · Regulation

India's Revolving Credit Ban

On 6 Aug 2026 the Reserve Bank proposed deleting the redraw: non-banks may lend term, or not at all. Comments close on the 28th. Nobody can say what is being deleted. No regulator, bureau or rating agency counts revolving exposure at Indian non-banks, and the one industry figure in circulation cannot be found from the bottom up.

0
Independent series counting the affected book ●
13–31%
Range on one lender's flexi share ◐
₹32kcr
Value erased, close to close, 7 Aug ◐
28Aug
Comment window closes ●
Gravitywell Research · GWR-2026-IN-004India's Revolving Credit Ban

ContentsIndia's Revolving Credit Ban

About Gravitywell Research

Gravitywell Research publishes independent research across five pillars — Credit Research, Capital Markets, Industry & Sector Research, Economy & Policy, and Risk & Analytics. Our work is analysis for readers who allocate capital, underwrite risk, or set policy across public and private markets, not regulated ratings or investment advice. We source to primary filings and official statistics, disclose our coverage and confidence, and mark every figure to its point in time.

Conventions & data vintage

Currency: ₹ = INR; 1 lakh crore = ₹1trn ≈ $11.4bn at ₹88/$. Period basis: FY = Apr–Mar, so FY27 is the year to Mar 2027. Figures nominal unless marked real. Data as of 20 Aug 2026. Estimates are ranges, not points, and this report publishes no estimate of the affected book because the inputs do not exist. Draft clause text is quoted from press reproduction of the draft, not from the RBI PDF, which sits behind a bot challenge; those quotations are tiered ◐ accordingly.

Confidence tiers on figures: ● filed/official   ◐ modeled/derived   ○ Gravitywell Research estimate.

How to read: the Executive Summary (p3–5) and the Quick Read (p7) stand alone. Consensus & the Variant (p6) states where this desk departs from the market, and what would prove the market right.

Gravitywell Research02 / 40
Executive Summary · IGWR-2026-IN-004

Executive Summary · IThe thesis

A credit limit is not a loan. It is a standing permission to borrow again, and on 6 Aug 2026 the Reserve Bank proposed to withdraw that permission from almost every non-bank lender in India.1 A finance company may offer credit "in the nature of term loans" and may not offer revolving products at all. Once the principal is repaid, the sanctioned limit "cannot be restored or replenished."2 Only non-banks the Reserve Bank has authorised to issue credit cards are exempt.3 Comments close on 28 Aug 2026, and the draft takes effect immediately once notified.1

The market priced it in a session. Bajaj Finance closed 4.5% lower the next day, erasing about ₹32,000 crore of value, and the sell-side called the impact manageable if lenders redesign the product.411 Our disagreement is not with that estimate. It is that nobody, including the people making the estimate, can say what is being removed.

No regulator, bureau or rating agency counts revolving exposure at Indian non-banks. The number everyone is quoting has no author, no method, and no bottom.Gravitywell Research Global Macro Desk

Three things follow. The ₹2 trillion that anchors the coverage was attributed to unnamed "senior NBFC executives" and carries no definition and no lender coverage. Nor can it be assembled from the bottom up, because every lender publicly sized adds to roughly ₹1 trillion at the outside.528 The one operating series still published is scoped to a single distribution channel and cannot settle anything, in either direction.732 The number that would settle it was published in 2020 and withdrawn in 2021.3537 And the precedents pull both ways. This regulator prohibited a revolving product at non-banks in Jun 2022, immediately and permanently.25 It has also narrowed three drafts since Dec 2024, each with a transition.4042

0
Independent series counting the book ●
₹1trn
Bottom-up total, every named lender ○28
4.1%
Flexi transaction growth vs 24% AUM ●32
3 of 3
Recent drafts this RBI narrowed ◐4042
What this report is, therefore

Not an impact estimate. The inputs for one do not exist, and this report shows why rather than supplying a number that would only inherit the problem. It is an argument about what can and cannot be known before 28 Aug 2026, and about what the right precedent says will happen. It also locates the dilution mechanism with evidence behind it, which is this regulator's own draft-to-final record since Dec 2024 rather than the industry's lobbying.4042

Gravitywell Research03 / 40
Executive Summary · IIGWR-2026-IN-004

Executive Summary · IINine findings

In one line · The rule will probably stick, and the book it removes cannot be measured by anyone.
₹0.71–1.37lakh cr
Bajaj Finance, broker range ◐4
<₹14,300cr
Tata Capital, company-stated ●28
~₹1trn
Every named lender, summed ○
₹2trn+
Claimed, unattributed ◐5

The gap between the third and fourth figures is the report's subject. It is either a long tail of lenders nobody has counted, or an inflated number, and no public series can tell the reader which.

Gravitywell Research04 / 40
Executive Summary · IIIGWR-2026-IN-004

Executive Summary · IIIWhat to do about it

In one line · Assume the rule sticks, discount every sizing including ours, and watch the definition rather than the lobbying.
The allocator: the rule is likelier to stick than to be withdrawn

Price both precedents. This regulator prohibited a revolving product at non-banks in Jun 2022 with immediate effect and never reversed it,25 and it has narrowed three drafts since Dec 2024.4042 Narrowing is modal at 50–60%, and it does not save the consumer product. The Feb 2025 easing usually cited as evidence of retreat did something else. It restored risk weights on bank lending to non-banks and on microfinance, and left the 125% on unsecured consumer credit untouched, where it still sits.24 Bajaj Finance at 5.9 times book is priced for a franchise whose flexi engine nobody has sized and whose own transaction metric is growing at 4%.1132 That is a position to trim into the notification rather than one to add to.

The lender: argue the definition, not the unfairness

Submissions close on 28 Aug 2026. The argument with evidence behind it is structural. The draft already admits bullet repayment as a term-loan form,3 so ask for that permission to be extended into a positive list of surviving facilities, and ask for a transition on the gold-loan template. Both are things this regulator has done in the last eighteen months.40 Disclose the flexi book in the next results presentation, since the estimates circulating in its absence span nearly two times.

The policymaker: the measurement gap is the cheapest thing here to fix

A prohibition is being consulted on without a series that counts what it prohibits. Account-level reporting of drawn against sanctioned would let the regulator detect utilisation-driven evergreening directly. It would also size the affected book before removing it, and show whether the exposure reappears at banks or outside the perimeter. None of those is possible today.

The call

Cautious into the notification, on the precedent rather than on the arithmetic. We publish no impact estimate because the inputs do not exist, and we treat any figure that claims one, including the ₹2 trillion, as an assertion until its author and method are known.

Prior house work on the same terrain: GWR-2026-IN-003 The Exit Window (14 Aug 2026), and GWW-2026-002 The Private Credit Unwind (14 Aug 2026).2021

Gravitywell Research05 / 40
Consensus & the VariantGWR-2026-IN-004

ConsensusWhere we depart, and what would prove us wrong

In one line · The market is estimating an impact; we say the inputs for an estimate do not exist.
TABLE 0.1 · The consensus, the delta, the mechanism, the falsifier
RowContent
The consensusManageable, if lenders redesign, and probably diluted in consultation. IIFL Capital's Viral Shah: the norms "can impact new customer acquisition, growth and customer stickiness, especially for NBFCs with meaningful exposure to such products." PL Capital reads the scope wide — "any reusable credit line, overdraft facility or flexi-loan product." Citi calls the intent "well-founded" while saying demand for revolving credit "may merit deliberation during the consultation period."4 Bernstein dissents on the channel: it argues the primary risk is asset quality rather than growth, because withdrawal "could expose pockets of borrower stress that have so far remained masked by easy access to additional liquidity."44 The sizing everyone works from is a ₹2 trillion figure attributed to unnamed executives.5
The house deltaTwo departures, both testable. First, on what dilution means: we agree narrowing is modal, at 50–60%, and we disagree that it is a reprieve, because the consumer revolving product goes in both likely branches and only fixed-tenor business lending survives. We also keep the as-drafted branch live at 30–40%, on the Jun 2022 prohibition.25 The Feb 2025 episode usually cited as the dilution precedent never touched the consumer risk weight.24 Second, on the sizing: we decline to publish an impact number, because the affected book cannot be located from the bottom up, the one checkable lender estimate is out by two times,28 and the only company metric available is undefined and growing at 4% against a 24% book.32
Why the mispricing persistsBecause a missing series is invisible in a model. An analyst who needs a number takes the only one in circulation, and its absence of provenance never enters the spreadsheet. The same reflex made Feb 2025 the default precedent: it is the most recent RBI reversal anyone remembers, and few have checked which instrument was reversed. The result is visible in the record. No broker has published an estimate cut on Bajaj Finance since the draft; consensus target sat at ₹1,152.69 on 20 Aug, above the ₹1,144.7 pre-draft close, and Jefferies wrote that the fall was "larger than the likely earnings impact".41 The street has priced the rule at approximately nothing.
What proves consensus rightThree observables, any one sufficient. The final Directions arrive with a carve-out for secured and working-capital lines and a transition window. Or Bajaj Finance discloses a flexi book at the top of the broker range with redraw frequency to match, which would mean the transaction metric never measured redraws and the growth engine is intact. Or the Reserve Bank publishes a facility-structure series showing the affected book at or above ₹2 trillion, which would make the sizing sound and this report's central objection wrong.
Where we sit with consensus, not against it

On direction of travel there is no house edge and we do not manufacture one. That displaced credit moves toward banks and the exempt card issuers is the industry's own public submission.5 We also agree the earnings impact is probably absorbable on the growth channel. Our objection is narrower and harder: an absorbable impact and an unmeasurable one are different claims, and the street has been making the first while only the second is supported. Note too that Bernstein's channel, asset quality, is one this report cannot price either, and §04 is the reason it should be taken seriously.44

The call

Trade the odds, not the arithmetic. The distribution of outcomes is wrong in the market's favour on dilution and wrong in nobody's favour on size, and only the first of those is tradeable today.

Gravitywell Research06 / 40
Investor Quick ReadGWR-2026-IN-004

Quick ReadFive calls, five risks

In one line · Roughly even odds the draft is notified as written, and no defensible way to size what it removes.
High-conviction callsAug 2026
Narrowing is modal, and does not save the retail productHigh

50–60% narrowed against 30–40% as drafted. The consumer revolving line goes in both branches; what a narrowed rule saves is fixed-tenor business lending.40

The narrowing lever is already in the draftHigh

The term-loan definition expressly admits bullet repayment.3 That is a product-structure permission inside an instrument otherwise built on a residual test, and it is what a submission should ask to extend.

No credible impact estimate exists yetHigh

The claimed ₹2trn has no author or method,5 every named lender sums to about ₹1trn, and the one broker row checkable against a company statement is out by two times.28

The exemption is an option, not an annuityMedium

It follows a licence status, not a fixed list. In 2022 the affected lender's endgame was to buy a bank and relaunch the product inside the perimeter.26

The de-rating has held, not bouncedMedium

Two weeks on, Bajaj Finance trades around ₹1,096, flat on its 7 Aug close and still well below its 6 Aug mark.11 The market has not decided this is noise.

Key risks
Notification without a transition
It comes into force immediately once notified.1 The 2022 precedent gave none; China's comparable measures gave three years.27
High
0–6m
The affected book cannot be measured
No regulator, bureau or rating agency publishes facility-structure data for non-banks.33 Every impact estimate, including any this desk could build, inherits that void.
High
standing
Our own panel is broker-sourced
The desk's panel failed its one calibration test by about two times against a company statement.28 We publish it with the miss recorded, not repaired.
High
standing
The redesign proves cheap
If repeat pre-approved term lending costs a large digital lender almost nothing, the friction case narrows to reported-growth optics.
Medium
6–12m
Credit leaves the perimeter unobserved
Displaced small-ticket demand can reach lenders that report to nobody, and no series would show it.
Medium
12m+
Thirty-second posture
Allocator

Trim the flexi-heavy compounder into the notification; the odds of dilution are lower than the market assumes.

Lender

Argue the residual definition by 28 Aug, ask for enumeration and a transition, and disclose your own flexi book.

Policymaker

Collect drawn against sanctioned before removing the facility it describes.

Gravitywell Research07 / 40
Part I · The Clause
Sections 01 — 04

The
Clause

Four pages on what the draft actually says, which products stop being legal, who is exempt, and why the Reserve Bank thinks the redraw is the problem.

Reading
Pages 08 — 11
Key figure
28 Aug 2026
What you'll take away
01
The test is replenishment, not product name
Anything that restores a limit on repayment is caught, however it is marketed.
02
The exemption is a licence, not a category
It follows credit-card authorisation, which two non-banks hold.
03
The regulator's case is evergreening
It is a better argument than the industry's answer to it.
Part I · The ClauseGWR-2026-IN-004

§ 01What the draft actually says

Takeaway · The permitted side of the definition admits bullet repayment — which decides what actually survives.

The draft does its work through a definition, and the definition is more permissive than the coverage suggests. A term loan is a facility of fixed principal, disbursed in one or more instalments and "repayable in accordance with a predetermined amortisation schedule, either as periodic instalments or as a bullet payment."3 Once repaid, the limit "cannot be restored or replenished."2 Revolving credit is the residual: anything failing that test. The prohibition is one line — a non-bank "shall only offer credit products which are in the nature of term loans and shall not offer any revolving credit products."2

TABLE 1.1 · The draft, clause by clause
ClauseWhat it doesBasis
Definition · term loanFixed principal, disbursed in one or more instalments, "repayable in accordance with a predetermined amortisation schedule, either as periodic instalments or as a bullet payment"; the limit does not come back on repayment3
Definition · revolving creditAny fund-based facility that is not a term loan2
ProhibitionTerm loans only; no revolving products at all2
ExemptionNon-banks authorised to issue credit cards may keep revolving credit through the card2
ConsequentialDeletes a Chapter II provision and removes the separate chapter on Demand/Call Loans, replacing them with "Restrictions on Revolving Credit Facilities"2
CommencementComes into force with immediate effect if notified in the present form; no transition period is written into the draft2
ConsultationComments to 28 Aug 2026 via Connect 2 Regulate or [email protected]1
FIG 1.1 · Three years of narrowing the non-bank product setinstrument · date
SIX STEPS, NOVEMBER 2023 TO AUGUST 2026
Risk weightsNov 2023Digital lending2024–25Partial rollbackFeb 2025Credit FacilitiesNov 2025Revolving ban6 Aug 2026Comments close28 Aug 2026
Each step narrowed what a non-bank may sell. The draft is the first to remove a product outright.
Source: RBI draft notifications and circulars ●114; clause text via press ◐23.
The call

Read the definition, not the headline, and read all of it. The residual test stops any lender renaming its way out. But the permitted side already admits bullet repayment, which is a product-structure permission rather than a pure residual, and it decides whether a fixed-tenor tranche with one maturity survives.

Gravitywell Research08 / 40
Part I · The ClauseGWR-2026-IN-004

§ 02Inside the perimeter

Takeaway · The rule reaches business lending far more than the coverage suggests, because working capital is revolving by construction.

The products caught by the draft are easier to list than the ones that survive, because the test is mechanical. If repayment restores the borrower's ability to draw, the facility is revolving. That sweeps in the consumer products the newspapers wrote about. More consequentially, it sweeps in the business facilities they did not. That means supply-chain and channel finance, drop-line overdrafts against property, and the working-capital lines a small manufacturer uses between paying for material and being paid for output.5

FIG 2.1 · Which facilities fail the term-loan test
THE MECHANICAL TEST: DOES REPAYMENT RESTORE THE LIMIT?
statusFlexi / hybrid personal loanCAUGHT — limit replenishesconsumerDrop-line overdraft, LAPCAUGHT — limit replenishessecuredMSME working-capital lineCAUGHT — revolving by designbusinessSupply-chain tranche, bulletSURVIVES — bullet is a term loanbusinessLoan against securitiesCAUGHT on the drafted testwealthEMI card, reusable limitCAUGHT unless card-authorisedconsumer
Bullet repayment is expressly admitted as a term-loan form, which is why a fixed-tenor tranche survives a rule aimed at revolving credit.
Source: Gravitywell Research classification against the drafted definitions ○2; segment list per FIDC submission scope ◐5. (● filed · ◐ modeled · ○ est.)

The product family matters here, and it is finer than the coverage suggests. Bajaj Finance's own description separates three things: "You could get a term loan. You could get a drop line flexi or you could take a flexi. Flexi is nothing but like a line, you could use it by drawing down and paying down."36 Plain flexi restores the limit on repayment and is unambiguously caught. Drop-line, where the limit steps down rather than replenishing, is the genuine open question, and its answer decides how much of the book is in scope.

The permitted side of the definition does real work. Supply-chain finance is typically written as discrete 30 to 180-day tranches with a single maturity, and a bullet-repayment tranche is a term loan on the draft's own wording.3 A block of what the industry counts as affected may never have been in scope, and the same question runs to gold overdrafts and loan against securities.

₹1.10L cr
Urban personal loans ●7
₹51,320cr
MSME lending ●7
₹36,564cr
Loan against securities ●7
₹45,220cr
Urban consumer finance ●7
The call

The consumer story is the visible half. If you are underwriting the credit consequences of this rule, the segment to watch is small-business working capital, where the revolving structure is not a convenience feature but the way the product works, and where the substitute, a term loan drawn each time material is bought, is a worse instrument for the borrower and a costlier one for the lender.

Gravitywell Research09 / 40
Part I · The ClauseGWR-2026-IN-004

§ 03The exemption, and the line it draws

Takeaway · Revolving credit is not being retired; it is being licensed.

The draft does not say revolving credit is unsafe. It says revolving credit is for entities the Reserve Bank has authorised to issue credit cards, which among non-banks means SBI Card and BoB Financial.3 That is a defensible position, since card issuance carries its own conduct rules, disclosure standards and billing discipline. It is also a competitive act, because it hands one product to two firms and takes it from everyone else in the same regulatory class.

₹58,269cr
SBI Card receivables, Jun 26 ◐8
2.26cr
Cards in force, +7% YoY ◐8
18.6%
Share of cards in force ◐8
19.5%
Share of card spends ◐8

The gift is worth less than it looks, and worth something. SBI Card's receivables grew 3% over the year to Jun 2026 while its spends grew 27% and its share of spending rose from 16.6% to 19.5%.8 A card business with flat balances and compounding spend is a payments business rather than a lending one. Customers are transacting and settling rather than revolving. Removing the flexi product will not make a settling customer borrow. It does remove the alternative for the ones who would.

FIG 3.1 · The exempt franchise is growing spends, not balances% YoY, year to Jun 2026
THE EXEMPT LENDER IS NOT THE FAST ONE
01020303%ReceivablesSBI Card7%Cards in forceSBI Card27%SpendsSBI Card24%AUMBajaj Finance
A card book whose balances are flat while spending compounds is a payments business rather than a lending one.
Source: SBI Cards Q1 FY27 results ◐8; Bajaj Finance Q1 FY27 investor presentation ●7. (● filed · ◐ modeled · ○ est.)
The call

Own the exemption on relative terms, not absolute ones. A licensing boundary that removes the only non-bank competitor is worth holding, and it does not turn a 3% receivables grower into a compounder. The trade is the spread between the two, and the binding constraint is underwriting capacity rather than addressable book.

Gravitywell Research10 / 40
Part I · The ClauseGWR-2026-IN-004

§ 04The regulator's case, at its strongest

Takeaway · Evergreening is a real defect of the product, and the industry has not answered it.

State the Reserve Bank's case properly and it is stronger than the industry's reply. Take a borrower with a ₹1 lakh limit who has drawn ₹50,000. The unused half can be drawn again and used to pay the interest or principal that has fallen due. The account never ages, the loan is never classified, and the lender's book shows a performing exposure to a borrower who has stopped generating cash.6 That is evergreening, and a revolving limit is the instrument that makes it invisible.

FIG 4.1 · How the undrawn half hides the drawn half
SANCTIONED LIMIT ₹1,00,000 — THE SAME FACILITY, THREE MOMENTS
MONTH 1 · DRAWNDrawn ₹50,000 of a ₹1,00,000 limitBorrower is servicing from cash flowClassified: standard assetMONTH 4 · DUES FALLDrawn ₹58,000; cash flow has stoppedUndrawn half still availableClassified: standard assetMONTH 5 · REDRAWNDrawn ₹70,000 — the dues were metPaid from the lender's own limitClassified: still a standard asset
Classification never moves because the account never falls overdue. Only utilisation rises, and no Indian series collects it. Figures illustrative.
Source: mechanism per coverage of the RBI draft ◐6; Gravitywell Research illustration ○, figures illustrative. (● filed · ◐ modeled · ○ est.)
TABLE 4.1 · Four strands of the case, and the industry's answer
Regulator's concernIndustry answerDesk verdict
Evergreening — undrawn limit repays drawn balance6Not addressed directly in the public submissionUnanswered
Information asymmetry — non-banks see the line, banks see the current account6Account aggregator and bureau data have closed much of the gapPartly
Underwriting outside the card framework6The answer is to extend the framework, not to ban the productFair point
The regulator's best witness is the largest affected lender

In 2020 Bajaj Finance's managing director described flexi on the record as carrying "loan losses being marginally higher compared to term loan due to slower amortization", priced 25 to 50 basis points above term, with an annual maintenance charge of 25 to 100 basis points, and "distinctly more profitable" — adding, "I would like 100% of these clients to get flexi products because I get more fees."36 A lender conceding higher loss rates on the product being prohibited for delayed loss recognition is the strongest evidence in the file for the regulator, and it relocates the earnings exposure from margin to fees.

The call

The industry will lose this if it keeps answering a stability concern with a competition complaint. The reply that works is a recognition rule catching a redraw used to service a due: it reaches the evergreening and leaves the product alone.

Gravitywell Research11 / 40
Part II · The Book in Scope
Sections 05 — 08

The Book in
Scope

How big the affected book is, who carries it, and why the only published number comes from the party with the most to lose.

Reading
Pages 12 — 15
Key figure
₹2trn+, unverified
What you'll take away
01
No one counts this book
Not the regulator, not the bureaus, not the rating agencies.
02
The exposure is concentrated
One franchise carries most of the listed-market risk.
03
The borrower is a small business
That is who pays for the substitute product.
Part II · The Book in ScopeGWR-2026-IN-004

§ 05The sector the rule lands on

Takeaway · Non-banks are gaining share on banks, and this rule arrives at the point of maximum divergence.

Timing matters more than size here. India's non-bank lenders are on track to pass ₹50 lakh crore of assets under management by Mar 2027. Their credit grew about 17% in the first half of FY26, against roughly 12% at banks.16 The gap is the context for the draft. One class of lender is taking share in unsecured and small-business credit, funded increasingly by the banks the same regulator supervises. That is a reason to standardise the product set.

FIG 5.1 · Non-banks are growing faster than the system that funds them% YoY
THE SHARE SHIFT THE DRAFT ARRIVES INTO
010203017%NBFC creditH1 FY2612%Bank creditH1 FY2615.9%Bank creditFY2626%Bank lending to NBFCsFY26
Bank and non-bank series are compiled on different bases; read the gap as directional.
Source: India Ratings and sector trackers on FY26 credit growth ◐16. Bank and non-bank series are compiled on different bases; treat as directional. (● filed · ◐ modeled · ○ est.)
TABLE 5.1 · The scale the rule applies to
MeasureValueAs ofTier
Non-bank assets under management, sector~₹50 lakh crMar 2027E16
Revolving product in scope, industry estimate₹2trn+Aug 20265
In-scope book as share of sector assets, if the claim holds~4–5%Aug 2026○ inherits the anchor's defect5
Growth of the in-scope book, as claimed15–20% p.a.Aug 2026◐ same source, no method5
Bajaj Finance consolidated assets₹5,46,944 cr30 Jun 202679
Bajaj Finance share of sector assets~11%Jun 2026716
The call

Four to five per cent of sector assets is not a financial-stability number, and that is the point worth holding onto. This rule is not being written because the revolving book is large. It is being written because of what the product does to loan classification. A measure aimed at classification integrity cannot be bargained down by proving the book is small.

Gravitywell Research12 / 40
Part II · The Book in ScopeGWR-2026-IN-004

§ 06Sizing a book nobody counts

Takeaway · Every named lender sums to about ₹1trn against a claimed ₹2trn, and no series exists to settle it.

Start with what the ₹2 trillion actually is. Business Standard attributes it to "senior NBFC executives", not to the trade body often credited with it, whose chief executive is quoted only on process.5 It arrives with no definition, no lender coverage and no as-of date. That is a reason to test it, not to ignore it.

FIG 6.1 · Everything anyone has actually sized, against the claim₹ trillion
THE GAP BETWEEN WHAT IS SIZED AND WHAT IS CLAIMED
Bajaj Finance, broker range₹0.71–1.37trnTata Capital, company-stated<₹0.14trnCholamandalam~₹0.03trnL&T Finance, Poonawalla~nilThe claim, unattributed₹2trn+
Roughly ₹1trn of retail and MSME revolving credit would have to sit at lenders nobody has named or sized.
Source: broker estimates via Business Standard, 7 Aug 2026 ◐429; Tata Capital per its chief executive, 11 Aug 2026 ●28; the claim per Business Standard, 17 Aug 2026 ◐5. (● filed · ◐ modeled · ○ est.)

Two readings survive that chart and the report cannot choose between them. Either the figure is dominated by a long tail of unlisted lenders nobody has counted, which is plausible when around 90% of the affected lending is said to be MSME and individual, or it is inflated. The direction of the definitional error is unknown too: supply-chain finance is usually written as discrete 30 to 180-day tranches that pass the term-loan test, while the residual definition catches gold overdrafts, demand loans and drawing power nobody put in the total.2

The calibration test, and the bottom-up cross-check

On 7 Aug brokers put Tata Capital's revolving exposure at high single to low double digits; on 11 Aug its chief executive said below 5%.428 That is the only row where a company statement exists to check the method, and the method was out by about two times in the direction that flatters the story. Against that, a bottom-up build from Bajaj Finance's own 2020 penetration guidance — loan against securities and gold loans described as flexi by construction, salaried personal loans about 90%, property and lease-rental around 60%, small-business around 65%36 — applied to today's segment assets lands near 23% of consolidated AUM. Loan against securities and gold alone are 10.6%. The consensus 13–15% sits close to the floor of that build rather than at its centre. We adopt neither. The point is that the estimates disagree in both directions. And one large lender is missing from every estimate: Shriram Finance, which is seeking a meeting with the Reserve Bank on this draft alongside Bajaj Finance and Tata Capital, has never been sized by anyone.43

The negative finding, stated plainly

No independent series counts revolving exposure at Indian non-banks. The Reserve Bank's sectoral data covers banks by sector; its Trend and Progress covers non-banks by category; TransUnion CIBIL and CRIF High Mark publish product categories, not facility structures.33 No rating agency published a sizing in the fortnight after the draft, and the Reserve Bank's own supervisors had to write to non-banks in Nov 2024 for supply-chain finance details.34 A prohibition is being consulted on without a count of what it prohibits.

Gravitywell Research13 / 40
Part II · The Book in ScopeGWR-2026-IN-004

§ 07The number the company stopped publishing

Takeaway · Bajaj Finance disclosed its flexi book twice in 2020, at 25% and 31% of assets, then closed the disclosure.

The press has reported throughout this episode that Bajaj Finance has never disclosed the size of its flexi book. That is not correct. It disclosed the figure twice, in 2020, and then stopped. The Q1 FY21 presentation filed with the exchanges put the existing flexi book at ₹36,846 crore as at 31 Mar 2020, a quarter of assets under management.35 On the Q2 FY21 call in Oct 2020 the managing director put total outstanding flexi at around ₹43,000 crore, "across mortgages to personal loans to gold loans, LAS" — about 31% of the book.36

TABLE 7.1 · What Bajaj Finance said, and when it stopped saying it
DateVenueFlexi bookShare of AUMWhat was said
21 Jul 2020Q1 FY21 presentation, filed₹36,846cr25.0%"the existing book under flexi stood at ₹36,846 crore"
21 Oct 2020Q2 FY21 earnings call~₹43,000cr~31.4%"across mortgages to personal loans to gold loans, LAS"
20 Jan 2021Q3 FY21 earnings calln/an/a"I think flexi is history now"; on tracking conversions, "No way. We are not tracking even that"
20 Jul 2021Q1 FY22 earnings calln/an/a"probably today is the last time I am responding to a direct question on flexi"
SinceEvery presentation and callnot disclosedIncluding through a draft rule that would prohibit the product
The distinction nobody has priced: drawn against sanctioned

On the same 2020 call the managing director gave the utilisation rate: flexi lines run "between 60% and 85% of the approved line."36 So ₹43,000 crore was a drawn balance, and the sanctioned lines behind it were ₹50,600–71,700 crore. Every broker estimate in circulation is a drawn figure too. But the re-papering, consent and acquisition cost this rule imposes runs against sanctioned limits, a base up to two-thirds larger. The debate is currently sizing the wrong quantity.

Two cautions on using the 2020 marks. The book has gone from ₹1.47 lakh crore to ₹5.47 lakh crore since, and consolidating Bajaj Housing Finance added a mortgage block that mechanically dilutes the flexi share. Neither 25% nor 31% transfers forward, and we do not claim they do. What survives is a fact about the company rather than the book: the last number it put its own name to was about 31%, and it closed the disclosure four months after opening it.

The call

Put the question to management, and put it in the right units. Not "what is the impact" but "what is the sanctioned flexi limit, drawn and undrawn, and why did you stop publishing a number you published in 2020?" On the current record the market is arguing over four broker inferences about a quantity the company itself has measured and declines to restate.

Gravitywell Research14 / 40
Part II · The Book in ScopeGWR-2026-IN-004

§ 08The metric everyone is quoting is a channel metric

Takeaway · Flexi transactions are an App-and-Web throughput line, and they cannot carry an argument in either direction.

One operating number for this product is still published, and it is being read as something it is not. Bajaj Finance reports "flexi loan transactions" once per deck, on a slide headed Omnipresence metrics, inside a block titled App + Web business metrics, between EMI cards acquired and DMS receipts.732 It is distribution-channel throughput for one channel. It is not a count of redraws across the franchise, and the company has never presented it as one.

FIG 8.1 · What the channel series actually showsflexi loan transactions, mn
SIX QUARTERS — AND THE YEAR-ON-YEAR READS DIFFERENTLY EACH TIME
012342.84Q4 FY252.96Q3 FY252.90Q1 FY263.00Q2 FY263.10Q3 FY263.10Q4 FY26
Q3 on Q3 +5% · Q4 on Q4 +9% · Q1 on Q1 +4.1% · FY25 11.62mn to FY26 12.10mn, +4.1%. Reported under App + Web business metrics, so this is channel throughput and not a measure of the book.
Source: Bajaj Finance investor presentations, Q3 FY26 p.63, Q4 FY26 p.75 and Q1 FY27 p.67 ●732. Q1 FY27 is 3.02mn. All four year-on-year comparisons shown. (● filed · ◐ modeled · ○ est.)

An earlier version of this report indexed this line against consolidated assets and called the difference a twenty-point wedge held across eight quarters. That was wrong three times over. The numerator is one distribution channel and the denominator is a whole company, a third of which is mortgages. There are six quarterly observations, not eight. And the year-on-year comparisons range from 4.1% to 9% depending on which quarter you pick, so a single "about 4%" reading is a selection, not a result.

What it can and cannot support

It cannot support a cost multiplier, because it is not a count of redraws and the company does not define the term. It cannot support the claim that the redraw engine is decaying, because a channel line says nothing about the book. What it does support is narrower and still useful: on the one channel the company reports, flexi activity is growing in mid-single digits while the balance sheet compounds in the mid-twenties. That is a question worth putting to management, not an answer.

The call

Distrust every use of this number, including the one this desk made. It is the only operating series in public and it is scoped to a channel, which is precisely the condition that produces confident wrong readings. The number that would settle the argument is the sanctioned flexi limit, and the company stopped publishing it in 2021.

Gravitywell Research15 / 40
Part III · The Economics
Sections 09 — 12

The
Economics

What a limit earns that a loan does not, why a term book has to be re-sold to stand still, and what the difference is worth in a valuation.

Reading
Pages 16 — 19
Key figure
33.4% cost-to-income
What you'll take away
01
A limit earns fees a loan cannot
Availability is a priced service, and the draft removes the thing being priced.
02
Growth changes character
Stock compounds itself; flow has to be bought.
03
The valuation swing sits in the growth line,
which is the input nobody can verify.
Part III · The EconomicsGWR-2026-IN-004

§ 09What a limit earns that a loan does not

Takeaway · A line sells availability; a term loan can only sell money.

A revolving facility has three revenue lines where a term loan has one. It earns interest on the drawn balance, as any loan does. It earns fees for keeping the undrawn portion available, which is why Bajaj Finance charges an annual maintenance fee on flexi facilities. And it earns the option value of a customer who stays on the book between borrowings, because the relationship outlives the balance. Non-interest income at Bajaj Finance was ₹2,653 crore in the Jun 2026 quarter against ₹2,232 crore a year earlier. It grows with the franchise rather than with the balance.7

FIG 9.1 · Three revenue lines against one
WHAT THE LENDER EARNS, LINE BY LINE, UNDER EACH STRUCTURE
REVOLVING FACILITYInterest on the drawn balanceFees for keeping the undrawn limit availableRetention value of a live relationship at zero balanceRedraw at no acquisition costTERM LOANInterest on the disbursed principalProcessing fee, once per originationRelationship ends at repayment unless re-soldEvery repeat is a fresh acquisition
Fee income on an availability product does not survive the product.
Source: Gravitywell Research, from product terms and Bajaj Finance disclosure of flexi fee structure ○7. Schematic, not to scale. (● filed · ◐ modeled · ○ est.)
TABLE 9.1 · Where the income sits at a flexi-heavy lender
LineQ1 FY27Q1 FY26Change
Net interest income (₹ cr)12,57110,228+23%
Non-interest income (₹ cr)2,6532,232+19%
Net total income (₹ cr)15,22412,460+22%
Operating expenses (₹ cr)5,0874,124+23%
Cost to net total income33.4%33.1%+30bp
The call

Watch the non-interest line, not the margin. Fee income on an availability product dies with the product. It is the line where a term-only rule shows up first and most cleanly. If flexi fees are material and undisclosed, the first full quarter after notification will disclose them by their absence.

Gravitywell Research16 / 40
Part III · The EconomicsGWR-2026-IN-004

§ 10Limit stock against origination flow

Takeaway · A limit book defends itself; a term book has to be re-sold every year just to stand still.

This is the mechanism the market is not pricing, and it is arithmetic rather than opinion. A book of limits carries its own replacement. When a borrower repays, the limit stays open and the same customer can restore the balance without the lender selling anything. A book of term loans amortises to zero. The lender must originate a volume equal to that run-off before a single rupee of growth appears. The draft converts the first kind of book into the second.

FIG 10.1 · The same 20% growth, bought two different waysindexed, ₹100 opening book
TO GROW A BOOK 20%, HOW MUCH MUST BE SOLD?
0255075₹20Limit book₹60Term book
On a two-year average term book, 40% of the opening balance runs off each year and must be originated again before a single rupee of growth appears.
Source: Gravitywell Research illustration ○. Run-off assumption stated on the chart; the ratio scales with tenor, not with the size of the book. (● filed · ◐ modeled · ○ est.)

The multiplier is set by tenor, and short-tenor lending is where the flexi book lives. On a two-year average life, growing a term book by a fifth takes three times the origination that growing a limit book by a fifth takes. That is why the guidance line matters more than the margin line. Nothing here changes what a rupee of assets earns. Everything here changes how many rupees have to be sold to get there.

The call

That is the mechanism in principle, and no public series tells you how hard it is running at any particular lender. §08 shows why the one operating number in circulation cannot answer it. Hold the guidance rather than the quarterly print, and treat any confident claim about redraw intensity, including one this desk previously made, as unsupported.

Gravitywell Research17 / 40
Part III · The EconomicsGWR-2026-IN-004

§ 11The cost of buying the same customer twice

Takeaway · Even on the harshest assumption the visible cost is 120 basis points, which is why the market shrugged.

This page is conditional, and says so at the top. It asks what the re-origination burden would cost if the flexi transaction count were a count of redraws. §08 shows it is not, because it is channel throughput, so what follows is a bound on a hypothetical. Cost per origination is not disclosed by any Indian non-bank either. What is disclosed is the aggregate: operating expenses of ₹5,087 crore against net total income of ₹15,224 crore in the Jun 2026 quarter, a cost-to-income ratio of 33.4%.7 On the redraw reading, 3.02mn events would lift the origination workload by roughly a fifth against 16.13mn loans booked.7

TABLE 11.1 · What the extra origination load does, on three assumptions about marginal cost
Marginal cost of a repeat originationExtra eventsAdded quarterly opexCost-to-incomeChange
Near zero — fully digital, pre-approved, no incremental servicing3.02mn~₹0cr33.4%unchanged
₹100 per event3.02mn₹30cr33.6%+20bp
₹300 per event3.02mn₹91cr34.0%+60bp
₹600 per event3.02mn₹181cr34.6%+120bp
Marginal-cost assumptions are the desk's, not disclosed by any lender ○. The point of the table is its shape. Even at ₹600 an event the earnings damage is contained, which is why the market calls it manageable. The damage is in what the same cost does to the pace of growth the franchise can afford.
FIG 11.1 · Why the cost line reads as manageablecost to net total income, %
THE VISIBLE COST IS SMALL — 120bp AT THE HARSHEST ASSUMPTION
01020304033.4today33.6₹100/event34.0₹300/event34.6₹600/event
Conditional on the flexi transaction count being a count of redraws, which §08 shows it is not.
Source: Bajaj Finance Q1 FY27 disclosure ●7; marginal-cost scenarios by Gravitywell Research ○. (● filed · ◐ modeled · ○ est.)
The call

Read the table as a bound, not a forecast. Its top row is the honest one: at a marginal cost near zero, which a large digital lender may well achieve, nothing happens to the cost line at all. If you run a lending book, build the cost per repeat origination yourself — no filing will give it to you.

Gravitywell Research18 / 40
Part III · The EconomicsGWR-2026-IN-004

§ 12Sensitivities: what moves the fair value

Takeaway · The de-rating held for ten sessions while the street cut nothing — the price moved and the estimates did not.

Bajaj Finance traded at ₹1,096 and about 5.9 times book on 20 Aug 2026, two weeks after the draft and effectively flat on its 7 Aug close of ₹1,093.11 The de-rating has held rather than bounced, which is itself information. Work backwards through the standard identity that links a bank-like multiple to its returns, and that price embeds a durable growth rate of about 11.6% at a 20% return on equity and a 13% cost of equity. The precision is spurious; the gearing is the point. At these levels the multiple is a bet on the growth term, and the growth term is what a term-only rule taxes.

TABLE 12.1 · Implied multiple as durable growth moves, ROE held at 20% and cost of equity at 13%
Durable growthImplied P/BChange vs todayWhat it would mean
11.6%, the rate embedded today5.9×Draft withdrawn or narrowed to nothing
11.1%4.7×−20%Definition narrowed; consumer flexi still re-papered
10.6%3.9×−34%Notified broadly as drafted, with a transition
9.6%3.1×−48%Notified with immediate effect, no carve-out

The identity is P/B = (ROE − g) ÷ (COE − g). It is unstable when g approaches the cost of equity, which is exactly the region a 5.9× multiple sits in, so read the direction and the order of magnitude rather than the decimal. A reader who prefers a 14% cost of equity gets a lower implied growth rate and a shallower slope; the sign does not change.

FIG 12.1 · The price moved, the estimates did not₹ per share, sell-side targets
A 60% SPREAD, AND NOT ONE ESTIMATE CUT SINCE THE DRAFT
₹815low₹850Macquarie₹910UBS₹1,096spot 20 Aug₹1,270HSBC₹1,300high
Targets predate the 6 Aug draft and none has been publicly restated. Consensus sat at ₹1,152.69 on 20 Aug, above the ₹1,144.7 pre-draft close.
Source: target dispersion as reported by Business Today, 31 Jul 2026 ◐10; spot price and book multiple as at 20 Aug 2026 ◐11. Targets predate the 6 Aug draft and none has been restated publicly since. (● filed · ◐ modeled · ○ est.)
The call

If you own this at 5.9 times book, you are not being paid for the rule risk. And the scenario carrying most of that downside is the one we think the market is under-weighting, because it is reading the wrong precedent. That combination is worth a hedge rather than a short.

Gravitywell Research19 / 40
Part IV · Where the Credit Goes
Sections 13 — 17

Where the Credit
Goes

Demand for a line does not end when one class of lender is barred from selling it. Four destinations, in order of how well lit they are.

Reading
Pages 20 — 24
Key figure
$12.4bn private credit
What you'll take away
01
Banks are the first destination and the least interesting
They keep the overdraft, and they always did.
02
The redesign is legal and partial
Repeat term lending survives; the self-restoring limit does not.
03
The last destination is the one that should worry the regulator
Credit that leaves the perimeter stops being counted.
Part IV · Where the Credit GoesGWR-2026-IN-004

§ 13Banks keep the overdraft

Takeaway · Eight months before proposing to bar the product at non-banks, the Reserve Bank widened it at banks.

The industry's arbitrage complaint is usually asserted. It can be dated. On 11 Dec 2025, after a draft floated on 1 Oct, the Reserve Bank issued revised directions on current, cash-credit and overdraft accounts, effective 1 Apr 2026. They let banks provide cash credit "without restriction" and lifted to ₹10 crore the exposure threshold below which any bank may run a customer's current or overdraft account.23 Eight months later the same regulator proposed removing the equivalent product from every non-bank without a card licence.1 Note what that sequence also shows: the Reserve Bank took a draft to a final and dropped the restriction on cash credit along the way.

FIG 13.1 · Two directions, eight months apart, opposite signs
THE SAME PRODUCT, TWO REGULATORY CLASSES, EIGHT MONTHS APART
BANKS · 11 Dec 2025, live 1 Apr 2026Cash credit permitted without restrictionCurrent and OD accounts free below ₹10cr exposureSix-monthly review and core-system flagging addedNet effect: wider access, tighter monitoringNON-BANKS · 6 Aug 2026, draftRevolving credit prohibited outrightTerm loans only; the limit cannot be replenishedExemption only for card-licensed entitiesNet effect: the product is withdrawn
Read charitably the two are consistent, because banks hold the current account and see cash flows a non-bank cannot.
Source: RBI revised directions on current, cash credit and overdraft accounts, 11 Dec 2025 ◐23; RBI draft amendment, 6 Aug 2026 ●1. (● filed · ◐ modeled · ○ est.)

Read charitably, the two are consistent. Banks hold the borrower's current account, so they see cash flows a non-bank cannot, and the Dec 2025 directions added six-monthly reviews and system flagging to make that visibility bite.23 That is the information-asymmetry argument stated as policy rather than as commentary. Read commercially, it is a transfer. Bank credit grew 15.9% in FY26 and bank lending to non-banks grew 26%,16 so the sector that would receive this book is both willing and already funding the sector that would lose it.

The call

The arbitrage is real, dated and defensible, which is a harder combination for the industry than a simple unfairness. If you are drafting a submission, argue the asymmetry of information rather than the asymmetry of treatment: the December directions concede that the current-account view is what justifies the difference, and that is a condition a non-bank can meet through account aggregator consent rather than a status it can never acquire.

Gravitywell Research20 / 40
Part IV · Where the Credit GoesGWR-2026-IN-004

§ 14What the exemption is actually worth

Takeaway · One lender's flexi exposure is larger than the entire book of the franchise meant to receive it.

Size the receiving vessel, since the displaced book cannot be sized. A borrower wanting a revolving facility after this rule has exactly two non-bank places to go, and SBI Card's entire receivables book is ₹58,269 crore.8 Bajaj Finance's flexi exposure alone runs ₹71,000 crore to ₹1,37,000 crore on the broker range, and higher on the company's own 2020 marks.435 The exempt franchise is smaller than the thing that would flow toward it.

FIG 14.1 · One lender's flexi exposure against the exempt franchise₹ '000 crore
THE RECEIVING VESSEL IS SMALLER THAN ONE LENDER'S EXPOSURE
050100150₹58k crSBI Card bookreceivables, Jun 26₹71k crBFL flexi, lowbroker floor₹137k crBFL flexi, highbroker ceiling₹1.7k crOne year of growthat 3% receivables
The binding constraint is underwriting capacity rather than addressable book.
Source: SBI Cards Q1 FY27 ◐8; Bajaj Finance flexi range from broker estimates ◐429. One year of growth at the disclosed 3% receivables growth ◐. (● filed · ◐ modeled · ○ est.)

The binding constraint is underwriting capacity and risk appetite rather than demand. A card issuer cannot absorb a flexi borrower without underwriting them as a card customer. That borrower is often exactly the profile a card issuer has been declining. So the realistic transfer is a fraction of the displaced book, concentrated at its prime end. The rest moves to a bank overdraft, converts to term borrowing, or goes unserved.

The call

The exempt issuers are a relative call and a slow one. Own them against the flexi-heavy lenders rather than outright. Size the position on underwriting capacity, because the credit filter binds and the addressable book does not.

Gravitywell Research21 / 40
Part IV · Where the Credit GoesGWR-2026-IN-004

§ 15The funds the Directions do not reach

Takeaway · A rule written for non-bank lenders does not bind the funds that lend beside them.

The Directions apply to non-banking financial companies. They do not apply to an alternative investment fund lending from a pooled vehicle, which is how most Indian private credit is structured. That channel cleared $12.4bn across 166 transactions in CY2025, up 35% on the year,18 against total alternative-fund commitments of ₹15.74 lakh crore.17 It is smaller than the non-bank sector and it is growing faster, and nothing in this draft touches how it structures a facility.

TABLE 15.1 · Four destinations for a displaced working-capital line
DestinationCan it offer a revolving facility?Visible to RBI?Cost to borrower
Bank cash credit or overdraftYes, and the Dec 2025 directions widened access23FullyLowest
Card issuer with RBI authorisationYes, by exemption; two non-banks qualify3FullyHigh
Private credit fund or AIFYes; the Directions do not apply to the vehiclePartly, via SEBIHigh
Non-bank lender, redesigned productNo — repeat term loans onlyFullyModerate, plus delay
Unregulated or informal lenderYes, without conditionsNot at allHighest
$12.4bn
India private credit deal value, CY2025 ◐18
166
Transactions ◐18
+35%
Growth on CY2024 ◐18
₹15.74lakh cr
Total AIF commitments ◐17

The mismatch is one of ticket size rather than of principle. Private credit in India underwrites at a scale that suits a mid-market corporate refinancing, not a trader drawing ₹8 lakh against receivables. The fund channel will absorb the top of the displaced book and none of the bottom. Our own scenario work on private credit stress made the same point in reverse. The channel is deep where the cheques are large and thin everywhere else.21

The call

For an allocator to Indian private credit, this rule is a mild tailwind at the large end and irrelevant at the small end. The mispricing to avoid is treating a displaced ₹2 trillion as an addressable pipeline. Most of it is too small to be worth a fund's underwriting, and the rest was already being competed for.

Gravitywell Research22 / 40
Part IV · Where the Credit GoesGWR-2026-IN-004

§ 16Redesign: what survives a term-only rule

Takeaway · Repeat lending survives; the self-restoring limit does not, and the difference is a consent screen.

The draft does not stop a lender from lending to the same customer again. It stops the facility from restoring itself, and the distinction decides how much of the flexi franchise can be rebuilt. A pre-approved offer the customer accepts, creating a fresh term loan, is a different legal object from a limit that comes back on repayment. The customer experience can be nearly identical.

TABLE 16.1 · Redesign options against the drafted test
OptionMechanismSurvives?What it costs
Repeat pre-approved term loansNew sanction each time, accepted by the customerYesA consent and disbursal event per draw
Short-tenor term loans in seriesSequential term facilities mimicking a lineYesDocumentation load, and supervisory attention
Auto-renewing limit on repaymentThe limit restores itselfNoDirectly contrary to the definition2
Co-lending with a bankThe bank holds the revolving facility, the non-bank sources and services itPartlyEconomics shared, the customer becomes the bank's
Sourcing for a card issuerNon-bank originates, the licensed issuer holds the revolving creditPartlyFee income replaces spread income
Renaming the productSame replenishment, different labelNoThe test is mechanical, not nominal
FIG 16.1 · Where the customer journey gains a step
THREE STEPS BECOME FIVE, AND TWO OF THEM ARE CREDIT EVENTS
TODAY · REVOLVING LIMIT1 · Open the app2 · Choose an amount3 · Money movesThree steps, no credit eventAFTER · TERM ONLY1 · Open the app2 · Choose an amount3 · Fresh sanction4 · Accept new terms5 · Money moves
Steps 3 and 4 are the rule: each is a document, a disclosure and a decision the customer can decline.
Source: Gravitywell Research, from the drafted definitions ○2; event count from Bajaj Finance Q1 FY27 ●7. (● filed · ◐ modeled · ○ est.)
The call

The redesign is available to everyone, which is why it will not differentiate anyone. The lenders that come out ahead are the ones whose customer already lives inside an app opened for other reasons. Inserting two steps into a journey already under way costs far less than re-acquiring a customer who has left.

Gravitywell Research23 / 40
Part IV · Where the Credit GoesGWR-2026-IN-004

§ 17The edge outside the perimeter

Takeaway · The exposure the regulator is worried about is the one it can measure; this rule moves some of it out of reach.

Every destination in the previous section except one reports to somebody. A bank overdraft appears in the Reserve Bank's returns. A card balance appears in the card statistics. A fund's exposure appears in filings with the securities regulator. Credit that ends up with a trade creditor, a local financier or a merchant advance appears nowhere, and the borrower most likely to end up there is the one at the bottom of the displaced book: the small trader whose ticket is too small for a fund and whose file is too thin for a card.

The measurement problem, stated plainly

India does not publish drawn against sanctioned at the account level for non-bank lending. That single absence is why the ₹2 trillion cannot be checked and why utilisation-based evergreening cannot be detected from public data. It is also why a claim that credit has left the perimeter will be unprovable for at least a year after notification. It is also the cheapest thing in this whole debate to fix.

FIG 17.1 · Visibility falls as ticket size falls
WHERE A DISPLACED BORROWER COULD GO, AND WHO WOULD SEE IT
visible to RBI?Bank cash credit / overdraftAny size · in the RBI returnsfullyCard issuer, prime onlyMid ticket · in the card statisticsfullyNon-bank, repeat term loansMid ticket · fully supervisedfullyPrivate credit fundLarge ticket · reports to SEBIpartlyUnrecorded creditSmall ticket · reports to nobodynot at all
The borrower most likely to end up in the last row is the one at the bottom of the displaced book.
Source: Gravitywell Research schematic ○, from channel characteristics described in sections 13–16. (● filed · ◐ modeled · ○ est.)
The call

If you set policy, this is the cost side of the ledger and it is not in the draft's own reasoning. A prohibition can improve classification integrity at supervised lenders and still push the marginal borrower toward lenders nobody sees. That buys visibility in one place by spending it in another. Whether that trade is worth making is a judgement, but it should be made explicitly.

Gravitywell Research24 / 40
Part V · Precedent & the Policy Trade
Sections 18 — 20

Precedent and the Policy
Trade

This regulator has prohibited a revolving product at non-banks before, and has softened three drafts in the eighteen months since. Both records are on the table before 28 Aug 2026, and they point different ways.

Reading
Pages 25 — 27
Key figure
Zero transition, ×3
What you'll take away
01
Every comparable regulator gave a runway
This one has given none, three times running.
02
The imported analogues are the wrong class
We say where they stop working, and retier them.
03
It has narrowed three drafts since Dec 2024
Gold loans, project finance, liquidity coverage.
Part V · PrecedentGWR-2026-IN-004

§ 18The precedent that actually fits

Takeaway · Every comparable regulator gave a runway. This one has given none, three times running.

On 20 Jun 2022 the Reserve Bank barred non-bank prepaid issuers from loading their instruments with credit lines, with immediate effect and no transition. Banks were untouched.25 The structure is the one in front of us: a revolving product removed from non-banks and left alive at banks, by this regulator, in this decade. It was never diluted. Slice shifted "its provision to term loans and away from credit lines", which was not enough, and its terminal answer was to buy North East Small Finance Bank in Oct 2024 and relaunch the product from inside the perimeter in Jun 2025.26

FIG 18.1 · Transition runway granted, by interventionmonths from rule to effect
THE IMPORTED ANALOGUES CLUSTER AT THE GENEROUS END. THIS REGULATOR SITS AT ZERO.
Japan, money lenders 200636 monthsChina, micro-lending 202036 monthsFCA overdrafts 201910 monthsUS CARD Act 20099 monthsUK payday cap 20147 weeksRBI · PPI credit lines 2022immediateRBI · risk weights 2023immediateRBI · this draft 2026immediate, as drafted
Every comparable regulator gave a runway. This one has given none, three times running.
Source: Japan Money Lending Business Act 2006 and FRBSF commentary ◐39; China micro-lending measures ◐27; FCA PS19/16 and PS14/16 ●1338; CARD Act 2009 ●12; RBI circulars ●14251. (● filed · ◐ modeled · ○ est.)
The contrast case, and why we do not lead with it — corrected

The two Western interventions usually cited point the other way, and both are weaker than their headlines. The FCA's "£1bn of consumer savings" appears nowhere in its own evaluation; the press release adds a £500m annual run-rate to a £486m cumulative 22-month figure, and the evaluation's own table puts ongoing benefits nearer £600m a year.38 The £17.40 is per consumer with an overdraft facility or eligibility, which is 54% of current-account holders, not per user. And the measured window begins Nov 2020, excluding the first seven months the rules were live. What the intervention demonstrably did was raise the arranged rate: mean equivalent annual rate went from about 7.7% to 31%, and the Bank of England's effective overdraft rate went from 10.32% in Mar 2020 to 19.00% in Aug 2020, the highest in that series.38 On the CARD Act, the $11.9bn is an extrapolation from eight banks to the whole market, and the design identifies off small-business cards at those same banks — an exempt twin of the product surviving at the same institutions, which is precisely the condition this draft removes.12 Both are conduct remedies on surviving products at banks. Neither is a prohibition at one lender class.

The call

Underwrite the 2022 outcome and the zero runway. If a large flexi-heavy lender applies for credit-card authorisation in the next two quarters, that is not a curiosity, it is the precedent repeating.

Gravitywell Research25 / 40
Part V · PrecedentGWR-2026-IN-004

§ 19The precedent everyone cites, corrected

Takeaway · Feb 2025 never touched the consumer risk weight — it is still in force, 33 months on.

The case for dilution rests on a misreading. In Nov 2023 the Reserve Bank raised the risk weight on unsecured consumer credit from 100% to 125% for banks and non-banks alike, and in Feb 2025 it is widely said to have backed down.14 What it actually restored was the surcharge on bank exposures to non-banks, and the weight on microfinance loans classified as consumer credit. The 125% on unsecured personal loans and credit-card receivables was not rolled back, and remains in force today.24

FIG 19.1 · What was reversed, and what was not
THREE INSTRUMENTS, ONE HEADLINE, TWO OUTCOMES
status, Aug 2026Unsecured consumer credit100% → 125%, Nov 2023still in forceBank exposure to NBFCs+25pp surcharge, Nov 2023restored Feb 25Microfinance as consumer125%, Nov 2023restored Feb 25
The retail-conduct core has held for 33 months. What was eased was a bank-capital channel, when bank credit to non-banks had halved.
Source: RBI circulars of 16 Nov 2023 and 25 Feb 2025 ●14; scope of the Feb 2025 restoration per Khaitan & Co ◐24. (● filed · ◐ modeled · ○ est.)
TABLE 19.1 · The macro that produced the Feb 2025 easing, against today31
ConditionFeb 2025Aug 2026
Repo rate6.25%, first cut delivered5.25%, held a fourth time
Stanceeasing into weaknessneutral
GrowthQ2 FY25 GDP 5.4%, a seven-quarter lowFY27 forecast raised to 6.7%
Bank credit to non-banks6.7% YoY, down from 15%credit growth broad-based
What that impliesa growth scareno growth scare

One further caution on the Nov 2023 episode. The often-quoted slowdown from 22% to 11% compares a three-year compound rate off a COVID-depressed FY21 base against a single half-year, on a banks-only series, with no counterfactual.15 It supports a direction and cannot carry a sized elasticity, so this report does not use it as one.

Gravitywell Research26 / 40
Part V · PrecedentGWR-2026-IN-004

§ 20What a better redraft would say

Takeaway · This regulator has narrowed three drafts in eighteen months, and it has already enumerated once inside this one.

The case for narrowing does not rest on the industry's objection, which this regulator has heard and rejected before. It rests on its own recent conduct. Since Dec 2024 the Reserve Bank has materially softened at least three drafts between consultation and notification, and granted a transition in each.4042 The closest structural match is the gold-loan file: a product-structure rule landing on a segment non-banks dominate, with the trade body circulating the draft to members and an external representation from the finance ministry. The binding parameter moved, small tickets were carved out, and the industry got ten months.

TABLE 20.1 · What this regulator did the last three times it consulted
DraftFinalWhat movedTransition
Lending against gold collateral, Apr 20256 Jun 2025Flat 75% loan-to-value became tiered, up to 85% below ₹2.5 lakh; small-ticket carve-outs on end-use and appraisal10 months
Project finance, May 202419 Jun 2025Under-construction standard-asset provisioning cut from 5% to 1% for infrastructure~3 months
Liquidity coverage on digital deposits, Jul 2024Apr 2025Additional run-off factor 5% to 2.5%; wholesale unsecured run-off 100% to 40%deferred a year
Revolving credit, Aug 2026pendingThe question this report is aboutnone drafted

There is a second reason to expect narrowing rather than repeal, and it sits inside the draft. The term-loan definition already admits bullet repayment as well as instalments.3 That is a product-structure permission written into an instrument otherwise built on a residual test, which means the Reserve Bank is willing to enumerate when it wants a structure to survive. Asking it to enumerate further is asking for more of something it has already done, not for a concession of principle.

A claim we withdrew, and why it matters that we say so

An earlier version of this report argued that the draft contradicts its own parent Direction, because the Nov 2025 Directions require a board policy for demand and call loans that a term-only rule would prohibit.30 That argument is wrong and we withdraw it. The draft's consequential amendments remove the separate chapter on demand and call loans, along with a Chapter II provision, and replace them with the new restrictions section.2 The Reserve Bank saw the conflict and repealed the older provisions in the same instrument. We had disclosed that we could not open the draft itself; we should not then have built a claim on what it omits.

The call

If you are drafting a submission before 28 Aug 2026, argue structure, not fairness. Ask for the bullet-repayment permission to be extended into a positive list of surviving facilities — fixed-tenor supply-chain tranches, secured working capital drawn against current assets — and ask for a transition on the gold-loan template. Both are things this regulator has done inside the last eighteen months.

Gravitywell Research27 / 40
Part VI · Scenarios, Risk & the Trade
Sections 21 — 25

Scenarios, Risk and the
Trade

Three ways the consultation ends, what each does to the numbers, and the positions that follow, with the levels at which each one stops working.

Reading
Pages 28 — 32
Key figure
50–60% narrowed
What you'll take away
01
Narrowing is modal, and it does not save the retail product
Both likely branches take the consumer line.
02
The as-drafted branch is a different regime
It runs through funding, not through fees.
03
Every position here has a price
A position without a level is an opinion.
Part VI · ScenariosGWR-2026-IN-004

§ 21Three ways the consultation ends

Takeaway · Narrowing is modal at 50–60% — and the consumer flexi product goes in both likely branches.

Two precedents pull in opposite directions and both are real. The Reserve Bank prohibited a revolving product at non-banks in Jun 2022, immediately and permanently.25 But its draft-to-final record since Dec 2024 runs the other way: gold loans, project finance and liquidity coverage were each materially softened, each with a transition.4042 The 2022 case is the better structural match; the 2025 cases are the better behavioural base rate, three against one.

The Feb 2025 episode usually cited as evidence of retreat belongs in neither column: two circulars the same day restoring a bank-capital surcharge and a microfinance weight, never touching the consumer measure.24 One live signal cuts toward narrowing. On 14 Aug 2026 Reserve Bank officials met senior non-bank executives and framed the message around internal audit, compliance and risk systems "particularly for products witnessing rapid growth" — supervisory language, not that of a prohibition being defended. In 2022 the regulator notified and did not engage.43

The strongest evidence against us runs the other way and we weight it. Between Apr and Jun 2025 the Reserve Bank's gold-loan draft was materially narrowed on its way to final form, losing roughly a fifth of its paragraphs, dropping end-use tracking and adding proportional compliance below ₹2.5 lakh.40 That is a recent, domestic instance of this regulator consulting and then substantially rewriting. It is why the narrowed branch carries 50–60% rather than the 20% the 2022 precedent alone would imply, and why the as-drafted branch sits at 30–40% rather than higher.

FIG 21.1 · Where the probability sitsoutcome of the consultation
GRAVITYWELL RESEARCH WEIGHTS
Definition narrowed50–60%Notified as drafted30–40%Withdrawn5–10%
Desk judgement, not a market-implied probability. No instrument prices this outcome.
Source: Gravitywell Research ○, on the Jun 2022 prohibition ◐25 and the 2025 draft-to-final record ◐4042.
TABLE 21.1 · What each branch looks like, and what tells you early
BranchShape of the final ruleEarly tell
As drafted
30–40%
Term-only, residual definition intact, in force on notificationFinal Directions with the definition unchanged, as in Jun 2022
Narrowed
50–60%
Bullet repayment extended into a positive list, so fixed-tenor tranches and secured working capital survive; the consumer product still goes; a transition appearsRBI enumerating surviving facilities, a second draft, or a transition clause. The 14 Aug meeting is the first signal
Withdrawn
5–10%
Deferred or replaced by conduct rulesSilence past the FY27 policy calendar
The call

We land where the market lands on the headline and disagree on what it means. Narrowing is modal, but the consumer flexi product goes in both likely branches. What a narrowed rule saves is business lending, not the retail engine. Reading "probably diluted" as "probably fine" is the error.

Gravitywell Research28 / 40
Part VI · ScenariosGWR-2026-IN-004

§ 22What each case does to the numbers

Takeaway · The as-drafted branch is not the narrowed one scaled up; it runs through funding, not through fees.

The narrowed and withdrawn branches are arithmetic. Growth slows by the flexi that cannot be re-papered, costs rise by the number of new origination events, and the multiple adjusts to a lower durable growth rate. Nothing about the lender's funding or its access to the bond market changes, and the franchise is intact at the end of it. Note that all the figures below inherit the sizing problem in Part II, so read them as the shape of a response rather than as forecasts.

TABLE 22.1 · The three branches on the same five lines
LineNarrowedAs draftedWithdrawn
FY27 AUM growth at a flexi-heavy lender18–21%14–18%22–24%
Cost to net total income+20–60bp+60–120bpunchanged
Non-interest income−5 to −10%−10 to −20%unchanged
Durable growth embedded in the multiple~11.1%~10.6% or below~11.6%
Implied price-to-book~4.7×3.1–3.9×~5.9×

Guidance ranges are Gravitywell Research estimates ○ built from the §11 cost table and the §12 identity, applied to Bajaj Finance's disclosed FY27 corridors ●7. They are the desk's arithmetic, not company guidance, and they are conditional on a flexi book whose size is contested across a 13–31% range.435 They also hold credit cost flat, which is an assumption we cannot defend against our own §04: if a revolving limit masks borrower stress, withdrawing it surfaces the stress. Bernstein reads that as the primary risk.44 Read the table as the growth channel only.

Why the as-drafted branch is a different animal

Notified as drafted, with immediate effect and no transition, a second channel opens that the other two branches do not have. A lender re-papering a fifth of its book inside a quarter cannot grow it at the same time. Reported asset growth then falls sharply for two or three quarters, whatever demand is doing. That matters because Indian non-bank funding is priced off growth and asset quality together. The wholesale market treats a sudden growth stop as an asset-quality signal until proven otherwise. Bank lending to non-banks grew 26% in the last cycle;16 a funding market that has been extending credit on the strength of the sector's compounding is the same market that reprices when compounding pauses. The mild branches cost fees. This branch costs funding, and funding costs are what turn a product problem into a balance-sheet one.

The anomaly we cannot explain

Bajaj Finance disclosed its flexi book twice in 2020 and has not disclosed it since, through a draft rule that would prohibit the product and a fortnight in which four brokers guessed at it across a 13–25% range, against its own last published mark of about 31%.3537 We have no explanation for the withdrawal and we are not going to invent one. It stays unresolved here and in the conclusions.

The call

Underwrite the funding channel, not just the fee channel. The branch that hurts is the one where a growth pause reads as a credit signal to the lenders funding the sector. That reading gets made by people who will not wait for the second quarter's data.

Gravitywell Research29 / 40
Part VI · ScenariosGWR-2026-IN-004

§ 23The trade, with prices

Takeaway · Four positions, each with the level at which it stops being one.

A view without a price is half a view, so here are the levels. Bajaj Finance traded at ₹1,096 and about 5.9 times book on 20 Aug 2026, against sell-side targets running from ₹815 to ₹1,300 that were all set before the draft appeared and none of which has been publicly restated since.1011 On the §21 weights our scenario-weighted fair value is roughly 4.2 to 4.6 times book, which is a low-to-mid ₹800s share price on the current book. The gap to spot is the trade, and it comes from the weights in §21 rather than from any new arithmetic.

TABLE 23.1 · Positions, entry logic and the level that invalidates each
PositionLogicEntry / referenceInvalidated if
Underweight the flexi-heavy compounderThe multiple pays for growth the rule taxes; scenario-weighted fair value is 4.2–4.6× book against 5.9× spot5.9× book
₹1,096
The narrowing extends to consumer flexi, or FY27 growth prints inside 22–24% after notification
Relative long, exempt card issuersThe exemption removes the only non-bank competitor in revolving creditvs the flexi-heavy names, not outrightCard receivables growth stays at 3% through two quarters after notification
Own the negligible-exposure lendersCholamandalam under 1%, L&T Finance and Poonawalla effectively nil on broker estimates that the Tata row warns should be discounted428on the sector de-rateThe rule is withdrawn, removing the handicap
Avoid: unlisted working-capital lendersRevolving by construction, no disclosure, no public pricen/aEnumeration preserves secured business lines
FIG 23.1 · Scenario-weighted value against the spot multipleprice-to-book, ×
THE GAP IS THE TRADE, AND IT IS ONE NOTCH, NOT THREE
3.1×as drafted4.2–4.6×weighted4.7×narrowed5.9×spot ₹1,096
Weighted at 30–40% as drafted, 50–60% narrowed, 5–10% withdrawn. A reader who reads narrowing as reprieve gets a fair value nearer 4.7×.
Source: Gravitywell Research, from the §12 identity and the §21 weights ○; spot multiple ◐11. (● filed · ◐ modeled · ○ est.)
The call

Held through the notification, this position turns on two things and neither is the announcement itself. The first is whether the final Directions keep the residual definition. The second is whether any lender finally discloses its flexi book, because the day one of them does, every estimate on this page is replaced by a fact.

Gravitywell Research30 / 40
Part VI · Risk RegisterGWR-2026-IN-004

§ 24Risk register

Takeaway · The operational risk of re-papering a live book is under-rated relative to the commercial one.
Regulatory · notification without a transition
The draft comes into force immediately once notified and no grandfathering is written into it.1 Every live facility then has to be converted against a clock the lender does not set.
TRIGGER: final Directions published with no transition clause
High
0–6m
Operational · mass re-papering of agreements and mandates
Converting a limit book touches agreements, key-fact statements, mandates and consent records at account level. The market is not discussing it.
TRIGGER: any lender discloses a customer-consent or documentation failure tied to product migration
High
0–12m
Commercial · growth guidance cut
A term book must be re-originated to stand still, and Bajaj Finance guides to 22–24% asset growth for FY27.7 Note the wedge in §08 cuts the other way on how large that engine still is.
TRIGGER: FY27 guidance revised within two quarters of notification
High
6–18m
Credit · withdrawal surfaces the impairment the product was hiding
If §04's mechanism is right, removing the facility stops masking borrower stress and credit cost rises. Bernstein calls this the primary risk, above growth.44
TRIGGER: unsecured stage-2 or 30+ delinquency rises at a flexi-heavy lender within two quarters of notification
High
6–18m
Financial · funding reprices on a growth pause
Wholesale lenders read a sudden stop in asset growth as an asset-quality signal. Bank lending to non-banks grew 26% into this cycle16 and can decelerate faster than it accelerated.
TRIGGER: spread widening on a large non-bank's paper without a change in reported asset quality
Medium
6–18m
Measurement · nobody counts the affected book
One unattributed industry figure, no facility-structure series,533 and a desk panel that missed its one calibration test by ~2×.28 Every sizing inherits it, ours included.
TRIGGER: RBI or a bureau publishes a drawn-against-sanctioned series for NBFCs
High
standing
Competitive · the arbitrage entrenches
Banks were given wider cash-credit access in Dec 202523 and two card issuers keep revolving credit by licence. The asymmetry is durable, not transitional.
TRIGGER: bank working-capital growth accelerates while non-bank unsecured growth decelerates for two consecutive quarters
Medium
12m+
Technological · the redesign is cheap and the thesis narrows
If a repeat pre-approved term loan costs a large lender almost nothing, the friction this report prices does not exist. The story then reduces to reported-growth optics.
TRIGGER: a flexi-heavy lender reports flat cost-to-income through two post-notification quarters
Medium
6–12m

Also carried at low severity: borrowers denied a line taking a larger term loan than needed and paying interest on all of it; and the redesign migrating facilities into fund structures with different tax treatment and reporting, same borrower and same risk.

HighMediumLowHorizon measured from notification, not from the draft date
Gravitywell Research31 / 40
Part VI · FalsificationGWR-2026-IN-004

§ 25Benchmarks that would change our view

Takeaway · Five dated observables, and the claim we would defend last.

This report makes two claims that can be checked rather than debated. The first is that the affected book cannot currently be measured by anyone. The second is that the Jun 2022 prohibition, not the Feb 2025 easing, is the precedent that governs what happens next. Here is what would break each of them.

TABLE 25.1 · What would falsify the house view
ObservableBy whenWhat it would mean
The Reserve Bank, a bureau or a rating agency publishes a facility-structure series for non-banksany timeThe central claim of this report stops being true, and the ₹2trn becomes checkable rather than assertable
A large non-bank discloses its flexi or line book, with redraw frequency per active accountQ2 FY27 resultsReplaces the whole of Part II with a fact. If it lands near the top of the broker range, or nearer its own 2020 mark, the sizing debate resolves and §08 is wrong
Final Directions arrive with a transition window and a positive enumeration of prohibited facilitiesFY27The narrowed branch, and our 45–55% weight on notification as drafted was too high
The prohibition is withdrawn outrightFY27The Jun 2022 precedent does not govern, and the reference class in §18 is wrong
A flexi-heavy lender applies for or receives credit-card authorisationtwo quartersThe 2022 sequence is repeating, and the exemption is an option with a strike rather than a two-name annuity
The weakest load-bearing claim in this document

It is the reference class. Everything in Part VI rests on Jun 2022 being the right analogue for Aug 2026, and there are real differences: prepaid instruments loaded with credit lines were a fintech workaround of recent origin, while flexi lending and working-capital lines are decades-old products at systemically significant lenders with a trade body organising against the draft. If the Reserve Bank treats those as different cases, our weights are wrong in the market's favour. We think the structural match outweighs the differences: a revolving product removed from non-banks and left with banks. A reader who disagrees should move weight to the narrowed branch, and will get a fair value close to spot.

Two things left open

First, the flexi transaction wedge in §08. Transactions have grown 4.1% a year against a 24% book for two years, and we cannot say which of the three explanations is right because the metric is undefined.32 Second, whether supply-chain finance is in scope at all. It is typically written as discrete 30 to 180-day tranches, each of which passes the term-loan test, which would take a large block out of the claimed ₹2 trillion. The draft's own clause text would settle it and we could not obtain the document.1 Neither question is resolved in the conclusions.

Gravitywell Research32 / 40
Investment ConclusionsGWR-2026-IN-004

ConclusionsWhat we think, stated as positions

In one line · Medium conviction, Cautious, and the sizing everybody is using does not exist.

The Reserve Bank has proposed removing the redraw from India's non-bank lenders. It will probably narrow the instrument rather than withdraw it, on its own record of the last eighteen months,40 and narrowing does not save the consumer product. What nobody can tell you, including this desk, is how large the thing being removed is. That is the argument of this report, and it is a smaller claim than the one we set out to make.

Four positions
Narrowed is modal; as-drafted is live at 30–40%High

The market treats dilution as reprieve. Both likely branches still take the consumer revolving product, and the Feb 2025 precedent usually cited does not say what it is thought to say.24

Trim the flexi-heavy compounderMedium

Scenario-weighted fair value 4.2–4.6× book against 5.9× spot.11 Medium, not high, because the weights carry the call and the book size is unknown.

Ask for the permission the draft already containsHigh

Bullet repayment is admitted as a term-loan form.3 Extending that into a positive list is the submission worth writing before 28 Aug.

Credit relocates rather than retiresHigh

Banks keep cash credit and had it widened in Dec 2025;23 two card issuers keep revolving by licence. Neither outcome reduces the exposure the draft is aimed at.

What we got wrong on the way here

This report was drafted around a different argument: that the rule taxes a self-renewing origination engine, and that the cost is measured in redraw events. Bajaj Finance's own eight-quarter series does not support it. Flexi transactions have compounded at 4.1% while the book compounded at 24%,32 which either means the engine is idling or means the metric never counted redraws. We have demoted that framing rather than defending it, and §08 prints the series that did the demoting. Our earlier work has the same failure mode: in The Exit Window we read a policy-driven queue as a pricing problem and were slow to say the binding constraint was administrative.20

The house view

GWR-2026-IN-004 · Medium conviction · Cautious. Cautious into the notification on the precedent rather than the arithmetic. We publish no impact estimate, because the inputs for one do not exist. The first lender to disclose its flexi book will do more for this debate than any note written before it.

Gravitywell Research33 / 40
MethodologyGWR-2026-IN-004

MethodologyHow this was built, and what it cannot see

Coverage is disclosed, uncertainty ships as ranges, every figure is dated and tiered, and no prior number is silently restated. Where a figure is the desk's own construction it is marked ○ and its derivation is printed rather than described.

Coverage, and its limits
TABLE M.1 · What is in scope, what is out, and where the data is thin
AreaStatusEffect on the analysis
The draft's clause textReproduced from press reporting; the RBI document itself sits behind a bot challenge and could not be opened directly. The draft's existence, title, date and comment deadline are confirmed on rbi.org.in1All clause quotations tiered ◐
Size of the in-scope bookOne unattributed industry figure; no regulator, bureau or agency series exists on facility structure533No estimate published
Lender-level revolving exposureDisclosed by one company only; broker estimates elsewhere, blanks left blank. The broker method missed the one checkable row by about two times428Panel incomplete, and calibrated once
Cost per originationNot disclosed by any Indian lender; and the transaction count it would multiply is undefined32Conditional grid, not a forecast
Bank cash-credit and overdraft outstandingsNot obtained at this vintage; the migration section is argued on ratios and policy, not on an RBI facility-type seriesUnder-covered, stated here
Whether supply-chain finance is in scopeTurns on clause text we could not obtain; discrete tranches may pass the term-loan test1Left open
Consultation outcomeWeighted against the Jun 2022 prohibition, not the Feb 2025 easing2524Judgemental ○
How the constructed figures were derived

Why there is no desk estimate of the in-scope book. An earlier draft built one, stacking the industry's ₹2trn with consumer flexi, loan-against-securities limits and reusable card lines, then haircutting for overlap. We withdrew it. Three of the four rungs rested on broker estimates of a single lender, and the overlap haircut was a judgement with no evidence behind it. The exercise dressed an unattributed figure in arithmetic that made it look sourced. §06 now shows what is actually sized, against what is claimed, and leaves the gap open. A range built on an unverifiable anchor is not more honest than declining to publish one.

The valuation sensitivity. §12 uses P/B = (ROE − g) ÷ (COE − g) at 20% return on equity and 13% cost of equity, solving for the growth implied by the observed multiple then re-solving at lower growth. The identity is unstable as growth nears the cost of equity, which the page discloses. The direction survives; the decimals do not.

The origination-cost grid, and its condition. §11 applies four assumed marginal costs to the disclosed count of flexi transactions and the disclosed operating expense base. It is conditional twice over: no lender publishes a marginal cost per origination, and Bajaj Finance does not define what a flexi transaction is, so the count may not be redraws at all. The page says so at the top. Its usable conclusion is the bound, not the point.

The calibration record. On 7 Aug 2026 brokers put Tata Capital's revolving exposure in the high single to low double digits; on 11 Aug its chief executive said below 5%.428 That is the one row in the desk's panel where a primary disclosure exists to check against, and the broker method was out by roughly two times. We have not applied a correction factor to the other rows, because one observation does not support one. The miss is recorded in the dataset README rather than quietly repaired in the table.

Point-in-time policy

Every figure carries its as-of date. If the final Directions differ from the draft, this report is superseded rather than quietly amended, and material corrections are appended as a dated erratum with the permalink serving the current version.

Gravitywell Research34 / 40
Source Register · IGWR-2026-IN-004

SourcesRegister · 1 to 18

Numbers match the superscripts in the body. Tier per entry: ● filed/official · ◐ modeled/secondary · ○ desk construction.

Gravitywell Research35 / 40
Source Register · IIGWR-2026-IN-004

SourcesRegister · 19 to 44

  1. Whalesbook, "RBI Draft on Revolving Credit Hits NBFCs, Bajaj Group Loses ₹55,000 Cr", 7 Aug 2026 — the circulating figure this report rejects; see Data Appendix A.4 ◐
  2. Gravitywell Research, GWR-2026-IN-003 The Exit Window, 14 Aug 2026 ●
  3. Gravitywell Research, GWW-2026-002 The Private Credit Unwind, 14 Aug 2026 ●
  4. Gravitywell Research desk panel, "NBFC revolving-book exposure", dataset at data/nbfc-revolving-book/, 20 Aug 2026 ○
  5. Reserve Bank of India, revised directions on current accounts, cash credit and overdraft accounts, issued 11 Dec 2025 after a draft of 1 Oct 2025, applicable 1 Apr 2026 — cash credit permitted without restriction; current and overdraft accounts unrestricted below ₹10 crore of banking-system exposure; six-monthly review and core-system flagging added ◐
  6. Khaitan & Co, "RBI eases risk weight on consumer credit and bank loan", Feb 2025 — establishes that the 25 Feb 2025 restoration covered bank exposures to NBFCs and microfinance, and that the 125% on unsecured consumer credit was not rolled back ◐
  7. Business Standard, "Non-banks can't load PPIs with credit lines, says RBI", 21 Jun 2022, and Cyril Amarchand Mangaldas, FIG Paper No. 13 — the prohibition, its immediate effect, and the exclusion of bank-issued instruments ◐
  8. Medianama, "Slice enters Indian small finance banking with NESFB merger", Oct 2024 — the merger completed 27 Oct 2024; product relaunched as a bank in Jun 2025 ◐
  9. Caixin, "Ahead of Ant Group's record IPO, China tightens online microlending rules", 3 Nov 2020, and "In depth: the rectification and remaking of Ant Group", 27 Apr 2021 — provincial licence confinement, the 30% joint-funding minimum, and a three-year transition ◐
  10. Business Standard, "Tata Capital's revolving credit exposure below 5% of loan book: Sabharwal", 11 Aug 2026 — MD & CEO Rajiv Sabharwal on the record ●
  11. Business Standard, "Bajaj Finance share price crash erodes ₹33k cr from m-cap: what's behind the fall", 7 Aug 2026, 12:17pm IST — Bonanza ~25% of AUM; states that Bajaj Finance has not disclosed the share of flexi loans; the ₹33,000cr is an intraday figure ◐
  12. Reserve Bank of India (Non-Banking Financial Companies – Credit Facilities) Directions, 2025, effective 28 Nov 2025 — the parent instrument, which requires a board-approved policy covering demand and call loans ◐
  13. Reserve Bank of India, monetary policy statement, Aug 2026 — repo held at 5.25% for a fourth consecutive meeting, neutral stance, FY27 GDP forecast raised to 6.7%, FY27 CPI forecast cut to 5.0% ◐
  14. Bajaj Finance Limited, Q3 FY26 and Q4 FY26 Investor Presentations — the flexi loan transaction series: 11.62mn in FY25, 12.10mn in FY26, and 2.90mn in Q1 FY26 ●
  15. CRIF High Mark, How India Lends, May 2026 edition, and TransUnion CIBIL's Credit Market Indicator — product-category series, not facility-structure series; cited here for what they do not contain ◐
  16. Business Standard, "RBI asks shadow banks to give information on supply-chain financing", 15 Nov 2024 — the regulator requesting facility-level data it did not hold ◐
  17. Bajaj Finance Limited, Q1 FY21 Investor Presentation, 21 Jul 2020, filed with BSE and NSE — "As of March 31st, the existing book under flexi stood at ₹36,846 crore", against consolidated assets of ₹1,47,153 crore ●
  18. Bajaj Finance Limited, Q2 FY21 earnings call transcript, 21 Oct 2020 — Rajeev Jain: total outstanding flexi "around 43,000 crores… across mortgages to personal loans to gold loans, LAS"; utilisation "between 60% and 85% of the approved line"; product taxonomy of term loan, drop-line flexi and flexi; segment penetration; and flexi economics, including "loan losses being marginally higher compared to term loan due to slower amortization" and "I get more fees" ●
  19. Bajaj Finance Limited, Q3 FY21 earnings call, 20 Jan 2021 ("I think flexi is history now"; "No way. We are not tracking even that") and Q1 FY22 earnings call, 20 Jul 2021 ("probably today is the last time I am responding to a direct question on flexi") ●
  20. FCA, Evaluation Paper 23/1 Technical Annex, and Bank of England, Money and Credit, Aug 2020 — the £500m annual run-rate and £486m 22-month figures, the Nov 2020 start of the measured window, the arranged-rate rise, and the effective overdraft rate moving from 10.32% in Mar 2020 to 19.00% in Aug 2020 ●
  21. Japan, Money Lending Business Act as amended 13 Dec 2006, with Federal Reserve Bank of San Francisco commentary; and FCA PS14/16, the high-cost short-term credit price cap in force 2 Jan 2015, which expressly excluded overdraft charges — two further cases of a regulator binding non-banks and leaving banks outside ◐
  22. Reserve Bank of India, gold-loan directions: draft of Apr 2025 and final guidelines of Jun 2025 — the draft was materially narrowed between consultation and notification, dropping end-use tracking and adding proportional compliance below ₹2.5 lakh ◐
  23. Broker record since the draft, via Trendlyne consensus and press reporting, 6–20 Aug 2026 — no published estimate cut on Bajaj Finance; consensus target ₹1,152.69 on 20 Aug against a ₹1,144.7 pre-draft close; Jefferies (flexi ~14% of portfolio, 1–1.5% EPS impact, the fall "larger than the likely earnings impact"); Morgan Stanley, Kotak Institutional Equities, Bernstein and SMC Global views ◐
  24. Reserve Bank of India draft-to-final record under the current governor: Project Finance Directions (draft May 2024, Directions 19 Jun 2025 — under-construction standard-asset provisioning cut from 5% to 1% for infrastructure) and the liquidity-coverage treatment of digital deposits (draft Jul 2024, circular Apr 2025 — additional run-off factor cut from 5% to 2.5%, implementation deferred a year), via Vinod Kothari Consultants, Khaitan & Co and Business Standard ◐
  25. Outlook Business, "NBFCs Seek RBI Review Of Proposed Curbs On Revolving Credit", 17 Aug 2026 — Reserve Bank officials met senior non-bank executives on 14 Aug 2026 and stressed internal audit, compliance and risk-management systems "particularly for products witnessing rapid growth"; Bajaj Finance, Tata Capital and Shriram Finance seeking meetings ◐
  26. Bernstein note on the 6 Aug draft, circulated 7 Aug 2026 — argues the primary risk is asset quality rather than growth, because revolving products "increase the risk of borrowers relying on incremental borrowing capacity to manage existing obligations" and withdrawal "could expose pockets of borrower stress that have so far remained masked" ◐
Gravitywell Research36 / 40
Glossary & AcronymsGWR-2026-IN-004

ReferenceGlossary & acronyms

Revolving credit
A facility a borrower can draw, repay and draw again up to a sanctioned limit. Under the draft, defined as any fund-based facility that is not a term loan.
Term loan
A fixed principal, disbursed once or in instalments and repaid on a schedule. On repayment the limit does not come back.
Flexi loan
Bajaj Finance's family of revolving personal and business facilities, sold with an annual maintenance fee on the available limit.
Drop-line overdraft
An overdraft whose sanctioned limit steps down on a schedule. Revolving within each step, so caught by the drafted test.
Cash credit
The bank equivalent of a working-capital line, drawn against current assets. Widened rather than restricted by the Dec 2025 directions, effective Apr 2026.
Bullet repayment
Repaying principal in a single payment at maturity. The draft expressly admits it as a term-loan structure, which is why a fixed-tenor tranche can survive a rule aimed at revolving credit.
Evergreening
Using new credit to service old credit so an account never falls overdue and is never classified as impaired.
Utilisation
Drawn balance as a share of sanctioned limit. Not published at account level in India, which is why line-type exposure cannot be independently sized.
Re-origination
Selling the same customer a fresh loan after the previous one is repaid. The cost a revolving limit avoids and a term book cannot.
Co-lending
A bank and a non-bank sharing a loan, with the bank holding the larger share and the non-bank sourcing and servicing it.
Durable growth
The long-run growth rate embedded in a valuation multiple. In §12 it is solved from the observed price rather than assumed.
Grandfathering
Allowing existing contracts to run to maturity under the old rule. Absent from the draft, which is the source of the transition risk.
Residual definition
Defining a prohibited class as everything that is not the permitted one. The draft defines revolving credit as any facility failing the term-loan test, which is why its perimeter is contested.
Demand or call loan
A loan repayable when the lender calls it, with no predetermined schedule. The draft removes the parent Directions' separate chapter governing it.
Prepaid payment instrument
A stored-value instrument such as a wallet or prepaid card. Loading one with a credit line was prohibited for non-banks in Jun 2022, the closest domestic precedent to this draft.
Positive enumeration
Listing the facilities a rule prohibits, rather than prohibiting everything outside a definition. The likeliest form of a narrowed final rule.
Account aggregator
India's consent-based framework for sharing bank-account data with a lender. It is how a non-bank could acquire the cash-flow visibility a bank has by default.
AIFAlternative Investment Fund, registered with SEBI AUMAssets under management BFLBajaj Finance Limited CC/ODCash credit and overdraft COECost of equity FIDCFinance Industry Development Council, the self-regulatory body for Indian non-bank lenders FYFinancial year to 31 March; FY27 is the year to Mar 2027 LAPLoan against property LASLoan against securities MSMEMicro, small and medium enterprise NBFCNon-banking financial company NTINet total income P/BPrice to book value RBIReserve Bank of India ROEReturn on equity SEBISecurities and Exchange Board of India
Gravitywell Research37 / 40
List of ExhibitsGWR-2026-IN-004

ReferenceList of exhibits

Figures19
1.1Three years of narrowing the non-bank product set08
2.1Which facilities fail the term-loan test09
3.1The exempt franchise is growing spends, not balances10
4.1How the undrawn half hides the drawn half11
5.1Non-banks are growing faster than the system that funds them12
6.1Everything anyone has actually sized, against the claim13
8.1What the channel series actually shows15
9.1Three revenue lines against one16
10.1The same 20% growth, bought two different ways17
11.1Why the cost line reads as manageable18
12.1The price moved, the estimates did not19
13.1Two directions, eight months apart, opposite signs20
14.1One lender's flexi exposure against the exempt franchise21
16.1Where the customer journey gains a step23
17.1Visibility falls as ticket size falls24
18.1Transition runway granted, by intervention25
19.1What was reversed, and what was not26
21.1Where the probability sits28
23.1Scenario-weighted value against the spot multiple30
Tables17
0.1The consensus, the delta, the mechanism, the falsifier06
1.1The draft, clause by clause08
4.1Four strands of the case, and the industry's answer11
5.1The scale the rule applies to12
7.1What Bajaj Finance said, and when it stopped saying it14
9.1Where the income sits at a flexi-heavy lender16
11.1What the extra origination load does, on three assumptions about marginal cost18
12.1Implied multiple as durable growth moves, ROE held at 20% and cost of equity at 13%19
15.1Four destinations for a displaced working-capital line22
16.1Redesign options against the drafted test23
19.1The macro that produced the Feb 2025 easing, against today26
20.1What this regulator did the last three times it consulted27
21.1What each branch looks like, and what tells you early28
22.1The three branches on the same five lines29
23.1Positions, entry logic and the level that invalidates each30
25.1What would falsify the house view32
M.1What is in scope, what is out, and where the data is thin34
Gravitywell Research38 / 40
Data AppendixGWR-2026-IN-004

Appendix AUnderlying data

The series behind the exhibits, so the arithmetic is reproducible. Tier: ● filed · ◐ modeled · ○ estimate.

A.1 · Bajaj Finance segment assets, 30 Jun 2026 — behind FIG 2.1 and the §06 build
Series₹ croreShareTierSource
Urban personal loans1,09,80220.1%7
Mortgages1,73,62431.7%7
MSME lending51,3209.4%7
Urban consumer finance45,2208.3%7
Loan against securities, including margin trade facility36,5646.7%7
All other segments1,30,41423.8%7
Total consolidated assets under management5,46,944100%7
A.2 · Operating series behind FIG 8.1, TABLE 9.1 and TABLE 11.1
SeriesQ1 FY27Q1 FY26TierSource
New loans booked (mn)16.13n/d7
Flexi loan transactions (mn)3.022.907
Net total income (₹ cr)15,22412,4607
Operating expenses (₹ cr)5,0874,1247
Cost to net total income33.4%33.1%7
A.3 · The flexi transaction series behind FIG 8.1
PeriodFlexi transactions (mn)AUM growth YoYTierSource
FY25 full year11.62n/a32
FY26 full year12.10n/a32
Q1 FY262.90n/a32
Q4 FY263.10n/a7
Q1 FY273.0224%7
Growth, FY25 to FY26 and Q1 to Q1+4.1%+24%7; 32

Bajaj Finance does not define "flexi loan transaction" in its disclosure. The series is reported as a flow, and the FY26 quarters sum to the printed annual figure. That is why we treat it as a count of events rather than a stock of accounts. What kind of event it counts is unknown.

A.4 · A figure we rejected, and the arithmetic

A widely circulated report put the value destroyed on 7 Aug 2026 at ₹55,000 crore across Bajaj Finance and Bajaj Finserv.19 Close to close, Bajaj Finance fell 4.52% to ₹1,093 on about 6.22bn shares, roughly ₹32,000 crore; Bajaj Finserv fell 3.67%, about ₹12,200 crore. The sum is ₹44,400 crore, not ₹55,000 crore. Adding the two also double-counts: Bajaj Finserv holds 51.32% of Bajaj Finance, so its look-through share of the subsidiary's loss is larger than its own entire market-value decline. This report uses ~₹32,000 crore at Bajaj Finance.11

Machine-readable copy: nbfc-revolving-credit-data.csv ships alongside this report. The maintained panel behind TABLE 7.1 lives at data/nbfc-revolving-book/ with its own sources and update cadence.22

Gravitywell Research39 / 40
Disclosures & GovernanceGWR-2026-IN-004

DisclosuresCertification & governance

Analyst certification

The Gravitywell Research desk responsible for this report certifies that the views expressed accurately reflect its independent judgement about the subjects and securities discussed, and that no part of its compensation was, is, or will be directly or indirectly tied to the specific recommendations or views expressed herein.

Positioning & conflicts

As of the publication date, Gravitywell Research and its analysts do not hold positions in the securities or assets discussed. Gravitywell Research has no advisory, banking, or commercial relationship with the entities named. This report was not commissioned or reviewed by any issuer named, and no draft was shown to any entity discussed before publication.

House rating scale

Conviction High · Medium · Low, set by coverage depth, source tier and how far the thesis has been stress-tested. Stance Constructive · Neutral · Cautious, the direction of the house view. This report: Medium conviction · Cautious. Conviction is Medium rather than High because the call rests on a reference class rather than on measured data, and because the size of the affected book cannot be verified by anyone. Both limits are stated in §25.

Distribution

Prepared for readers who allocate capital, underwrite risk, or set policy across public and private markets. Not for general retail distribution, nor for readers who lack the expertise to assess the assumptions. Intended recipients may not redistribute without attribution. Availability of this research in some jurisdictions may be restricted; recipients are responsible for their local rules.

Credits

Cover art commissioned and graded by the Gravitywell art desk. Typeset in Playfair Display and IBM Plex; all typefaces are embedded, so this document renders without a network connection. © 2026 Gravitywell Research. All rights reserved.

Governance

Material errors are corrected in a dated erratum appended to this report and noted in the next edition; the report's permalink always serves the current version. Permalink: gravitywellresearch.xyz/research/nbfc-revolving-credit. Desk: [email protected]. Version: GWR-2026-IN-004 · v1.0 · as of 20 Aug 2026. This report reads a draft regulation; if the final Directions differ, the report is superseded rather than amended.

Prior calls
Prior callWhat we saidWhat has happenedVerdict
GWR-2026-IN-003 · Aug 2026CY2026 India IPO proceeds land ₹/$11–14bn, roughly half the sell-side consensusOpen; resolves 15 Jan 2027Too early
GWR-2026-IN-003 · Aug 2026NSE lists inside Q4 CY2026Open; resolves 15 Jan 2027Too early
GWW-2026-002 · Aug 2026US private-credit seizure activates on a three-leg test inside 18 months, at 15–25%Open; resolves 14 Feb 2028Too early
First editionNo prior Gravitywell Research call on non-bank product regulationn/a

Every falsifiable call here is logged on publication to the desk's append-only calls register, with a metric, a horizon and a resolution date. Each is scored publicly when it matures. Misses are recorded with the same care as hits.

Gravitywell Research40 / 40
Gravitywell
Research
CREDIT RESEARCH
GWR-2026-IN-004
AUGUST 2026

A credit limit is not a loan. It is a standing permission to borrow again, and the draft withdraws the permission rather than the credit.

India Non-Bank Credit Series

Independent research for capital allocators, risk officers, and policymakers. Not regulated ratings, not investment advice, not an offer or solicitation. Figures are marked to the dates shown and may be revised; Gravitywell Research is under no obligation to update. Sources are cited in the register; while drawn from sources believed reliable, accuracy is not guaranteed.

© 2026 Gravitywell Research gravitywellresearch.xyz