NBFCs & Housing Finance

Phase 5 · Week 43 · ~20 hours

Where it sits. The banking playbook ended by telling you that an NBFC is "a bank minus the deposit franchise and plus a liability time bomb". Here is the full account of that sentence. This is the second sector playbook, and the most India-critical module in Phase 5. Everything you learned about lenders still applies: leverage is the product, the P&L is a spread machine, earnings lag the truth, equity is valued directly on P/B against ROE. What changes is the liability side, and the change is total. A bank funds itself with insured, granular, sticky deposits and stands behind them with a central bank that will lend against good collateral in a panic. An NBFC funds itself with wholesale money (bank lines, bonds, commercial paper) that is uninsured, concentrated, and free to say "no" every 90 days, with no lender of last resort standing behind it. That single difference reorders the entire analysis: for a bank you study the asset side first and the funding second; for an NBFC you read the liability side first, because that is where the company dies. Ahead: the asset-liability management (ALM) statement built from scratch and drilled until you can read one cold, the full KPI panel with thresholds, autopsies on IL&FS and DHFL (the 2018–19 failures that taught India this lesson at a cost of a few lakh crore), the five great NBFC segments mapped (housing, vehicle, gold, microfinance, unsecured consumer), valuation end-to-end, and the US analogues at the close, because America learned the identical lesson in 2008 when its wholesale-funded lenders died or became banks. On the master map's five-questions spine: for a bank, "is it a good business?" and "what can kill it?" are the same question in different tenses; for an NBFC, question 4 comes first. You get to admire the asset side only after the liability side clears.

Suggested hour budget. Core content: 8h · worked examples: 1.5h · practice set: 3h · applied mini-project (the NBFC workup): 5h · flashcards + mastery check: 1.5h · teach-back + journal: 1h.

App link. The NBFC wing of the sector case engine runs on this material: the timed ALM-gap drill (a bucketed liquidity statement appears; you compute per-bucket and cumulative gaps, test them against the regulatory caps, and classify survive/stress/dead), the borrowing-mix classifier (funding pies with maturity profiles; you grade the refinancing risk), the KPI-panel drill carried over from the banking playbook with NBFC thresholds, and the workup grader that scores your mini-project against the rubric below: the gate for the insurance playbook. Your completed panel seeds the second page of the sector KPI cheat-sheet artifact.

Learning objectives

By the end you can:

1. Explain what an NBFC is and why India built a huge one-of-a-kind non-bank lending sector: the 50-50 principal-business test, the licensing history, the specialization economics (last-mile underwriting banks cannot do), and the funding plumbing (mutual-fund money → CP/NCDs) that inflated the sector into 2018.
2. State the structural fragility from first principles: borrow wholesale, lend niche, with no insured deposits, no depositor granularity and no lender-of-last-resort backstop, and therefore why an NBFC's survival is a rollover decision made by other people every quarter.
3. Build and read an ALM/structural-liquidity statement: bucket inflows and outflows, compute per-bucket and cumulative gaps and gap-as-%-of-outflows, apply RBI's granular-bucket caps, strip out rollover and behavioral assumptions, and diagnose the negative short-bucket gap that killed IL&FS and DHFL.
4. Compute and classify the full KPI panel from published accounts: AUM growth and its reconciliation (including off-book), NIM/spread and risk-adjusted spread, cost of funds and borrowing mix, GNPA/Stage-3 and credit cost under Ind AS ECL, gearing, CRAR, ROA/ROE, each against its healthy/stress threshold, and place the lender in its RBI Scale-Based Regulation layer.
5. Narrate the complete anatomy of the IL&FS and DHFL failures: structure, timeline, mechanism, and the flags that were publicly visible in advance. Map them to the US 2008 wholesale-funding analogues (GE Capital, Northern Rock, Countrywide) and to the consumer-finance survivors that solved the problem by becoming banks.
6. Apply the five segment lenses (housing finance, vehicle finance, gold loans, microfinance, unsecured consumer), knowing each segment's specific economics, normal KPI ranges, and characteristic way of dying.
7. Value an NBFC on P/B against ROE: justified P/B, the excess-return frame, reverse-reading a franchise multiple into its implied ROE-growth-duration claim, and stress-testing stated book value for a wholesale lender before calling anything cheap.
8. Run the red-flag panel in fifteen minutes on ALM mismatch and CP reliance, AUM growth outrunning disbursement quality, spread squeeze, low provisioning on unsecured books and promoter pledge, naming the exact disclosure where each is caught, and deliver the Phase 5 NBFC workup at rubric standard.

Prerequisites & connections

Builds on. M5.01 is mandatory and comes immediately before this. You'll reuse its ROA-×-leverage identity, the KPI-panel habit, P/B against ROE with the excess-return machinery of M3.10, and the credit-cycle logic of why lender earnings lag the truth, all of it now with NBFC numbers. From Phase 1, M1.09 (section 4.9) matters more here than anywhere so far, because Indian NBFCs report under Ind AS and banks don't. The three-stage expected-credit-loss model, EIR income recognition and the derecognition rules for assigned loan pools are all live analytical issues here. M1.10's Indian annual-report anatomy extends to the NBFC-specific notes: the borrowings schedule, the ALM maturity note, the stage-wise loan disclosure. From Phase 2 you need the quality-of-earnings reflexes of M2.06 to M2.08. Upfronted assignment income and coverage-ratio games are Schilit patterns wearing Indian clothes. From Phase 4, moat language. You'll find that an NBFC's true moat usually belongs to the liability side: who lends to you, at what price, in a bad month. That's a switching-cost-and-trust story you can measure in basis points.

Feeds into. M5.03 (insurance) completes the financials trilogy with the third funding model: float. Phase 7's macro modules plug straight in. The 2018 NBFC crisis is India's cleanest case study of the credit cycle and liquidity transmission, and the funding-freeze mechanics reappear in the GFC case work. Phase 8's rapid teardown uses this playbook whenever the two-minute triage says "non-bank lender," which in India is roughly a fifth of the credit system and a large share of listed mid-caps. Competency C6 continues its certification here. The workup format you build becomes the second page of the sector KPI cheat-sheet artifact.

Core teaching content

4.10 The red-flag panel

Here are five flags, each with its mechanism and the document where you catch it. Run all five on every NBFC. It takes fifteen minutes, and there are no exceptions.

1. ALM mismatch + CP reliance. This is the compound flag. Either one alone is a watch item, and together they're the IL&FS/DHFL signature. The mechanism is the whole ALM section. You catch it in the annual report's maturity-pattern note (re-derive it with zero rollover), the borrowings note (CP share >10 to 20%), the rating rationale's liquidity table (undrawn lines against 3-month outflows) and the LCR disclosure. One more tell corroborates it. A lender that stops disclosing granular ALM tables in a stressed year has answered your question.

2. AUM growth ≫ disbursement quality. Look for growth above ~40%, growth into segments/geographies where the lender has no vintage history, or growth through channels that separate the underwriter from the risk (DSA-sourced, fintech-partnered, top-up-heavy). The mechanism is an unseasoned book + compounding funding needs + adverse selection, as in KPI 1. You catch it by rebuilding the AUM bridge and by reading the deck's disbursement mix by product and vintage. Yield expansion during hyper-growth is another sign, because borrower quality is falling faster than the price is rising. Finally, recompute GNPA on the lagged denominator, the way you did for banks.

3. Spread squeeze. This is CoF rising faster than yields for 2+ quarters, the market repricing the lender's credit faster than the lender can reprice its borrowers. It comes before the visible crisis because bond investors read ALM notes before equity investors do. You catch it in the marginal borrowing costs in the deck, or in new NCD coupons on the exchanges, which are public. Track the CoF trend against the segment's yield ceiling too. Then cross-check whether a "stable NIM" is being manufactured by a riskier asset mix (flag 2) or by upfronted DA gains.

4. Low provisioning on unsecured. The first sign is Stage-3 coverage under ~40 to 50% on an unsecured or microfinance book, where recoveries after 90 dpd are structurally poor. Others are a thinning stage-2 disclosure, ECL "management overlays" being released into a growing book, or a fat impairment reserve, which is the IRAC floor catching a light ECL model. Here the mechanism is simple. Unsecured LGD is 70 to 90%, so under-coverage means losses denied. Calling it conservatism deferred gets it backwards. You catch it in the stage-wise disclosure (coverage by stage, by product), the ECL policy note's migration criteria, the impairment-reserve line, write-off velocity in the roll-forward.

5. Promoter pledge. This is the NBFC-specific reflexivity bomb, the India thread from your quality-of-earnings work turned into a weapon. The promoter pledges shares to fund other ventures, and then the stock falls. Margin calls force more pledging or invocation, and invocation dumps stock. The falling stock frightens the NBFC's lenders, who read pledge data as governance information, and CoF rises. It's a funding-cost spiral set off entirely outside the company. You catch it in the pledge disclosures in the shareholding pattern, published quarterly on the exchanges, where BSE/NSE show promoter pledge percentages. Treat >20 to 30% of promoter holding pledged as amber and >50% as red. Practice 11 runs the spiral arithmetic. Two corollary flags come from the same family. Watch for promoter-group entities borrowing from the NBFC (the RPT note), and for complex group structures where the listed lender is the group's funding organ. That's the IL&FS lesson, so compute double leverage as in Practice 8.

The meta-flag. Management that answers liquidity questions with capital ratios ("our CRAR is 19%, we are extremely comfortable") either doesn't understand its own business or hopes you don't. The correct answer to "how do you survive a shut market?" is denominated in months and crores of unencumbered, saleable, committed resources. Count what they count, and distrust adjectives.

Common mistakes & how experts think differently

Mistake 1: "CRAR is 19%, so it's safe." This is the most expensive confusion in the sector. Capital is loss-absorption. It answers "can the balance sheet survive assets going bad slowly?" Liquidity answers "can the company pay Tuesday's CP?" Those are different questions with different instruments, and the second kills faster. IL&FS entities and DHFL both reported compliant capital ratios within months of dying. Experts triage in survival order: the ALM note and borrowing mix first, capital second, profitability third. And they treat any management that answers a liquidity question with a capital ratio as a red flag in itself.

Mistake 2: reading GNPA with one universal threshold. A 4% Stage 3 is alarming at a prime HFC. At a vehicle financier it's routine, because delinquency there is a rhythm of rural cash flows and ultimate loss is a fraction of peak GNPA. At a gold lender, where LGD ≈ 0, it's almost meaningless. Meanwhile a low GNPA on a fast-growing unsecured book may just be the denominator effect plus fast write-offs. Experts read credit cost, LGD and coverage together against the segment and the growth rate. They rebuild the roll-forward before believing any ratio.

Mistake 3: admiring NIM. A 12% NIM is a price, and prices in lending are bids for risk. Don't mistake it for a moat. A high-NIM lender may simply be collecting tomorrow's credit losses as today's interest. Experts quote risk-adjusted NIM (NIM − credit cost, through-cycle). Then they look for the franchise where an NBFC actually keeps it, on the liability side, which means cost of funds against peers and how it behaves in a stress quarter. The lender with the boring 7% NIM and a 60 bp funding advantage that persisted through 2018 has the moat. The exciting 13%-NIM lender funded at CP rates has an option on the weather.

Mistake 4: taking reported ROE at face value and pricing it on P/E alone. An 18% ROE built from 4% ROA at 4.5× leverage and one built from 2% ROA at 9× are different species. The second has half the error margin against asset shocks and twice the funding dependence. Growth also has to be funded with book, through g = ROE × retention, dilution, or leverage, and there's no fourth option. So a consumer-company P/E slapped on a lender silently assumes free growth. Experts decompose ROE into ROA × leverage, translate any P/E back to P/B against ROE, and read the multiple as a claim about how long the spread lasts, as in the Falcon arithmetic.

Mistake 5: analyzing the on-book balance sheet and calling it the business. Direct assignment, securitization, and co-lending move assets off the balance sheet, along with their upfronted income. The origination risk culture, servicing obligations, and retained strips stay behind. Growth headlines use AUM while gearing optics use on-book, so a Q4 spike in DA "improves" both at once. Experts rebuild the AUM bridge and restate earnings with assignment gains spread over pool life. To them, a rising ratio of upfront gains to PBT, especially in a slowing year, is a Schilit chapter with an Indian accent.

Mistake 6: believing the published ALM statement. The disclosed maturity note usually embeds behavioral generosity: CP assumed to roll, prepayments assumed brisk, developer bullets assumed refinanced on schedule, "sanctioned" lines counted as money. Experts re-derive the table under the freeze convention: CP rollover zero, prepayments slowed, wholesale bullets stressed, only committed lines counted. Reality runs that version when it matters. If the re-derived first-year cumulative gap is negative and CP-funded, no other page of the annual report can rescue the thesis.

Mistake 7: extrapolating three benign years on an unsecured book. Unsecured retail in India has compounded through a decade with no full-scale unemployment cycle in it. Seasoning, survivorship (the failures merged or pivoted) and the denominator effect all flatter the observed loss history. Experts demand vintage/cohort curves. They price the business on through-cycle credit cost, adding 100 to 150 bp to the recent average to see what survives. They size the tail the way you sized it for microfinance. And they remember that the regulator now leans on this segment pro-cyclically, so the rules tighten just when the growth is best.

The deepest difference. A generalist sees an NBFC P&L and asks "how fast can earnings grow?" A financials specialist reads the same page and asks "who funds this, at what tenor, and who says yes in a bad month?" In this sector the income statement is the output of the liability side's permission. Every 2018 survivor had boring liabilities, and every casualty had exciting ones. Assets kill you slowly. Liabilities kill you overnight.

Worked examples

Worked example 1: Bhima Finance: the full KPI panel from raw statements

Bhima Finance (synthetic) is a vehicle financier. Here's what its Ind AS accounts show (₹ cr).

• P&L: interest income 1,710, interest expense 780, other income 90, opex 400, impairment 300, tax 25%.
• Averages: AUM 12,000, borrowings 9,200, total assets 12,900, net worth 2,250.
• Closing: AUM 13,100 (opening 10,900), gross on-book loans 12,300 + assigned pools 800, Stage 3 660, total ECL 380 of which Stage-3 ECL 264, borrowings 9,900, net worth 2,350, Tier 1 capital 2,140, Tier 2 100, RWA 12,600.
• A borrowing mix of banks 55%, NCDs 30%, securitization payables 10%, CP 5%.
• A 1-month cumulative ALM gap of +6% of outflows, and an LCR of 118%.
• Prior-year credit cost of 1.8%.

Build the panel with every number shown:

PAT comes from NII 930 + 90 − 400 − 300 = 320 PBT, then × 0.75 = 240.

The read. Bhima has the opposite disease from Tapti's. Its liability side is boring and safe (short vehicle assets, positive gaps, 5% CP). So this is a value case about asset quality, and survival isn't the question. The trouble is a cluster: Stage 3 of 5.4% with only 40% coverage, while credit cost is still rising. Now put the provisions where they belong. Raising Stage-3 coverage to 50% needs 0.10 × 660 = 66 of extra provisions ⇒ PAT 240 − 66×0.75 = 190 ⇒ ROA 1.48%, ROE 8.5%. Even normalizing credit cost back to 2.0% only gets PAT to (620 − 240)×0.75 = 285 ⇒ ROE 12.7%, still below a ~14% cost of equity. The verdict is solvent, liquid, and currently value-destructive. Any equity thesis here is entirely a bet on the CV cycle turning plus NIM repair, so the stock should price below book until slippage turns. Each quarter, watch credit cost against that 2.0% through-cycle line and look past the GNPA headline. Vehicle GNPA is a rhythm, and cost is the truth.

Worked example 2: five segments on one page: same ROE, different nightmares

Build each segment's economics as a stack, per ₹100 of AUM. These are illustrative mid-cycle calibrations from the segment lenses, and every line is a % of AUM:

Each ROE multiplies the unrounded ROA by the leverage, and the displayed ROA is rounded. Gold shows why that matters. Its ROA is 4.95, shown as 5.0, and 4.95 × 3.5 = 17.3 where 5.0 × 3.5 would give 17.5. Two-tenths of a point on one segment is nothing. But multiply rounded figures into a capital ratio or a payback year, and that's how a model acquires headroom it doesn't have.

Three lessons the table teaches better than paragraphs:

1. Competition equalizes ROE, and leaves ROA free to differ. Yields span 9.6% to 22%, a 12-point range, yet every model lands at 15 to 20% ROE. Capital chases returns until each segment's ROA × sustainable-leverage product converges toward the market's required return. So never rank lenders by yield, NIM, or even ROA in isolation. Each is one gear in a machine tuned to the same output.

2. What differs is the construction, and the construction is the risk. The HFC's 15% is a low-margin annuity levered 9×, so its earnings variance is tiny and its funding variance is as large as it gets. The MFI's 16.7% is a fat margin at low leverage. Its funding variance is modest and its earnings variance extreme. Gold's 17.3% is serene on both sides but exposed to regulatory and gold-price jumps. Unsecured's 20.3% is the best number with the least history. Identical ROEs don't make identical investments, so ask which side of the balance sheet carries the variance.

3. Price in the tail. Re-run the MFI in an event year, with credit cost at 8.5 instead of 2.5. PBT = 7.8 − 8.5 = −0.7 ⇒ PAT ≈ −0.5 ⇒ ROA −0.5%, ROE ≈ −2.2%. Suppose a year like that arrives once in five, which is roughly the sector's record (AP 2010, demonetization 2016, COVID, the 2024-25 stress). Then through-cycle ROE = (4 × 16.7 − 2.2)/5 = 12.9%, well below the 16.7% you started with. At r = 14%, g = 5%, justified P/B on the good-year ROE = (16.7−5)/(14−5) = 1.30×, and on through-cycle ROE = (12.9−5)/(14−5) = 0.88×. That's a one-third valuation error from simply forgetting that the tail is part of the mean. The same logic applies to unsecured consumer in a softer dose, and there the tail isn't in anyone's observed data yet.

Worked example 3: DHFL by the numbers: the melting ice cube

All figures here are illustrative and order-of-magnitude, modeled on public reporting from September 2018 to June 2019, so pull the actual disclosures for precision. The lesson is the mechanism.

Position, September 2018. The loan book is ~₹1,00,000 cr, with individual mortgages ~70% and developer/wholesale ~30%. Borrowings are ~₹85,000 cr. Of that, CP is ~₹9,000 cr, and another ~₹24,000 cr of NCD/bank maturities fall due within 9 months ⇒ ~₹33,000 cr of obligations by mid-2019. Cash and liquid buffers are ~₹3,500 cr. Monthly customer inflows (EMIs + prepayments + developer receipts) are ~₹2,300 cr. Gearing is ~9×, and reported CRAR is comfortably above the floor. Keep Mistake 1 in mind.

Step 1: the arithmetic of a shut market. After IL&FS, assume zero rollover, because that's what happened. Nine months of organic sources come to 3,500 + 9 × 2,300 = ~₹24,200 cr, against ~₹33,000 cr due. That leaves an ~₹8,800 cr hole that only asset sales can fill, and that's before a single rupee of new lending, opex, or the deposits that also started leaving.

Step 2: the fire sale that looks like strength. For months DHFL "met all obligations", and the press releases were technically true. It managed it by selling and assigning its best retail pools (₹30,000+ cr over the period) and halting disbursements. Watch what that does to the residual balance sheet. The saleable 70%, granular priority-sector-eligible mortgages, shrinks. The unsaleable 30% stays, because no one buys lumpy developer credit in a freeze. The book slides from ~₹1,00,000 toward ~₹65,000 cr while its developer share rises from ~30% toward ~45%. DHFL was performing adverse selection on itself.

Step 3: the P&L dies of the cure. Interest income falls ~35% with the book. Interest expense falls more slowly, because CP was the cheapest funding and it's the part that disappeared. What's left is the expensive term debt. Opex is sticky. And the halted developer projects stop paying. A construction loan is only "performing" while the next tranche arrives, so stopping disbursements manufactures your own NPAs. Pre-provision profit collapses just as provisions explode.

Step 4: the solvency arithmetic completes. The developer book is ~₹28,000 cr, grown as a share of the book (Step 2). Stress 25% of it to default at 50% LGD ⇒ ~₹3,500 cr of losses. Add the fictitious-asset write-offs the forensic audit later surfaced, ~₹14,000 cr of "Bandra Books" retail that never existed. Set all that against net worth of ~₹8,000 to 10,000 cr ⇒ equity < 0, long before the June 4, 2019 interest default made it official. Recovery in IBC came to ~₹38,000 cr against ~₹87,000 cr of claims.

Here's the checklist to carry away from the walk:

• Compute months-of-survival = (cash + committed lines + N months of collections) ÷ N months of obligations, at zero rollover. DHFL's was ~9 months, and it used every one of them.
• Ask which assets are saleable in a freeze, and what the balance sheet looks like after they're gone.
• Treat "we have met all obligations" during a freeze as a countdown announcement. It isn't reassurance.
• Remember that solvency computed on stated book (Mistake 1's twin) is meaningless when the book contains a fraud you can't see. So the fragility flags (funding mix, ALM, growth-vs-quality) have to do the work the audit can't.

Worked example 4: Chambal Housing Finance: build the ALM ladder, then price the freeze (India, ₹ cr)

Chambal Housing Finance (synthetic) is a mid-size HFC. Its loan book is ₹30,000. Individual home loans make up 22,000 of it, contractually 15 to 20 years but behaviorally ~7 to 9 because of prepayment. LAP + developer/construction finance make up 8,000 (lumpy, bullet-heavy). Borrowings are ₹26,400: CP 4,000 (15.2%), NCDs 15,000, bank term loans 6,000, NHB refinance 1,400. Cash + liquid investments come to 900 and fixed/other assets to 300. Other liabilities are 700 and net worth is 4,100. That makes gearing (borrowings ÷ net worth) 6.4×, unremarkable for an HFC. The balance sheet foots: 900 + 30,000 + 300 = 26,400 + 700 + 4,100 = 31,200. ✓

Here's the structural liquidity statement on a contractual basis, with CP rollover set to zero (the freeze convention) and the first bucket aggregated to a month:

Read it in four moves:

1. Plumbing. The final cumulative gap ties: 30,900 − 27,100 = +3,800 = net worth 4,100 − unbucketed fixed assets 300. ✓ If a published statement doesn't roughly foot to net worth, an outflow has been dropped, often undrawn disbursement commitments.
2. Shape. Assets are long, with 60%+ of receipts beyond three years, and liabilities are front-loaded, with all CP inside six months. Again it's borrow-short-lend-long drawn in numbers, so the gap is negative early and positive late.
3. Caps. The ≤1m cumulative gap is −30.1% of the month's outflows, against a 10 to 20% granular tolerance. That's a breach, and the 1-7-day sub-bucket where the CP clusters looks worse.
4. The hole. The twelve-month cumulative gap is −3,310, so Chambal must raise ₹3,310 crore of new money in a year merely to stand still.

Now run the freeze. Say the CP market shuts because someone else defaulted. Count only committed mitigants. Cash is already in the ladder, so add the ₹700 of committed undrawn lines. The running position = cumulative gap + 700, which gives +170 at month 1, −1,960 by month 3, −2,610 by month 12. Take month 1 first. Chambal survives it on a razor-thin ₹170 cushion. By month 3 it's ₹1,960 underwater and has to sell its cleanest unencumbered mortgages, so the good assets leave first and the quality of what's left degrades. It halts disbursements at the same time. That stalls the developer borrowers whose "scheduled" receipts the ladder assumed, and the hole gets deeper. At zero rollover the survival horizon is ~1 to 2 months before asset sales are forced, and Chambal needs ~₹2,610 crore of pool sales across the year to reach month 12. It's the same melting ice cube. Illiquidity arrives first, and insolvency is the paperwork.

Price the fix. Re-fund the identical book safely. Term the ₹4,000 of CP into 5-year NCDs (~+50 bp) and hold ₹1,500 more liquidity buffer (negative carry ~200 bp). Cost = 4,000 × 0.50% + 1,500 × 2.0% = 20 + 30 = ₹50 crore pre-tax ≈ ₹37.5 crore post-tax ≈ 12 bp of ROA ≈ 0.9 ROE points at Chambal's leverage. The early buckets turn positive and the freeze becomes a non-event. Survival cost about one point of ROE. So when a lender out-earns its peers on identical assets, read the ALM note before you applaud the P&L.

Worked example 5: Beacon Consumer Finance: a US non-bank lender's spread, leverage, and runway (US, $m)

Beacon Consumer Finance (synthetic) is a US branch-based personal-installment lender, the OneMain archetype. It takes no deposits and is funded entirely in the capital markets. All figures are $m and illustrative.

• Averages: net receivables 22,000, interest-bearing debt 18,500.
• P&L: interest income 5,060, interest expense 1,480, other income (insurance products, fees) 600, opex 1,900, provision for credit losses 1,500 (of which net charge-offs 1,430).
• Balance sheet: net receivables 22,000 + cash 900 + other assets 600 = 23,500. Funding comes from debt 18,500 (securitizations/ABS 9,000, unsecured senior notes 8,000, drawn conduit facilities 1,500), other liabilities 1,300 and equity 3,700.
• Allowance for credit losses 2,420 (CECL). Tax 25%.

Spread walk. Yield = 5,060 ÷ 22,000 = 23.0%. Cost of funds = 1,480 ÷ 18,500 = 8.0%, so the spread is 15.0%. NII = 5,060 − 1,480 = 3,580 → NIM = 3,580 ÷ 22,000 = 16.3%. Credit cost = 1,500 ÷ 22,000 = 6.8%, so risk-adjusted NIM = 16.3 − 6.8 = 9.5%. That's the number that survives contact with a US subprime borrower, and the fair figure to set beside an Indian NBFC's.

Returns. PBT = 3,580 + 600 − 1,900 − 1,500 = 780 → net income = 780 × 0.75 = 585. ROA = 585 ÷ 23,500 = ~2.5%, leverage = 23,500 ÷ 3,700 = 6.35×, and ROE = 585 ÷ 3,700 = 15.8% (ROA × leverage = 2.49% × 6.35). Read that against Worked example 2. It's the unsecured-consumer construction, a fat ~2.5% ROA levered a modest 6×, so earnings variance is high and funding variance is what you watch.

The CECL tell. The allowance is 2,420 ÷ 22,000 = 11.0% of receivables, against an annual charge-off rate of 1,430 ÷ 22,000 = 6.5%. So the reserve is ~1.7× a year of losses, because CECL books lifetime expected loss on day one. It's the fastest of the three recognition regimes you met earlier. That allowance is ~65% of equity and optically depresses book value. It's the accounting telling the truth early, and it doesn't signal distress. Don't mistake a US lender's heavy reserve for the under-provisioning flag you hunt in India.

The runway: the US answer to the ALM ladder. Beacon files no RBI-style ladder, but it does disclose a liquidity runway. Cash 900 + committed, multi-year, undrawn conduit capacity 6,500 = 7,400. That covers the 1,200 of 12-month unsecured maturities 6.2×, and the 3,800 of 24-month unsecured maturities 1.9×, before you count receivable run-off. The conduits are committed and secured on the receivables, so nobody can vote them away the way 90-day CP gets voted away. So Beacon can operate 12+ months with the unsecured market shut. It's the structural fix the 2018 Indian casualties lacked, and the deposit-funded converts, Ally and Synchrony, went further still. Same lesson, both hemispheres. A non-bank survives a freeze only to the extent its funding is committed and collateralized instead of confidence-sensitive.

Practice set

Work in ₹ crore unless stated, with tax at 25% throughout. The set moves from guided problems (1 to 5) to independent ones (6 to 11) and then a timed one (12).

Problem 1 (guided). Vetravati Finance: average AUM 20,000, interest income 2,900, average borrowings 15,500, interest expense 1,300, fee income 160, opex 720, impairment 640. Compute yield, cost of funds, spread, NIM, credit cost, risk-adjusted NIM, then ROA (average assets 21,000), and explain why the "fat NIM" verdict would mislead.

Solution. Yield = 2,900/20,000 = 14.5%. CoF = 1,300/15,500 = 8.39%, so the spread is 6.11%. NIM = (2,900 − 1,300)/20,000 = 1,600/20,000 = 8.0%. Credit cost = 640/20,000 = 3.2%, which makes risk-adjusted NIM = 8.0 − 3.2 = 4.8%. PBT = 1,600 + 160 − 720 − 640 = 400 ⇒ PAT = 300 ⇒ ROA = 300/21,000 = 1.43%. That's mediocre, despite a NIM nearly 3× a prime HFC's. Opex (3.6% of AUM) and credit cost (3.2%) ate almost all of the 500 bp of extra yield over a mortgage book. High-yield lending is high-cost lending twice over. NIM is vanity, and risk-and-opex-adjusted ROA is sanity. At 1.43% ROA this lender needs 10×+ leverage to clear a 14% cost of equity, and its asset quality can't safely carry that. The panel's rules interlock here. A weak KPI 9 rules out fixing it with KPI 7.

Problem 2 (guided). Kopili Finance, closing figures: borrowings 13,500, net worth 3,000, Tier 1 capital 2,850, Tier 2 150, RWA 15,000, expected retained earnings next year 400. (a) Gearing and CRAR, classified. (b) How much RWA growth does the capital position support before touching the 15% floor, excluding then including retention? (c) Its 1-month cumulative ALM gap is −18% of cumulative outflows. Which constraint binds, and what would you say to a CFO who answers ALM questions with the CRAR?

Solution. (a) Gearing = 13,500/3,000 = 4.5×, healthy (3 to 6). CRAR = (2,850+150)/15,000 = 20.0% (Tier 1 19.0%), comfortable against 15%/10%. (b) Surplus capital = (20% − 15%) × 15,000 = 750 ⇒ supports 750/0.15 = 5,000 of new RWA (+33%). With retention it's (750 + 400)/0.15 = 7,667 ⇒ +51%. (c) Liquidity binds. Capital headroom of 33 to 51% is irrelevant to a first-month funding hole 8 points beyond any regulatory tolerance. CRAR measures loss absorption over quarters, while the −18% gap measures cash shortfall over days. Tell the CFO: "You've answered a survival question with a solvency ratio. Walk me through the 1 to 30 day buckets at zero CP rollover, the committed undrawn lines, and the unencumbered pools you can sell inside four weeks." A 20% CRAR corpse is still a corpse, as IL&FS proved.

Problem 3 (guided). Sharavati Finance, this year: opening on-book loans 15,000, opening off-book (assigned) pools 1,000, disbursements 9,000, collections/runoff on the on-book book 5,200, write-offs 300, pools sold via DA during the year 1,500, runoff of off-book pools 400. The investor deck headlines "AUM +19%, gearing improved." (a) Compute closing on-book, closing off-book, closing AUM, and both growth rates. (b) Which base do gearing and CRAR use, and what did DA do to them? (c) Name the two follow-up checks the deck's headline demands.

Solution. (a) Closing on-book = 15,000 + 9,000 − 5,200 − 300 − 1,500 = 17,000 (+13.3%). Closing off-book = 1,000 + 1,500 − 400 = 2,100. Closing AUM = 19,100 against opening 16,000, which is +19.4%. (b) Gearing and CRAR are computed on the on-book balance sheet. Selling 1,500 of pools shrank the measured denominator/RWA while the franchise kept growing, along with the serviced customer base and the retained risk strips. So the deck's "gearing improved" is partly an accounting relocation. It isn't deleveraging. (c) The first check is timing. If the DA is concentrated in Q4, it's window-dressing the year-end ratios. The second is earnings quality. How much upfront assignment gain entered other income (restate it as in Problem 7)? And what risks were retained (MRR strip, credit enhancement, servicing obligations)? The securitization note tells you. An AUM headline that outgrows the on-book by 6 points every year belongs to a business that's becoming an originator-servicer. That's a different business, and not a worse one (think of co-lending economics), but it has to be analyzed as one.

Problem 4 (guided, ALM reading drill I). Arkavati Finance's structural liquidity statement, condensed (₹ cr):

Mitigants: HQLA 150, committed undrawn lines 350, unencumbered securitizable retail pools ~1,200 (4 to 8 weeks to monetize, in a normal market). Compute per-bucket gaps, cumulative gaps, and cumulative gap as % of cumulative outflows. Then give the regulatory verdict and the survival read.

Solution. The gaps are −200, −600, +200, +400, and the cumulative gaps −200, −800, −600, −200. Cumulative outflows run 1,100 / 3,000 / 4,500 / 7,100, so the cumulative gap % is −18.2%, −26.7%, −13.3%, −2.8%. On the regulatory verdict, the ≤1m bucket at −18.2% breaches the granular caps. The 10%/10%/20% caps apply to the 1-7/8-14/15-30-day sub-buckets, and a month aggregating to −18% can't have compliant sub-buckets unless outflows are all month-end-loaded, so demand the granular table. For the survival read, the trough is −800 at 3 months against immediate mitigants of 150 + 350 = 500. That leaves a −300 hole, and the only way to fill it is selling pools. Selling takes the same 4 to 8 weeks you're trying to survive, in a market that may be the reason you're stressed. A cumulative gap still negative at 12 months (−200) means the year as a whole runs on rollover. Classify it as stress, survivable only if the pool sale executes early. The right management action is to sell pools now, pre-need, and term out the 1-3m wall. If the borrowing mix shows the outflow wall is CP, move the classification to red. Prudence will shut that door before panic does.

Problem 5 (guided). Musi Finance: borrowings 15,000, of which CP 3,000 (20%) at an average original tenor of 75 days. It holds HQLA of 400 and committed undrawn lines of 500. (a) Compute annual gross CP rollover volume and the CP maturing in any 30-day window. (b) Does the company survive a 30-day market shutdown without touching assets? (c) What CP stock would its buffers safely support, and what does terming out the excess cost if 3-year NCDs price 120 bp over CP (average AUM 18,000)?

Solution. (a) Annual rollover ≈ 3,000 × 365/75 ≈ 14,600. So the market has to say "yes" to fifteen thousand crore of refinancing events a year. In any 30 days, 3,000 × 30/75 = 1,200 matures. (b) Immediate resources = 400 + 500 = 900 < 1,200, so the answer is no. A 30-day freeze forces ~300 of distressed asset sales even before NCD/bank maturities and opex stack on. (c) Safe CP ≈ resources × (75/30) = 900 × 2.5 = 2,250, or 15% of borrowings. Stay prudently below that, since the same buffers must cover other outflows too, so target ≤10%. Terming out 1,000 costs 1,000 × 1.2% = 12/yr pre-tax ≈ 9 post-tax ≈ 5 bp of ROA (9/18,000). Five basis points of ROA to remove a third of the refinancing cliff is the cheapest insurance in finance. So a lender that doesn't buy it is telling you something about how it thinks, or about what its ROE targets require it to pretend.

Problem 6. Purna Finance under Ind AS: opening gross loans 15,800 (gross Stage 3 of 800), opening total ECL 520 (Stage-3 ECL 360). During the year: fresh slippage into Stage 3 460, upgrades/recoveries out of Stage 3 120, write-offs 280 (fully provided), P&L impairment charge 310, closing gross loans 17,200, closing Stage-3 ECL 430, average AUM 16,500. Compute closing Stage 3 (stock and ratio), closing total ECL, Stage-3 coverage, net Stage-3 ratio, credit cost, slippage run-rate, and give the two-line verdict.

Solution. Closing Stage 3 = 800 + 460 − 120 − 280 = 860 ⇒ ratio = 860/17,200 = 5.0%. Closing total ECL = 520 + 310 − 280 = 550. So Stage 1&2 ECL = 550 − 430 = 120, i.e., 0.73% cover on the 16,340 performing book. Stage-3 coverage = 430/860 = 50%. Net Stage 3 = (860 − 430)/(17,200 − 430) = 430/16,770 = 2.56%. Credit cost = 310/16,500 = 1.88%. Slippage = 460/(15,800 − 800) = 3.07% of the opening standard book. The verdict is that the headline ratio "improved" only modestly, while write-offs did most of the containment (280 of the 400 gross reduction) and fresh slippage ran above 3%. Judge the inflow pipe and look past the stock. It's the same roll-forward habit you built with banks, unchanged under new labels. Coverage at 50% is adequate only if this book is secured. On an unsecured book it's a flag.

Problem 7. Tawa Finance reports PAT 400, PBT 533. Other income includes an 80 upfront gain from assigning a 1,000 loan pool (excess spread ~3%/yr, expected life ~3 years). (a) Restate PAT as if the spread accrued over life (~30/yr pre-tax). (b) Compute the DA-gain share of PBT. (c) When is upfront DA income not a red flag, and what pattern makes it one?

Solution. (a) Restated PAT = 400 − 80(0.75) + 30(0.75) = 400 − 60 + 22.5 = 362.5, so reported earnings overstate the run-rate by ~10%. (b) 80/533 = 15% of PBT from one derecognition event. (c) It's not a flag when a lender assigns a steady volume every year. That lender is in steady state. Its upfront and accrual methods converge, and DA is just a sensible funding/PSL-monetization channel. Flag patterns look different. Watch for step-ups in DA volume, since each increment borrows from future years' income. Watch for gains clustering in the very quarters that "meet guidance," and for growing dependence, with the DA share of PBT trending up as core spread stalls. And watch for selling while retaining most of the economic risk through fat credit enhancements, so the income is recognized and the risk is kept. The rule is to model the lender's earnings ex-upfront-gains first. Whatever story survives that is the real one.

Problem 8. Kolar Holdings (an IL&FS miniature): holdco net worth 5,000, holdco borrowings 3,000, holdco assets = investments in subsidiaries' equity 6,500 + other assets 1,500. The subsidiaries (wholly owned) carry combined equity 6,500 and external debt 32,500. Compute (a) holdco standalone gearing, (b) the double-leverage ratio and (c) consolidated gearing. Then (d) explain how every entity can simultaneously report a compliant capital ratio while the group is lethal, and name the two documents that catch it.

Solution. (a) Standalone gearing = 3,000/5,000 = 0.6×, which looks pristine. (b) Double leverage = investments in subs ÷ holdco net worth = 6,500/5,000 = 130%. The holdco has borrowed 1,500 and called it equity downstairs. Anything >100% flags it, and >120% is serious. (c) Consolidated external debt = 3,000 + 32,500 = 35,500 against true group equity of 5,000, which is 7.1×. That's a risky band, and it's invisible at every individual entity. Each sub shows 5× against "equity" that's partly the holdco's debt, and the holdco shows 0.6×. (d) Each entity's CRAR uses its own capital base, so the same rupee of ultimate equity gets counted at every layer it passes through. The name for this is capital cascading. Two documents catch it. One is the consolidated financial statements read against the standalone ones, since any group whose consolidated gearing ≫ every standalone gearing is cascading. The other is the rating rationale/CIC disclosures that compute group-level leverage. IL&FS ran this structure ~350 entities deep, and the arithmetic above is all the sophistication the fraud required.

Problem 9 (ALM reading drill II). (a) Kabini Gold Finance has a loan book of 8,000 at ~5-month average tenor, with inflows of ≤1m 1,600, 1-3m 3,200, 3-6m 3,200. Cash is 500, and borrowings are 6,000 with maturities ≤1m 150, 1-3m 450, 3-6m 700 (rest beyond). Opex runs 40/month. Compute the ≤1m gap and characterize the profile. (b) Return to Tapti from the ALM section. Replace its 5,400 of CP with term debt (≤1m NCD outflow unchanged at 300) and raise the cash buffer from 1,000 to 2,500 (funded by 5-year NCDs). Recompute the first four cumulative gaps and the annual cost if the term-out prices +50 bp on 5,400 and the extra 1,500 of buffer negative-carries 200 bp. State the moral in one sentence.

Solution. (a) For ≤1m, inflows = 500 + 1,600 = 2,100 and outflows = 150 + 40 = 190 ⇒ gap +1,910. Cumulative gaps stay positive throughout, with 1-3m adding +3,200 − 530 = +2,670 more. A gold lender's assets are shorter than its liabilities, so its ALM statement is an HFC's with the sign flipped. So gold NBFCs barely noticed September 2018, while HFCs died of it. The freeze survival question ("if new money stopped, how long do you live?") answers itself when the whole book comes back as cash in five months. (b) The new ≤1m bucket has inflows = 2,500 + 420 = 2,920 and outflows = 300 + 100 + 70 = 470 ⇒ +2,450. For 1-3m, 940 − (900+200+140) = −300 ⇒ cumulative +2,150. For 3-6m, 1,760 − (1,500+300+210) = −250 ⇒ +1,900. For 6-12m, 3,920 − 4,120 = −200 ⇒ +1,700. Every bucket passes with room. The cost is 5,400 × 0.5% = 27, plus 1,500 × 2.0% = 30 ⇒ 57 pre-tax ≈ 43 post-tax ≈ 10 bp of ROA ≈ 0.9 ROE points at Tapti's leverage. The moral is that about one point of ROE separated the 2018 casualties from the survivors. Fragility was a choice, paid for in multiple expansion until it was paid for in everything.

Problem 10. Betwa Microfinance, normal year (% of average AUM): NIM 12.6, fees 0.8, opex 6.0, credit cost 2.5; leverage 4.5× (assets/equity). Event years hit once in five, with credit cost 8.5 and other lines unchanged. (a) ROA/ROE in a normal and an event year. (b) Through-cycle ROE. (c) Justified P/B at r = 14%, g = 5% priced on the normal year vs the cycle, and the sizing lesson.

Solution. (a) In a normal year PBT = 12.6 + 0.8 − 6.0 − 2.5 = 4.9 ⇒ PAT 3.675 ⇒ ROA 3.7%, ROE = 3.675 × 4.5 = 16.5%. In an event year PBT = 4.9 − 6.0 = −1.1 ⇒ PAT −0.825 ⇒ ROA −0.8%, ROE −3.7%. (b) Through-cycle ROE = (4 × 16.5 − 3.7)/5 = 12.5%. (c) Justified P/B on the normal year = (16.5 − 5)/(14 − 5) = 1.28×, and on the cycle = (12.5 − 5)/9 = 0.83×. Pricing off good years overvalues the franchise by ~50%, and that's before you ask whether an event year impairs growth capital. It does. The −3.7% ROE year also consumes ~4% of net worth and pauses AUM growth, so the true through-cycle number is below 12.5%. Here's the sizing lesson. A business whose bad year is correlated across its whole book (politics, weather, borrower-behavior waves) deserves a smaller position at a bigger discount than its average-year economics suggest. In Phase 9 the phrase for this is "the tail sets the size."

Problem 11. Wainganga Finance's promoter owns 40% of the company and has pledged 55% of that holding (22% of the company) against a personal-venture loan at 50% loan-to-value, struck when the stock was ₹200. (a) Per 100 company shares, compute the loan and the collateral. (b) The stock falls 40%: compute the new LTV and the top-up shares needed to restore 50% LTV; can the promoter meet a second 20% fall? (c) Trace the reflexive loop to the NBFC's cost of funds, and name where you monitor this quarterly.

Solution. (a) Pledged 22 shares × ₹200 = 4,400 of collateral, so the loan = 2,200. (b) At ₹120, collateral = 2,640 ⇒ LTV = 2,200/2,640 = 83%. Restoring 50% needs collateral of 4,400 ⇒ top-up = 1,760/120 ≈ 14.7 shares. That leaves 36.7 of the promoter's 40 pledged (92%). Now take a further 20% fall to ₹96. Collateral = 36.7 × 96 = 3,523 against 4,400 needed ⇒ shortfall ≈ 877/96 ≈ 9.1 shares. With only 3.3 unpledged, that means invocation. The pledgee sells ~9% of the company into a falling market. (c) The loop runs like this. Invocation makes headlines and the stock gaps down. The NBFC's bond investors read promoter distress as governance risk and reprice its debt. CoF rises and the spread compresses. ROE falls and equity falls further. Then the next margin call arrives. Notice what happened. The operating company's funding cost was set on fire by a loan it never took. Monitor the quarterly shareholding pattern's pledge column on BSE/NSE, plus SAST pledge disclosures, against the bands above (>20 to 30% of promoter holding pledged = amber, >50% = red). Any rising trend in a falling market means the spiral is already turning.

Problem 12 (timed: 25 minutes, exam conditions). Kosi Finvest, FY summary (₹ cr, Ind AS). Averages: AUM 30,000 (opening 24,000, closing 36,000), borrowings 23,500, total assets 31,500, net worth 5,000. P&L: interest income 4,440, interest expense 2,090, other income 520 (of which upfront DA gains 210), opex 1,000, impairment 690. Closing: borrowings 28,300, net worth 5,400, Tier 1 capital 4,200, Tier 2 300, RWA 29,000, gross loans 30,600, Stage 3 700, Stage-3 ECL 266. Borrowing mix: banks 38%, NCDs 32%, CP 22%, other 8%. Prior-year CoF was 8.1%. ALM: 1-month cumulative gap −14% of cumulative outflows, LCR 92%. Unsecured = 35% of AUM, and opening gross loans were 20,400. Market cap 8,600. Deliver the ten-KPI panel with classifications, earnings restated for (i) normalized credit cost of 3.0% and (ii) ex-DA gains, a five-line verdict naming the two deadliest issues, and a fair-multiple range.

Solution. Yield = 4,440/30,000 = 14.8%. CoF = 2,090/23,500 = 8.9%, up from 8.1%. That's an 80 bp spread squeeze while the book grew, the market repricing Kosi faster than Kosi reprices its borrowers. NIM = 2,350/30,000 = 7.8%. AUM growth = 36,000/24,000 − 1 = +50%, red. C/I = 1,000/2,870 = 34.8%, which is efficient (fast books usually are, briefly). PPOP = 2,870 − 1,000 = 1,870. Credit cost = 690/30,000 = 2.3%. But lagged Stage 3 = 700/20,400 = 3.4% against a headline 700/30,600 = 2.3%, so the denominator is doing the reassuring. Coverage = 266/700 = 38%, low for a 35%-unsecured book. PAT = (1,870 − 690) × 0.75 = 885 ⇒ ROA = 885/31,500 = 2.81% ("strong"), ROE = 885/5,000 = 17.7%. Gearing = 28,300/5,400 = 5.2× (ok). CRAR = 4,500/29,000 = 15.5%, 50 bp above the floor while growing 50%. At this pace RWA outruns 17.7% internal accrual within months, so a dilutive raise isn't optional. ALM at −14% is breach territory, LCR at 92% < 100% floor is a breach, and CP at 22% is red.

Now the restatements. (i) At 3.0% credit cost, impairment is 900 ⇒ PAT = (1,870 − 900) × 0.75 = 727 ⇒ ROA 2.31%. (ii) Ex-DA, PAT = 885 − 210(0.75) = 727. With both, PAT = (1,870 − 900 − 210) × 0.75 = 570 ⇒ ROA 1.81%, ROE 11.4%. So roughly one-third of the reported 17.7% ROE is manufactured, from pulled-forward income + under-provisioning + denominator optics.

The verdict is that four of five red flags are lit (ALM+CP, growth ≫ quality, low coverage on unsecured, spread squeeze), with only promoter pledge unexamined. This is a 2018-cohort balance sheet operating after 2018. The deadliest pair is CP 22% against a −14% short gap, since that funding can be voted away in 90 days, and capital at 15.5% against 50% growth, since the rescue raise dilutes just as the multiple compresses. On valuation, ABV = 5,400 − (700 − 266) = 4,966 ⇒ market's P/B = 8,600/5,400 = 1.59×, and P/ABV = 1.73×. For a real ROE of ~11%, justified P/ABV ≈ (11 − 5)/(14 − 5) ≈ 0.7×. Kosi's stock is pricing the manufactured ROE. The fair range is ~0.7 to 0.9× ABV until the CP terms out, coverage rebuilds to 50%+, and growth decelerates into the healthy band. Ask one question before taking any position, in either direction: can it raise 3 to 5 year money, this quarter, at a spread that leaves the model alive? (Score yourself: 1 point per KPI computed and classified, 10 max; 3 for the two restatements; 3 for a verdict naming the CP/ALM + capital pair. ≥12/16 to pass.)

Problem 13 (guided). Damodar Finance: average AUM 25,000, interest income 3,500, average borrowings 20,000, interest expense 1,700, impairment 450, borrowing mix banks 40% / NCDs 33% / CP 21% / other 6%. Last year's cost of funds was 7.9%. Compute yield, cost of funds, spread, NIM, credit cost, and risk-adjusted NIM. Then classify the CP share and say what the year-on-year cost-of-funds move signals.

Solution. Yield = 3,500/25,000 = 14.0%. CoF = 1,700/20,000 = 8.5%, so the spread is 5.5%. NIM = (3,500 − 1,700)/25,000 = 1,800/25,000 = 7.2%. Credit cost = 450/25,000 = 1.8%, and risk-adjusted NIM = 7.2 − 1.8 = 5.4%. CP at 21% is red, over the 20% threshold. That's refinancing risk you aren't paid for unless the assets are sub-6-month. The CoF moved from 7.9% to 8.5%, +60 bp in a year. If yield didn't rise as much, that's a spread squeeze, the bond market repricing Damodar's own credit faster than Damodar can reprice its borrowers. It's the flag that comes before the visible crisis (flag 3 on the red-flag panel). Overall, the risk-adjusted spread is adequate today, but the story is the funding trend and the CP reliance. Read the maturity note before the P&L.

Problem 14 (guided, ALM reading drill III). Godavari Finance's structural liquidity statement, condensed (₹ cr):

Mitigants: HQLA 200, committed undrawn lines 400, unencumbered securitizable retail pools ~1,500 (4 to 8 weeks to monetize, normal market). Compute per-bucket gaps, cumulative gaps, and cumulative gap as % of cumulative outflows. Then give the regulatory verdict and the survival read.

Solution. The bucket gaps are −450, −500, +200, +500, and the cumulative gaps −450, −950, −750, −250. Cumulative outflows run 1,250 / 2,850 / 4,150 / 6,550, which gives cumulative gap percentages of −36.0%, −33.3%, −18.1%, −3.8%. The regulatory verdict is a breach. At −36.0% the ≤1m bucket can't satisfy the 10/10/20 granular caps, so demand the 1-7/8-14/15-30-day split. For survival, the trough is −950 at three months against immediate mitigants of 200 + 400 = 600. That leaves a −350 hole, and selling pools is the only way to fill it. A sale takes the same 4 to 8 weeks you're trying to survive, in a market that may be why you're stressed. With the cumulative gap still negative at 12 months (−250), the year runs on rollover. Classify it as stress, survivable only if the pool sale is executed early and pre-need. If the outflow wall proves to be CP, move it to red, because prudence shuts that door first.

Problem 15. Narmada HFC, closing figures: borrowings 32,000, net worth 4,000, Tier 1 capital 3,700, Tier 2 300, RWA 22,000, expected retained earnings next year 350, individual home loans 75% of the book, developer/LAP 25%. Its 1-month cumulative ALM gap is −16% of cumulative outflows. (a) Gearing and CRAR, classified. (b) RWA growth headroom before the 15% floor, excluding then including retention. (c) Does the prime-HFC gearing exception apply here, and which constraint actually binds?

Solution. (a) Gearing = 32,000/4,000 = 8.0×, at the top of the prime-HFC tolerance (8 to 10×). For an ordinary NBFC that number would be deep in the red zone. CRAR = (3,700 + 300)/22,000 = 18.18% (Tier 1 16.82%), comfortable against 15%/10%. (b) Surplus capital = (18.18% − 15%) × 22,000 ≈ 700 ⇒ supports 700/0.15 ≈ 4,667 of new RWA (+21.2%). With retention, (700 + 350)/0.15 = 7,000 (+31.8%). (c) Only partly. The 8 to 10× tolerance belongs to prime individual mortgages, with 20 to 50 bp of through-cycle loss. Narmada runs 25% developer/LAP, which is lumpy with 30 to 50% LGD. So a quarter of that 8× leverage rests on an asset class that doesn't earn the tolerance, which is the DHFL epitaph. Re-underwrite the mix before accepting the gearing. Either way, with a −16% first-month ALM gap, liquidity binds first. Capital headroom of 21 to 32% is irrelevant to a funding hole 6+ points beyond tolerance. Capital answers "can it absorb slow losses?" The −16% answers "can it pay next month?", and the second kills faster.

Problem 16. Kaveri Finance under Ind AS: opening gross loans 18,000 (gross Stage 3 of 720), opening total ECL 600 (Stage-3 ECL 430). During the year: fresh slippage into Stage 3 540, upgrades/recoveries out 160, write-offs 300 (fully provided), P&L impairment charge 380, closing gross loans 21,000, closing Stage-3 ECL 470, average AUM 19,500. The book is 50% unsecured. Compute closing Stage 3 (stock and ratio), closing total ECL, Stage-3 coverage, net Stage 3, credit cost, the slippage run-rate, and the lagged Stage-3 ratio, then the two-line verdict.

Solution. Closing Stage 3 = 720 + 540 − 160 − 300 = 800 ⇒ ratio = 800/21,000 = 3.81%. Closing total ECL = 600 + 380 − 300 = 680, so Stage-1&2 ECL = 680 − 470 = 210, ~1.0% on the performing book. Stage-3 coverage = 470/800 = 58.75%. Net Stage 3 = (800 − 470)/(21,000 − 470) = 330/20,530 = 1.61%. Credit cost = 380/19,500 = 1.95%. Slippage = 540/(18,000 − 720) = 3.12% of the opening standard book. Lagged Stage-3 = 800/18,000 = 4.44%, against the headline 3.81%, so the growing denominator flatters the ratio by ~0.6 pt. So coverage near 59% is adequate even for a half-unsecured book. But the inflow pipe (slippage >3%) and the lagged ratio (4.4%) are the truth the headline hides. Judge the roll-forward and look past the closing ratio, the same habit you built with banks, unchanged under Ind AS labels.

Problem 17. Cascade Auto Finance (synthetic) is a US non-bank auto lender ($m). Averages: net receivables 14,000, debt 11,500. P&L: interest income 2,660, interest expense 862, other income 300, opex 900, provision 780 (net charge-offs 700), tax 25%. Balance sheet: total assets 15,000, equity 2,600, allowance for credit losses 1,540 (CECL). Liquidity: cash 500, committed undrawn conduit capacity 3,800, debt maturing within 12 months 1,900. (a) Yield, cost of funds, spread, NIM, risk-adjusted NIM. (b) ROA, leverage, ROE. (c) Allowance as % of receivables vs the annual charge-off rate: what does CECL front-loading say about the "low" book value? (d) The 12-month funding runway if the unsecured market shuts.

Solution. (a) Yield = 2,660/14,000 = 19.0%. CoF = 862/11,500 = 7.5%, so the spread is 11.5%. NIM = (2,660 − 862)/14,000 = 1,798/14,000 = 12.8%, and risk-adjusted NIM = 12.8 − (780/14,000 = 5.6%) = 7.3%. (b) PBT = 1,798 + 300 − 900 − 780 = 418 ⇒ net income = 418 × 0.75 = 313.5. ROA = 313.5/15,000 = 2.09%, leverage = 15,000/2,600 = 5.77×, and ROE = 12.1%. (c) Allowance/receivables = 1,540/14,000 = 11.0%, against an annual charge-off rate of 700/14,000 = 5.0%. The reserve equals ~2.2 years of losses because CECL books lifetime expected loss on day one. A slower regime (incurred-loss/IRAC) wouldn't hold that reserve yet, and it's what optically depresses book value. It's an accurate reserve, and calling it conservative excess would be wrong. Don't read the fat allowance as distress. (d) Liquidity = cash 500 + committed conduit 3,800 = 4,300, which covers the 1,900 of 12-month maturities 2.26×. And the conduits are committed, multi-year, secured on the receivables, so they can't be voted away the way 90-day CP can. The 2018 casualties lacked that structural answer. A non-bank can be fair-weather-proof if its funding is committed and collateralized instead of confidence-sensitive.

Problem 18 (timed: 25 minutes, exam conditions). Tungabhadra Finance, FY summary (₹ cr, Ind AS). Averages: AUM 28,000 (opening 22,000, closing 34,000), borrowings 22,000, total assets 29,500, net worth 4,200. P&L: interest income 4,060, interest expense 1,870, other income 480 (of which upfront DA gains 180), opex 950, impairment 620. Closing: borrowings 27,000, net worth 4,600, Tier 1 4,000, Tier 2 250, RWA 26,500, gross loans 29,000, Stage 3 870, Stage-3 ECL 350. Borrowing mix: banks 42% / NCDs 30% / CP 20% / other 8%. Prior-year CoF 8.0%. ALM: 1-month cumulative gap −12% of outflows, LCR 96%. Unsecured = 38% of AUM, and opening gross loans were 21,000. Market cap 7,000. Deliver the KPI panel with classifications, earnings restated for (i) normalized credit cost 3.0% and (ii) ex-DA gains, then both, a five-line verdict naming the two deadliest issues, and a fair-multiple range.

Solution. Yield = 4,060/28,000 = 14.5%. CoF = 1,870/22,000 = 8.5%, up from 8.0%, which is a 50 bp spread squeeze while the book grew. NIM = 2,190/28,000 = 7.82%. AUM growth = 34,000/22,000 − 1 = +54.5%, red. C/I = 950/2,670 = 35.6%, efficient (fast books usually are, briefly). PPOP = 2,670 − 950 = 1,720. Credit cost = 620/28,000 = 2.21%. But lagged Stage 3 = 870/21,000 = 4.1% against a headline 870/29,000 = 3.0%, so the denominator is reassuring you. Coverage = 350/870 = 40.2%, thin for a 38%-unsecured book. Net Stage 3 = (870 − 350)/(29,000 − 350) = 1.82%. PAT = (1,720 − 620) × 0.75 = 825 ⇒ ROA = 825/29,500 = 2.80% ("strong"), ROE = 825/4,200 = 19.6%. Gearing = 27,000/4,600 = 5.87× (watch). CRAR = (4,000 + 250)/26,500 = 16.04% (Tier 1 15.09%), barely a point above the floor while growing 54%. RWA at this pace outruns 19.6% internal accrual within a year, so a dilutive raise isn't optional. ALM at −12% is breach territory, LCR at 96% < 100% floor is a breach, and CP at 20% is red.

For the restatements: (i) at 3.0% credit cost, impairment is 840 ⇒ PAT = (1,720 − 840) × 0.75 = 660 ⇒ ROA 2.24%. (ii) Ex-DA, PAT = 825 − 180(0.75) = 690 ⇒ ROA 2.34%. With both, PAT = (1,720 − 840 − 180) × 0.75 = 525 ⇒ ROA 1.78%, ROE 12.5%. Roughly a third of the reported 19.6% ROE is manufactured, from pulled-forward income + light coverage + denominator optics.

Verdict, in five lines. (1) Four of five red flags are lit: ALM+CP, growth ≫ quality, thin unsecured coverage, spread squeeze (only promoter pledge is unexamined). (2) The deadliest pair is CP 20% against a −12% short gap and a sub-100% LCR, because the funding can be voted away in 90 days, and CRAR 16% against 54% growth, because the rescue raise dilutes just as the multiple compresses. (3) The real ROE is ~12.5%, well short of 19.6%. (4) ABV = 4,600 − (870 − 350) = 4,080 ⇒ P/B = 7,000/4,600 = 1.52×, P/ABV = 1.72×, while justified P/ABV at r = 14%, g = 5% on a 12.5% ROE ≈ (12.5 − 5)/9 = 0.83×. (5) The fair range is ~0.8 to 1.0× ABV until CP terms out, coverage rebuilds past 50%, LCR clears 100%, and growth decelerates. Ask one question before any position: can it raise 3 to 5 year money this quarter at a spread that leaves the model alive? (Score: 1 per KPI computed and classified, 10 max; 3 for the two restatements; 3 for a verdict naming the CP/ALM + capital pair. ≥12/16 to pass.)

Applied mini-project: the NBFC workup (the Phase 5 gate)

Task. Run the full playbook on one Indian NBFC/HFC and one US consumer-finance company. Deliver a two-page workup for each, plus a one-paragraph cross-market comparison. For India, choose from Bajaj Finance, Cholamandalam Investment & Finance, Shriram Finance, M&M Financial, Sundaram Finance, Muthoot Finance, Manappuram, LIC Housing Finance, PNB Housing, Can Fin Homes, Aavas Financiers, Five-Star Business Finance, or CreditAccess Grameen. Pick at least one name that wasn't used as an example here. For the US, take OneMain Holdings, the purest surviving non-bank analogue. If you want a contrast, add one deposit-funded convert (Ally, Synchrony, or American Express). Working out why the US names are mostly banks now is part of the deliverable.

Data to pull (all free). For India, you need these:

• The latest annual report. You want the borrowings note, the maturity-pattern/ALM note, the stage-wise loan and ECL disclosures, and the securitization/assignment note.
• The last four quarterly investor presentations (AUM mix, disbursements, stage-wise movement, borrowing mix, LCR).
• The sector's cheat code, the credit rating rationales from the CRISIL/ICRA/CARE/India Ratings websites. They tabulate borrowing mix, ALM gaps, banker relationships, and unencumbered assets more candidly than any deck.
• screener.in for the 5-year ratio history.
• The RBI SBR Master Direction for the thresholds you cite.

For the US, pull the 10-K (the liquidity and borrowings sections of MD&A, the receivables/allowance notes under CECL) and the quarterly earnings supplement from SEC EDGAR.

Deliverables. Hand in all eight of these:

1. The 10-KPI panel for 5 years, with each cell classified healthy/watch/stress.
2. The rebuilt ALM table at zero CP rollover, with the survival horizon in months (Worked example 3's formula) and mitigants counted at committed-only values.
3. A funding-franchise paragraph: borrowing mix over 5 years, CoF against two peers, and what happened to its spreads in the last stress episode (2018-19 or 2020). The archives will tell you.
4. Asset quality: stage-wise stock and coverage, the roll-forward rebuilt for 2 years, write-off velocity, vintage/cohort disclosures if given, credit cost restated through-cycle.
5. Capital & regulation: CRAR/Tier 1, the growth headroom math, the SBR layer and what it obliges.
6. Valuation: a P/B-vs-ROE scatter against ≥8 lenders (mix banks and NBFCs deliberately), a two-stage excess-return value, and the reverse read. What ROE-duration does the current multiple imply?
7. The five-flag sweep, with evidence citations for each flag, including the pledge check.
8. The verdict: an adapted 8-line teardown ending with the one question you'd answer next. For the US name, add two sentences on how CECL and deposit/non-deposit funding changed what you watched.

Scoring rubric (0 = missing/wrong · 1 = present but shallow or partly wrong · 2 = complete, correct, evidence-cited). Pass = ≥12/16 with no criterion at 0. Passing this plus a mastery-check form at ≥85% unlocks the insurance playbook.

File the finished panels in your knowledge-base thread. This workup becomes the second page of the sector KPI cheat-sheet artifact, and Phase 8's teardowns reuse it whenever the triage says "non-bank lender."

Case Lab: run case CL-5.2 (The Rollover That Never Came) in the app's Cases tab: a synthetic HFC walked from a compliant-looking published ALM table through a CP-market freeze: you re-derive the ladder at zero rollover, count only committed mitigants, compute the survival horizon in months, and classify survive/stress/dead as the disbursement halt manufactures its own NPAs.

Reading & resources

• RBI, Master Direction — Non-Banking Financial Company — Scale Based Regulation Directions, 2023 (as amended), at rbi.org.in. The primary law: layers, CRAR, NPA norms, concentration rules. Skim it for structure and cite it for thresholds. It's where every threshold above is checked. [Free] [Advanced, reference]
• RBI, Liquidity Risk Management Framework for NBFCs (November 2019 circular): the ALM buckets, the 10/10/20 caps, and the LCR regime, born directly of the IL&FS autopsy. Short and readable. [Free] [Intermediate]
• RBI, Report on Trend and Progress of Banking in India (annual, NBFC chapter) and the Financial Stability Report (semi-annual, NBFC stress tests and bank–NBFC interconnectedness): the free system-level layer for every workup. [Free] [Intermediate]
• Tamal Bandyopadhyay, Pandemonium: The Great Indian Banking Tragedy. The IL&FS and DHFL chapters narrate the failure anatomy above with reporting you can't get elsewhere. [Paid] [Beginner–Intermediate]
• Credit rating rationales from CRISIL, ICRA, CARE, India Ratings (free on each agency's site, searchable by company). For any NBFC: borrowing mix, ALM tables, banker lists, unencumbered assets, gearing covenants. It's the densest free document in this sector. Read three for your project company across three years and watch the language shift. [Free] [Intermediate]
• HDFC Ltd annual reports, 2015–2022 (pre-merger; archived on BSE/NSE and web archives). Deepak Parekh's letters plus the sector's cleanest ALM and funding disclosure: the gold standard everyone else is graded against. [Free] [Beginner–Intermediate]
• Bajaj Finance annual report + investor presentations. The franchise archetype documenting its own liability strategy. Read the liquidity/ALM slides from FY19 (the crisis) and from now. [Free] [Intermediate]
• Muthoot Finance and Manappuram presentations; MFIN Micrometer (quarterly microfinance industry data, mfinindia.org): segment calibration for gold and MFI. [Free] [Intermediate]
• Aswath Damodaran, "Valuing Financial Service Firms" (stern.nyu.edu, free paper): the equity-direct valuation theory. Reread it after the valuation section. [Free] [Intermediate]
• Gary Gorton, Slapped in the Face by the Invisible Hand: Banking and the Panic of 2007 (free SSRN paper): the deepest short explanation of why wholesale/securitized funding runs. It's the American rhyme above, in academic form. [Free] [Advanced]
• UK House of Commons Treasury Committee, The Run on the Rock (2008), at parliament.uk, free. Northern Rock dissected: an HFC failure anatomy with British candor. Read chapters 1–3. [Free] [Intermediate]
• Gryta & Mann, Lights Out: Pride, Delusion, and the Fall of General Electric. The GE Capital chapters: how CP funding made and unmade the bluest chip. [Paid] [Beginner–Intermediate]
• OneMain Holdings 10-K (SEC EDGAR). Read the "Liquidity and Capital Resources" MD&A and the funding-structure exhibits: a live non-bank lender documenting its own survival math (runway targets, ABS ladders). Pairs with any Indian NBFC's rating rationale as a cross-market drill. [Free] [Intermediate]
• screener.in / Trendlyne (India) and stockanalysis.com (US): the 10-year KPI histories for panels and scatters. [Free] [Beginner]
• Going deeper: RBI, Discussion Paper / Revised Regulatory Framework for NBFCs: A Scale-Based Approach (2021) (rbi.org.in): the reasoning document behind SBR and the liquidity framework. Read it beside the Master Direction to see why each post-2018 rule exists: the case for granular ALM buckets, LCR, and the upper-layer regime argued straight from the IL&FS/DHFL autopsies. [Free] [Advanced]
• Going deeper: Vinod Kothari, Securitisation, Asset Reconstruction & Enforcement of Security Interests (and the free explainers at vinodkothari.com): the definitive Indian treatment of direct assignment, securitization, minimum retention, credit enhancement, and Ind AS 109 derecognition. That's the machinery under the upfront-DA-gain flag and the on-book-vs-AUM bridge, and it'll let you read a securitization note the way its author wrote it. [Paid] [Advanced]

The Modern Analyst's Addendum

Everything above teaches this skill from first principles, by hand, and that is how you learn it, and the mastery check still tests it that way. This addendum shows how a working analyst amplifies the same skill today. It adds; it never replaces. (R1/R10)

AI-Augment this skill

{
  "skill": "The NBFC/HFC playbook: the structural liquidity statement rebuilt on a freeze convention, the ten-KPI panel, Ind AS 109 stage migration and the impairment reserve, the AUM bridge across on-book and assigned pools, cost-of-funds and borrowing-mix discipline, the SBR layer map, the five segment lenses, and P/B against ROE with the franchise premium interrogated",
  "use": "The liability side of an NBFC is documented in prose scattered across four sources that no aggregator consolidates — the borrowings note, the maturity-pattern note, the rating rationale, and the securitisation/assignment disclosures — and assembling it is where the analytical hours go. That is the job to hand over. (1) LIABILITY RECONSTRUCTION: give it the annual report, the last four rating rationales from CRISIL/ICRA/CARE/India Ratings, and ask it to build one instrument-level table — instrument, amount, coupon, fixed or floating, contractual maturity date, covenant, secured or unsecured — with a page cite per row and an explicit NOT DISCLOSED wherever the document is silent. (2) The RATING-RATIONALE diff: rating agencies restate the same lender every six months and change one sentence; ask it to diff consecutive rationales and quote what moved, because a downgraded outlook is usually visible as a changed clause before it is visible as a changed rating. (3) STAGE-MIGRATION and overlay archaeology: extract stage-wise gross carrying amount, ECL and coverage per quarter and flag every change in the stage-2 migration criterion or 'management overlay' release. (4) The ASSIGNMENT question: find every direct-assignment and securitisation transaction, the pool size, the minimum retention and any credit enhancement, so the upfront-gain restatement in §4.3 has real inputs. What it must never do is tell you whether the ALM is safe. The gaps are arithmetic and the mitigants are a judgement about whose promises are committed — both are yours.",
  "tools": ["Document/PDF AI — for instrument-level liability reconstruction from annual reports and rating rationales, and for stage-wise ECL extraction", "Chat assistants — Claude, ChatGPT for diffing consecutive rating rationales and definition footnotes", "Coding assistants — for the bucketed ALM ladder, the repricing-gap engine and the rollover-survival arithmetic below", "Spreadsheet/notebook tooling — pandas for the bucket table, so the freeze convention is a parameter rather than a re-typed table"],
  "prompt": "From the attached annual report and rating rationales for <LENDER>, build one table of every borrowing instrument outstanding: type (CP / NCD / bank term loan / ECB / sub-debt / refinance / securitisation payable), amount, coupon, fixed or floating and the benchmark, contractual maturity date, secured or unsecured, and any financial covenant. Cite the document and page for each row. Separately list every UNDRAWN facility with the exact word the document uses — 'committed' or 'sanctioned' — quoted verbatim, because the difference decides survival. Write NOT DISCLOSED wherever the documents are silent; never interpolate. Do not build a maturity ladder, do not compute any ratio, and do not comment on liquidity.",
  "verify": "Three checks, all arithmetic. First, the LADDER MUST FOOT: total bucketed inflows minus outflows should equal net worth minus unbucketed fixed assets. Chambal's does — 30,900 - 27,100 = +3,800 = 4,100 - 300 — and a published statement that does not foot has dropped an outflow, usually undrawn disbursement commitments the lender has already promised borrowers. Second, the AUM BRIDGE must tie: opening + disbursements - collections and runoff - write-offs = closing (Chenab: 20,000 + 13,500 - 7,100 - 400 = 26,000). Third, re-derive the ladder on the FREEZE CONVENTION yourself — commercial-paper rollover set to zero, behavioural prepayment generosity stripped, only committed lines counted — because the published statement is built on management's rollover and behavioural assumptions and is therefore a statement about a world in which nothing has happened. Every figure a tool extracts is a candidate; only the identities are evidence.",
  "diy": "The gate is unaided. You build the structural liquidity statement bucket by bucket from a maturity-pattern note, foot it to net worth, compute cumulative gaps as a percentage of cumulative outflows against the 10/10/20 caps, run the freeze by hand and state the survival horizon in months, compute all ten KPIs, restate ROA for upfront assignment gains, and price the ALM fix in basis points of ROA and points of ROE. On paper. The reason is the one §4.5 states: capital ratios never predicted an NBFC failure and liquidity always did, so the arithmetic you must own is the arithmetic no ratio service publishes."
}

Modern Data Analysis

By hand first. You built Tapti's and Chambal's ladders bucket by bucket, footed both to net worth, ran the freeze month by month, and priced the fix at 12 bp of ROA. Keep all of it. The ladder's arithmetic is easy, but it has to be careful, and the care is a habit: refusing management's rollover assumption, refusing behavioral prepayment, counting only committed lines. That habit lives in your hands, and no library can hold it for you.

What code adds is a second reading of the same table. The structural liquidity statement answers one question: can this lender pay what it owes when it owes it? Read the identical bucket table with the word "maturity" replaced by the word "repricing", and it answers a different one: what happens to the margin when the price of money changes? A liability that matures has to be refinanced at tomorrow's rate. An asset whose principal comes back gets redeployed at tomorrow's rate. So the same cells that give the liquidity gap also give the repricing gap, and the repricing gap times a funding-cost shock is a NIM compression you can compute. KPI 3 names the mechanism: "the market repricing the NBFC's own credit faster than the NBFC can reprice its borrowers". It's the arithmetic behind every Indian NBFC crisis. The funding shock arrives long before the credit shock, and it hits the liability side of a book that reprices at one-third the speed of its funding.

Today's workflow. One frame, two readings, three parameters. The frame is the bucket table. The readings are liquidity_gap() and repricing_gap(). The parameters are the ones the published statement hides: cp_rollover (set it to zero), behavioural_prepayment (set it off), and lines (committed only, with the document's own word quoted in the code comment). Then the shock becomes a function argument instead of a scenario you re-type. You can ask the two questions that matter. How much of a market rate move survives, when both sides reprice? And how much of an idiosyncratic spread widening survives, when only your side reprices? The second is the one that kills lenders, and it's roughly two and a half times the first.

Tools & sources (IN + US). Use pandas for the bucket frame and numpy for the shock grid. Plain arithmetic handles the survival horizon. You don't need a library, and none should be trusted with the freeze convention. For India:

• The annual report's maturity-pattern of assets and liabilities note, the statutory version of the ALM table, and the borrowings schedule.
• The rating rationales, free on crisil.com, icra.in, careratings.com and indiaratings.co.in. They're the densest public document about any Indian NBFC's liability side, and the only one that names undrawn committed lines.
• The RBI Master Direction — Non-Banking Financial Company Scale Based Regulation Directions, 2023, for the layer map, the LCR phase-in and the 10/10/20 bucket caps.
• RBI's liquidity risk management framework for NBFCs (November 2019), for the statement's prescribed format.
• The National Housing Bank's Trend and Progress of Housing in India, for HFC-specific system data.
• RBI's DBIE, for system NBFC credit and the sectoral deployment series.
• The company's securitisation and direct-assignment disclosures, for the off-book half of AUM.

For the US:

• SEC EDGAR XBRL companyfacts and the 10-K liquidity section, for a non-bank lender's committed conduit capacity and unsecured maturity wall (the structural fix worked example 5 describes).
• Series from FRED (DGS2, DGS10, FEDFUNDS, and the ICE BofA option-adjusted-spread series such as BAMLC0A4CBBB), for the funding-cost shock a US non-bank actually faced.
• The SEC's ABS-EE / Reg AB II loan-level exhibits, for securitised consumer pools.

# One bucket table, two readings. Recomputed in-session — R3. Chambal HFC, §8 WE4.
import numpy as np
B = ["<=1m", "1-3m", "3-6m", "6-12m", "1-3y", "3-5y", ">5y"]
assets = {"cash":    [900, 0, 0, 0, 0, 0, 0],            # liquidity only, never reprices
          "home":    [330, 660, 990, 1980, 6300, 4300, 7440],
          "lap_dev": [0, 80, 400, 1000, 3200, 2320, 1000]}
liabs  = {"cp":      [1400, 2000, 600, 0, 0, 0, 0],      # rollover set to ZERO — the freeze convention
          "ncd":     [200, 600, 1000, 2000, 5500, 3700, 2000],
          "bank":    [100, 150, 250, 500, 2000, 2000, 1000],
          "nhb":     [0, 0, 0, 150, 550, 450, 250],
          "opex":    [60, 120, 180, 340, 0, 0, 0]}       # an outflow, but never reprices

A = np.array(list(assets.values())).sum(0); L = np.array(list(liabs.values())).sum(0)
liq = (A - L).cumsum()                                    # reading 1: can we pay?
rep_a = (np.array(assets["home"]) + np.array(assets["lap_dev"])).cumsum()
rep_l = (L - np.array(liabs["opex"])).cumsum()            # reading 2: what reprices?
h = 3                                                     # index of the 6-12m bucket
print("12-month liquidity gap ", f"{liq[h]:>8,.0f}")               # -3,310
print("12-month repricing gap ", f"{rep_a[h]-rep_l[h]:>8,.0f}")    # -3,510

def nim_hit(bps, idiosyncratic=True, book=30000, nw=4100, tax=0.25):
    d = bps / 10000
    dNII = (0 if idiosyncratic else d * rep_a[h]) - d * rep_l[h]
    return dNII, dNII / book * 10000, dNII * (1 - tax) / nw * 100
for kind in (True, False):
    n, nim, roe = nim_hit(200, kind)
    print(f"+200bp {'own-spread' if kind else 'market-wide'}: "
          f"NII {n:+8.1f} cr | NIM {nim:+6.1f} bp | ROE {roe:+5.2f} pp")

Verify. Prove the engine on the ladders you built by hand before it meets a real lender. Chambal should reproduce cell for cell:

• Inflows of ₹30,900 cr and outflows of ₹27,100 cr.
• Bucket gaps of −530 / −2,130 / −640 / −10 / +1,450 / +470 / +5,190.
• Cumulative gaps of −530 / −2,660 / −3,300 / −3,310 / −1,860 / −1,390 / +3,800.
• Cumulative gap over cumulative outflows of −30.1% / −57.5% / −49.5% / −34.3% / −10.5% / −5.8% / +14.0%.
• Footing to +3,800 = 4,100 − 300.
• A freeze running position, with ₹700 of committed lines, of +170 / −1,960 / −2,600 / −2,610.

Tapti's table foots the same way (41,000 − 36,840 = +4,160 = 4,500 − 340). The fix prices at ₹50 cr pre-tax, ₹37.5 cr post-tax, 12.0 bp of ROA on ₹31,200 cr of assets, 0.91 ROE points, which matches the "12 bp ≈ 0.9 ROE points" above. Chenab's AUM bridge ties (20,000 + 13,500 − 7,100 − 400 = 26,000), and its CP rollover arithmetic reproduces too. ₹2,300 cr at a 75-day average tenor is 4.87 rollover cycles a year and ₹11,193 cr of rollover volume, which matches the "~₹11,000 crore of annual opportunities for someone to say no."

Quantitative lens

The ladder holds most of this sector's quantitative content. It supports three results that the text above states in words and that you can't compute on paper. All figures are recomputed from the Chambal table above. They're properties of that table, and they don't measure any listed lender.

Chambal's book reprices at one third the speed of its funding, and that gap is the crisis mechanism. Read the same table twice. On the liquidity reading, the twelve-month cumulative gap is −₹3,310 cr. The repricing reading sets assets whose principal comes back and can be redeployed at new rates against liabilities that must be refinanced at new rates. On that reading the twelve-month gap is −₹3,510 cr, because by twelve months only 18.1% of the ₹30,000 cr loan book has returned cash, while 33.9% of the ₹26,400 cr of borrowings has come due. The mismatch persists. At three years it's 49.8% of assets against 64.4% of liabilities, and at five years 71.9% against 87.7%.

Now shock it. Take a +200 bp market-wide move first, where the lender reprices its borrowers as its funding reprices. It costs ₹70.2 cr of NII, which is 23.4 bp of NIM and 1.28 points of ROE. That's painful, but survivable and symmetric. Then take a +200 bp idiosyncratic move. Here the lender's own credit spread widens because someone else in the sector defaulted, and it can't reprice existing fixed-rate borrowers. That costs ₹179.0 cr, 59.7 bp of NIM and 3.27 points of ROE. The idiosyncratic shock is 2.55× worse than a market shock of identical size, and it's the one that actually happens. IL&FS didn't raise anyone's risk-free rate.

Push it through to the multiple. A lender starting at 16% ROE with a 14% cost of equity and 6% sustainable growth is worth 1.25× book on the one-stage identity. After the same 200 bp own-spread shock its ROE is 12.73% and justified P/B is 0.84×. That's a 33% fall in justified value from a funding event, before a single borrower has missed a payment. So the KPI panel lists the ALM row last but ranks it first in the decision order.

The commercial-paper rollover count converts "refinancing risk" into a survival probability. Chambal carries ₹4,000 cr of CP. At a 75-day average tenor that's 4.87 rollover cycles a year and ₹19,467 cr of annual rollover volume, nearly five times the outstanding stock, and every rupee of it is an occasion for an investor to decline. Model each cycle as an independent refusal with probability p, and the annual survival probability is (1 − p) raised to 4.87. At p = 1% that's 95.2% for one year and 86.4% across three. At p = 2% it's 90.6% and 74.5%, and at p = 5% it's 77.9% and 47.3%. Read those numbers the way the ALM section asks you to. Forget the idea of a cheap funding line, and see the CP stock as a string of 4.87 annual survival tests whose failure has no covenant-waiver dance. You pay at par on the day or you're in default. That's how the phrase "how much of the company's continued existence has been outsourced to other people's future decisions" becomes a quantity you can measure. Of course, independence across cycles is the wrong assumption, and it's wrong in the flattering direction. Refusals cluster, so treat these figures as ceilings on survival. They aren't estimates of it.

And the price of survival is about one point of ROE, computed two ways on two lenders. Chambal's fix terms ₹4,000 cr of CP into five-year NCDs at roughly +50 bp and holds ₹1,500 cr more liquidity at roughly −200 bp of carry. It costs ₹50 cr pre-tax, ₹37.5 cr after tax, 12.0 bp of ROA, 0.91 points of ROE. Tapti's identical fix on a larger book (₹5,400 cr of CP, the same ₹1,500 cr buffer) costs ₹57 cr pre-tax, ₹42.75 cr after tax, 10.3 bp of post-tax ROA on ₹41,340 cr of assets, 0.95 points of ROE. Two lenders, two balance sheets, the same answer to two decimal places: survival costs about one point of ROE. Now set that beside the 3.27 points of ROE that a 200 bp own-spread shock takes from an unfixed Chambal. The trade isn't close. The fix costs 0.91 points to avoid 3.27, a payoff of 3.6 to 1 against a single shock of a size the sector has delivered repeatedly. So why did lenders decline it anyway? The ALM section gives the reason. The point of ROE shows up every quarter, the shock shows up once a decade, and the market paid premium multiples for the one it could see.

These results have limits, and they matter. The repricing reading treats every maturing liability as refinancing at the shocked rate and every returning asset rupee as redeploying at it. For a fixed-rate book that's clean. It's wrong for the growing floating-rate share of Indian lending, where an externally-benchmarked loan reprices on a reset date that has nothing to do with its maturity bucket. If a lender discloses a floating-rate share, split the asset side before you run this. If it doesn't, treat the gap computed here as an upper bound on the compression. The shock is also applied to a static balance sheet, with no new disbursement, no repricing of new business and no management action. In a real freeze the halt in disbursement changes the ladder itself, as stalled developer borrowers turn "scheduled" receipts into zero. The rollover-survival numbers assume independent refusals, which is the opposite of how a freeze works. They illustrate the shape of rollover risk, and you can't read them as probabilities of failure. The bucket table is contractual and monthly-aggregated, and that flatters it. The regulation's 1 to 7 day sub-bucket is where commercial paper clusters, as the ladder above says. And nothing here touches credit. A lender can pass every one of these tests and still be insolvent, and Bhima is that lender, because KPI 10 answers the survival question and says nothing about whether the assets are worth what the balance sheet says.

Do it in code: hold one bucket frame and two functions over it. Make cp_rollover=0 and committed_lines_only=True defaults instead of scenarios. Run the shock as a grid and report the idiosyncratic case beside the market case, because quoting only the symmetric one understates the risk by roughly two and a half times. And price the ALM fix in basis points of ROA every time you praise a lender's ROE. See G1/FI1.02, G2/QD2.02 and G2/CS2.06.

Where this goes next: galaxy cross-links

• G1 · FI1.02 Duration & Convexity: the repricing gap above is a crude duration measure; that module gives you the exact one, including why a maturity bucket and a repricing bucket differ for a floating-rate asset and how to compute the sensitivity properly rather than by cumulating principal.
• G1 · FI1.03 Term Structure & Bootstrapping: the funding curve an NBFC actually borrows on, so a "+50 bp to term out the CP" assumption becomes a number read off a curve rather than a rule of thumb, and so the negative carry on a liquidity buffer is priced rather than guessed.
• G2 · QD2.02 Risk Systems: VaR & Expected Shortfall. Liquidity stress testing done as a system rather than a table: scenario libraries, the difference between a shock and a stress, and why a survival horizon is a better output for this sector than a loss distribution.
• G2 · CS2.06 Software Engineering & AI-Assisted Development: the freeze convention as a default argument rather than a scenario, the footing identity as a test, and a ladder that raises rather than silently rebalances when the buckets do not tie to net worth.
• G1 · MS1.02 Payment System Plumbing: where commercial paper, repo and the interbank market actually clear, which is the machinery whose seizure in 2018 turned a single default into a sector-wide funding freeze.

Teach it back & journal

Feynman prompt. Write one page for a smart 15-year-old titled "How a lender with no depositors can die in ninety days." It has to cover four things:

• A pocket-money example of borrowing short to lend long. You lend your friend ₹500 repayable next month, funded by ₹450 borrowed from three classmates repayable next week.
• What "rolling over" means, and why nobody has to withdraw anything for the lender to die. The classmates just stop saying yes.
• The difference between not having money (insolvency) and not having your money ready (illiquidity).
• Why the lender with the boring classmates, who always re-lend, can safely earn less than the one paying top rates to strangers.

Use no jargon except three terms you define yourself: ALM, rollover, capital.

Journal reflection. Three prompts, ten minutes. (1) Which segment's economics tempted you most: the gold lender's serene 5% ROA, the franchise's 20% ROE, or the microlender's fat normal-year margins? Write down which variance you were about to under-price, and the pre-commitment rule that stops you (e.g., "any lender growing >35% gets valued on restated earnings and a zero-rollover ALM read, no exceptions"). (2) Recall a time you extended "rollover" trust yourself, renewing a commitment by default instead of by decision. What would have made you stop rolling, and how much warning would the other side have had? That instinct is what a CP investor feels in a bad September. (3) Add one line in liability language to your permanent checklist thread. Most checklists are all asset questions, so fix yours.

This module's flashcards and mastery quiz are wired into the app: see the node's Quiz and Reviews.