Banks

Phase 5 · Week 42 · ~20 hours

Where it sits. Revenue growth can be the single most bearish fact on a bank's page. Apply consumer-company tools to a lender and every number misleads you: EBITDA is meaningless, FCFF is undefinable. That is why banking opens Phase 5's ten sector playbooks, because it is the sector where the phase's master rule (the business model dictates the KPIs and the valuation multiple) is most violently true. Phase 4 taught you to judge whether a business is good; the special-situations capstone gave you the valuation architecture for lenders (equity-direct valuation, the excess-return model, P/B read against ROE). Both are industrialized here into a complete working playbook: the banking P&L and balance sheet from first principles, the full KPI panel with healthy/stress thresholds, the credit cycle and why bank earnings lag the truth, the RBI and US regulatory maps, valuation run end-to-end, and the red-flag panel. Two autopsies, Yes Bank and Silicon Valley Bank, show the two classic ways banks die. It also sets up the NBFC playbook that follows, which is this playbook minus the deposit franchise and plus a liability time bomb. On the five-questions spine (from the master map), a bank playbook is questions 2 and 4 fused: in banking, "is it a good business?" and "what can kill it?" are the same question asked in different tenses.

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

App link. The Banks wing of the sector case engine runs on this material: the timed KPI-panel drill (a bank's summary financials appear; you compute the twelve KPIs and classify each healthy/watch/stress) and the red-flag spotter (quarterly-result vignettes hiding one deteriorating bank each). It also runs the excess-return builder carried over from the special-situations capstone (now with bank-specific fade patterns), and the workup grader that scores your mini-project against the rubric below. That score is the Phase 5 gate for the NBFC playbook. Your completed KPI table seeds the first page of the sector KPI cheat-sheet artifact.

Learning objectives

By the end of the week you can:

1. Explain from first principles why banks are different: why leverage is the product rather than a financing choice, why deposits are raw material rather than debt, why confidence is a balance-sheet input, and therefore why FCFF/WACC/EV multiples fail and equity-direct tools (P/B vs ROE, excess returns) take over.
2. Rebuild a bank's P&L and balance sheet from a blank page: interest income → NII → PPOP → PAT on one side; loan book, investment book (SLR/CRR), deposits (CASA vs term), borrowings, and capital on the other. Then trace how one rupee of deposits becomes profit.
3. Compute and interpret the full KPI panel from a bank's published financials: NIM, GNPA/NNPA, PCR, credit cost, slippage ratio, ROA, ROE, cost-to-income, CASA, CD ratio, CRAR/CET1, each against its healthy/stress threshold, with the NPA roll-forward that links them.
4. Explain why bank earnings lag the truth (growth-denominator optics, vintage seasoning, recognition discretion, provisioning smoothing), and use SMA buckets, restructured books, and RBI divergence disclosures as your early-warning instruments.
5. Navigate the regulatory map: RBI's IRAC norms, Basel III minima as applied in India (with buffers), PCA triggers, and priority-sector rules; and translate the panel into US terms (Fed/OCC/FDIC, CCAR/stress-capital buffers, efficiency ratio, net charge-offs).
6. Value a bank two ways: a P/B-vs-ROE cross-sectional read (including the regression logic and adjusted book value), and a multi-stage excess-return model run end-to-end. Then reverse either into "what the price implies."
7. Run the red-flag panel on any bank in fifteen minutes: rising SMA-2 + slippages, NIM compression alongside aggressive growth, falling PCR as GNPA rises, CD ratio above 90%, promoter/concentrated exposure, NPA divergence. Then explain, using Yes Bank and SVB, how each flag maps to a real failure mechanism.
8. Deliver the Phase 5 bank workup: a full KPI panel, five-year trend read, valuation, and one-page verdict on one Indian and one US bank, graded against the mini-project rubric.

Prerequisites & connections

Builds on. M3.10's section on valuing financials is the direct parent. It covers why lenders break the standard valuation machine, the dividend-discount variant with the regulatory-capital payout constraint (payout = 1 − g/ROE), the excess-return model and the P/B-vs-ROE shortcut. Those tools are assumed here. The time goes instead on the operating anatomy and the sector judgement, which M3.10 put off. From Phase 1 you need M1.09's financial-instruments section, because amortised cost versus FVOCI/FVTPL matters enormously and SVB is as much an accounting-classification story as a risk story. You also need M1.10, the Indian annual-report anatomy, whose bank-specific schedules you'll now read. From Phase 2, bring the quality-of-earnings reflexes (M2.06–M2.08), since a bank's stated book value is a hypothesis you still have to test. Bring the DuPont habit (M2.02) too, which comes back here as the ROA tree. From Phase 3, cost of equity (M3.03) is the r every bank tool discounts at. From Phase 4, bring moat language. A CASA franchise is a switching-cost-plus-brand moat you can measure in basis points of funding advantage.

Feeds into. M5.02 (NBFCs/HFCs) strips away deposit insurance and the RBI liquidity backstop, and shows what the same balance sheet looks like when the liability side can vanish in 90 days. Run this playbook first or that one won't land. M5.03 (insurance) reuses the "regulated financial with a levered investment book" pattern. Phase 7's macro work on rate cycles and liquidity plugs straight into NIM and credit-cost forecasting. Phase 8's rapid teardown uses this playbook as its "financials" branch. When the two-minute triage says lender, this is the checklist you run. Competency C6 (analyse any industry with the right lens) begins its certification here.

Core teaching content

4.5 The credit cycle: why bank earnings lag the truth

A bank's reported profit is an estimate stapled to a hope. The revenue is real enough. But the cost that matters, credit losses, gets recognised years after the lending decision that caused it. Four mechanisms create the lag. Learn them as a set, because they always operate together.

1. The denominator effect: growth hides sins. GNPA is a ratio, and fast growth inflates the denominator faster than sins surface in the numerator. Watch it happen. A bank with advances of ₹1,00,000 cr and GNPA of ₹2,000 cr (2.0%) grows 40% in a year. Fresh slippage of ₹1,000 cr arrives and ₹480 cr is recovered or written off, so closing GNPA is ₹2,520 cr on advances of ₹1,40,000 cr, or 1.8%. The headline says "asset quality improved." In truth the NPA stock rose 26%. Two years later growth stalls and the new ₹40,000 cr vintage seasons at a 4% slippage rate, which is ₹1,600 cr. GNPA jumps toward 3.3% with no new lending sin. The sin was committed during the applause. Hence the rule. For any bank growing >25%, recompute asset quality on the lagged book, GNPA today ÷ advances 2 years ago, before you believe the ratio.

2. Vintage seasoning. Loans default on a schedule. Retail unsecured losses peak 12 to 24 months after disbursal. Corporate and project loans take 3 to 5 years. A book that doubled recently is mostly unseasoned. You can't know its true loss rate yet, and its current credit cost is structurally understated.

3. Recognition discretion. The 90-day rule sounds mechanical. In practice it's negotiable, through fresh loans that pay old interest ("evergreening"), restructuring that resets the clock, and moratoria. India's pre-2015 corporate cycle was the masterclass. System restructured books swelled past 6% of advances while reported GNPA looked tame. In September 2015 the RBI's Asset Quality Review (AQR) re-inspected the big exposures and forced recognition, and system GNPA roughly tripled to a peak near 11.2% in FY18 (PSU banks ~14.6%). Those loans hadn't gone bad in 2016 to 18. They'd gone bad in 2011 to 14 and were finally named. Looking back, the earnings of FY12 to FY15 had been substantially fictional.

4. Provisioning smoothing. Even once a loss is recognised, the P&L hit is an estimate. India's incurred-loss regime sets floors, which the regulatory map covers, but it leaves headroom above them. Managements under-provide in bad years to protect reported ROE, and falling PCR is the tell. In fat years they over-provide. The US moved to CECL, booking lifetime expected losses on day one, after 2008 taught it the same lesson. RBI has proposed a similar expected-credit-loss framework for banks, so check where implementation has got to.

Put the four together and you get the banking cycle in one arc. It starts with a credit boom. India's infra and corporate boom of FY04 to FY11 ran at ~20%+ credit CAGR. GNPA ratios hit cyclical lows just when embedded risk peaks (FY08: ~2.3%). Then growth slows and vintages season, until discretion runs out. A recognition avalanche follows. In India that was the AQR, with GNPA at 11.2% in FY18. Then comes the clean-up: IBC 2016, recapitalisation, write-offs. Ratios look pristine again (~2.3 to 2.8% in the mid-2020s; check in the current FSR). Memories fade and unsecured retail speeds up. Round it goes.

That's the lesson for question 4 of the spine. A bank's P&L this year describes its lending 2 to 4 years ago. Its lending this year describes its P&L 2 to 4 years from now. So the KPI panel leans hard on leading indicators (slippage, SMA-2, restructured stock, growth mix). And "record profits at a fast-growing lender" should send you to the underwriting data, and nowhere near the buy button.

4.7 Valuation: P/B against ROE, and the excess-return engine

A bank whose ROE equals its cost of equity is worth book at any growth rate. That's the anchor identity you met in relative valuation. For a stable financial, P/B = (ROE − g) / (r − g), so ROE = r ⇒ P/B = 1. Everything from the special-situations work on financials now gets put to industrial use. Every bank valuation conversation comes down to three variables. There's the spread (ROE − r) and the growth it can fund, where g = ROE × retention, capped by capital rules. And there's the duration, how long the spread survives competition and credit cycles.

Step 1. The cross-sectional read, your two-minute tool. Plot P/B (y) against ROE (x) for the 10 to 15 listed Indian banks. screener.in gives you both in one export. Fit the line by eye or by least squares. The line is descriptive. It shows how the market prices a point of ROE today. Banks far below the line are cheap relative to today's pricing rule. That makes them a research trigger. It doesn't make them a verdict, because the regression is blind to the one thing that matters, whether the book and the ROE are real. So before you trust any outlier, re-run it on adjusted book value: ABV = net worth − NNPA. That's the conservative full knock. Some analysts deduct only (1 − tax) of NNPA. State your convention and keep it constant. A bank at P/B 0.9 with NNPA at 4% of book may be at P/ABV 1.3. That discount was the market pricing in thin provisioning. There was no opportunity in it. Practice 10 and 11 drill both moves.

Step 2. The excess-return model, end to end. Start with the frame Value of equity = Book value today + Σ PV of excess returns, where excess return in year t = (ROEₜ − r) × Bookₜ₋₁. Run it on a franchise shaped like ICICI Bank. All figures are illustrative, modelled on ~FY25 scale, so pull the actual annual report and current price before you use this. Standalone book value per share ≈ ₹400. Sustainable ROE ≈ 16%. Cost of equity r = 13%, built the cost-of-capital way from ~7% Rf + ~1.0 β × ~6% ERP. Refresh all three inputs from live sources. Stage-1 growth g = 12%, which forces payout = 1 − 12/16 = 25%. That's the regulatory-capital constraint at work, because book has to grow with RWA. Stage 1 lasts 10 years. After that ROE fades to 14% and g to 5%, giving a terminal P/B = (14 − 5)/(13 − 5) = 1.125.

PV of stage-1 excess returns = ₹102.07. Closing book B₁₀ = 1,109.23 × 1.12 = ₹1,242.34. Terminal excess value at t=10 = (1.125 − 1) × 1,242.34 = ₹155.29, and its PV = 155.29 ÷ 3.3946 = ₹45.75.

Add it up and V₀ = 400 + 102.07 + 45.75 ≈ ₹548 per share ≈ 1.37× book.

Step 3. Reverse it. What is the price implying? Suppose the stock trades at ~₹1,425. The listed and unlisted subsidiaries (insurance, AMC, securities) are worth ~₹200 per share on a sum-of-the-parts basis, using the SOTP mechanics from the special-situations capstone. Then the core bank is being priced at ~₹1,225, or ~3.0× book. Your base case said 1.37×. Don't shout "overvalued". Solve for the assumptions that make 3.0× fair, then judge them:

The market price is a claim: ~18% ROE, book compounding ~13% a year, for twenty years, discounted at 11%. India's best private banks have actually delivered two such decades, and they trade there because of it. But the mystique ("great bank, deserves premium") has now turned into a testable sentence with four falsifiable parts. Your job is to say which part you'd bet against, and what evidence would change your mind: deposit share, unsecured mix, credit cost through one full cycle. Valuation is a range. And the excess-return engine is how you find where today's price falls inside it.

Step 4. The scale contrast, HDFC Bank against JPMorgan. Same species, different ecosystems. All figures are illustrative, at ~FY25/CY24 scale and ~₹85/$, so pull the filings:

The last three rows tell the story. Similar multiples, opposite machines. JPMorgan earns its premium as a fee-diversified return-of-capital machine. It's a fortress that can't grow much faster than US nominal GDP plus share gains, so it mails the spread back to its owners. HDFC Bank earns a similar multiple as a return-on-capital machine. It's a narrower-spread business reinvesting three-quarters of its earnings into a banking system that's still formalising. There, deposits and credit can compound at low double digits for another decade. So when you compare P/B across geographies, you're comparing durations and reinvestment runways as well as quality. Notice, too, what scale buys JPMorgan. There's the perception of a G-SIB backstop, ~$2.4tn of deposits that ran toward it in March 2023, and stress-tested capital. Safety is part of its multiple too.

4.8 Red flags, and how banks actually blow up

This is the red-flag panel, made mechanical. If any two fire together, underweight the reported numbers. If three or more fire, assume the equity is a call option, using the distressed-valuation logic from the special-situations capstone, and analyse survival first:

1. Rising SMA-2 + rising slippage. That's tomorrow's GNPA queuing at the door, the EW1 signal in the panel. Check whether slippage guidance is being cut while SMA-2 grows. If it is, that's denial.
2. NIM compression alongside aggressive loan growth. The bank is buying market share on price, and underwriting quality almost never survives that combination. Growth >25% with NIM down >30 bp is the classic vintage-time-bomb setup.
3. Falling PCR while GNPA rises. The bank is under-providing into deterioration to protect reported ROE, so the P&L is borrowing from the future. Worked Example 2 shows it in numbers.
4. CD ratio > 90%. Loans are outrunning the deposit franchise, and growth is being funded by confidence-sensitive money. Above 100% (Yes Bank FY19: ~106%) it's a lit fuse.
5. Promoter / concentrated / connected exposure. Think loans to promoter-group entities, loans against shares, top-20 borrowers above 150% of net worth, or lending to the borrower's other lender to keep both afloat. In India, read the related-party note, the pledging data you learned to find in the full-filing capstone, and the "advances against shares/guarantees" schedule.
6. NPA divergence vs RBI. The referee re-marked the book, which is EW3 in the panel. One material divergence is a governance event. Two in a row are a pattern. Put that together with flags 1 to 5 and you have the whole story.

Then add the supporting cast. Watch for treasury gains or one-off recoveries propping up a falling PPOP, and for "other income" growing mysteriously faster than the business. Watch for rising reliance on bulk deposits (top-20 depositors above 15% of deposits). Watch for auditor changes and for senior credit or risk officers leaving. And watch for the CEO on television more often than usual.

Autopsy 1. Yes Bank (credit risk + confidence, India, 2020). These figures are approximate and come from public filings and RBI actions, so verify them in the case sources on the reading list. Yes Bank grew advances ~4× in five years, from ~₹55,000 cr to ~₹2.41 lakh cr over FY14 to FY19, a ~34% CAGR. It did it by dominating the very lending other banks were shedding: leveraged promoter groups, real estate and infra (IL&FS, DHFL, ADAG, Essel, Cox & Kings…). Reported GNPA stayed magically near 1.5 to 2% while RBI divergence disclosures kept arriving. For FY17, reported GNPA was ₹2,018 cr against an RBI-assessed ₹8,373 cr, a 4× understatement. Its CD ratio crossed 100%, and growth was increasingly funded by bulk deposits and borrowings. In September 2018 the RBI refused to extend the founder-CEO's term, the referee's ultimate red card.

Confidence did the rest. Deposits fell from ~₹2.28 lakh cr in March 2019 toward ~₹1.37 lakh cr in March 2020 as institutions fled. GNPA marched toward ~16 to 17% as the book was finally marked at its true value. On 5 March 2020 RBI imposed a moratorium with a ₹50,000 withdrawal cap and superseded the board. Then it engineered an SBI-led reconstruction, in which ₹8,415 cr of AT1 bonds were written down to zero while the equity survived, diluted. AT1 investors worldwide learned that capital-structure lesson again with Credit Suisse in 2023. Now map the flags. There's #2, hyper-growth with the highest NIM in the peer set, which is priced-for-risk lending. There's #4 (CD >100%), #5 (promoter-group concentration) and #6 (serial divergence). Then #1 and #3 joined in as the end approached.

Every flag was public years before March 2020.

Autopsy 2. Silicon Valley Bank (duration risk + confidence, US, 2023). SVB had no credit losses at all. It died of interest-rate risk and depositor concentration. Deposits tripled in two years, from ~$62bn in 2019 to ~$189bn at end-2021, as venture-funded clients parked the money they'd raised. SVB bought long, putting ~$91bn into mortgage-backed securities and Treasuries classified as held-to-maturity, yielding ~1.6%, with duration ~5.7 years and largely unhedged. Then the 2022 rate shock (+~450 bp) did the arithmetic duration promises. MTM loss ≈ duration × Δrate, which gave roughly $15bn of unrealised losses by end-2022, about equal to the bank's entire equity. Two things kept the insolvency out of the reported ratios. One was the accounting, with HTM held at amortised cost. The other was Category IV tailoring, which kept AOCI out of regulatory capital and waived the full LCR. On paper the bank was regulatorily well capitalised. Economically it was broke.

Then the liability side struck the match. Roughly 90%+ of deposits were uninsured, concentrated in one chattering industry. On 8 March 2023 SVB sold its $21bn AFS book at a $1.8bn loss and announced an equity raise. That was the signal that forced everyone to do the HTM maths. On 9 March depositors tried to pull $42bn in one day, a fifth of the bank, at fibre-optic speed. On 10 March the FDIC took it over. Map the flags and none of the six credit flags fires. What fires instead is deposit concentration, uninsured share, duration mismatch and reliance on an accounting label. For your panel, the moral is to ask two questions of any bank. What happens to book value if rates move 200 bp? Check the AFS/HTM split and the disclosed MTM to find out. And who are the depositors, and are they insured? SVB's public filings answered both questions, plainly, all along.

Two deaths, one grammar: a bank fails when a hole in the asset side (credit or duration) meets a liability side willing to run. Your KPI panel is there to find the hole while it's small. The funding metrics (CASA, CD, concentration, insured share) tell you how fast the run can be.

Common mistakes & how experts think differently

1. Valuing a bank like a company. EV/EBITDA screens, FCFF DCFs and "net debt" adjustments are all meaningless for lenders, for the reasons the opening section gave. Experts switch to the equity-direct toolkit (P/B against ROE, excess returns, capital-constrained payout) the moment the triage says lender. They treat any sell-side bank note built on EV multiples as a failed competence test.

2. Celebrating loan growth. For most companies revenue growth takes risk out of the story. For a bank it adds risk you can't see yet, because revenue is the one thing a lender can always buy. Experts ask "who is losing the business this bank is winning, and why?" Then they run the lagged-denominator recomputation on every fast grower.

3. Reading GNPA as current information. GNPA describes lending decisions made 2 to 4 years ago. Experts read the leading stack (slippage, SMA-2, restructured stock, growth mix, unsecured share), and they treat GNPA as the obituary column.

4. Taking book value literally. P/B 0.7 looks cheap until you knock off NNPA, the under-provided restructured book and a divergence. Then it's P/ABV 1.2 on a shrinking franchise. Experts adjust book before every P/B comparison. They also remember the special-situations capstone's rule that a sub-book multiple is a verdict to investigate. It isn't a discount coupon.

5. Comparing ROE without leverage and capital context. 16% ROE at 8× leverage ≠ 16% at 16×. Experts quote ROA alongside ROE and check CET1 headroom. A high-ROE bank at minimum capital is one bad cycle away from a dilutive rights issue, and that issue takes value from the very shareholders who admired the ROE.

6. Treating NIM expansion as pure skill. A mix shift toward unsecured lending raises NIM now and credit cost later. Experts break NIM moves into rate, mix and funding before they applaud. And they get suspicious when NIM and growth rise together late in a cycle, which is red flag #2.

7. Dismissing regulation as compliance trivia. PCA thresholds, PSL drag, CRR/SLR, divergence rules, category tailoring: these are the economics. You can't make sense of SVB without Category IV, of Yes Bank without divergence disclosures, or of PSU-bank valuations without PCA. Experts read the regulator's publications (the FSR, the Fed's stress results) as primary research. And they're free.

8. Anchoring on deposit franchises as permanent. CASA stickiness is a real moat, but it's a confidence asset. SVB's deposits tripled and then fled within 18 months. Yes Bank lost a third of its deposits in two quarters. Experts grade funding on granularity, insured share and behaviour under stress as well as on cost.

Worked examples

Worked example 1: Meridian Bank: the full KPI panel from raw statements

(A synthetic bank, with all figures in ₹ crore. It's the drill you'll run on real banks in the mini-project: statements in, panel out, then a verdict.)

P&L, FY25: interest income 23,600 (on advances 18,000, on investments 5,600) · interest expense 13,740 (deposits 12,240, borrowings 1,500) · NII 9,860 · other income 3,000 (fees 2,400, treasury 400, recoveries from written-off accounts 200) · operating expenses 5,400 · PPOP 7,460 · provisions 1,300 (specific 1,250, standard 50) · PBT 6,160 · tax 1,540 · PAT 4,620.

Balance sheet, period-end: gross advances 2,04,000, specific provisions held 3,574, net advances 2,00,426 · investments 82,000 · cash and bank balances 24,000 · fixed/other assets 8,574 · total assets 3,15,000 ‖ deposits 2,45,000 (current 26,000, savings 72,000, term 1,47,000) · borrowings 25,000 · other liabilities 13,000 · net worth 32,000.

Disclosed averages: total assets 3,00,000 · interest-earning assets 2,80,000 (advances 2,00,000, investments 80,000) · deposits 2,40,000 · net worth 30,000.

Asset-quality note: opening GNPA 3,796 · slippages 3,000 · recoveries + upgrades 1,500 · write-offs 400 · closing GNPA 4,896 · opening standard advances 1,88,000 · opening specific provisions 2,724. Capital: RWA 2,10,000 · CET1 27,000 · AT1 1,000 · Tier 2 3,500. Early warning: SMA-1 2,400 · SMA-2 1,800 · restructured standard book 1,600.

Step 1. Check the NPA roll-forward before anything else. 3,796 + 3,000 − 1,500 − 400 = 4,896, so it ties. So does the provisions roll: 2,724 + 1,250 − 400 = 3,574. With the plumbing confirmed, here's the panel:

Step 2. The ROA tree, as % of average assets: NII 3.29 + other income 1.00 − opex 1.80 − provisions 0.43 = pre-tax 2.05 × 0.75 = 1.54 ROA × 10× leverage = 15.4% ROE. The engine is margin-led. NII does the work and fee income is modest. Efficiency is fine and credit is quiet.

Step 3. The verdict, in five lines, the format the app grades. This is a margin-led, deposit-funded franchise earning excellent ROA on conservative leverage. Asset quality is clean on both stock (GNPA 2.4%, PCR 73%) and flow (slippage 1.6%, SMA-2 0.9%). Funding is at the edge of comfort (CASA at 40% on the nose, CD 83%), so watch whether growth ambitions push CD past 85%. Capital supports ~15% RWA growth without dilution, and nothing in the early-warning stack contradicts the reported numbers. Fair value anchors near P/B = (ROE − g)/(r − g) = (15.4 − 5)/(13 − 5) ≈ 1.3× book on one stage, and more with a credible decade of 15%+ ROE, but pay above 2× only with evidence that this ROA survives a full credit cycle.

Worked example 2: Axiom Bank: watching a lender rot in plain sight

(Synthetic. Three fiscal years, ₹ crore. Every number below was published in "record profit" press releases.)

Check the roll-forwards, always. For FY2, 1,200 + 2,100 − 840 − 300 = 2,160 (recoveries 840, write-offs 300). For FY3, 2,160 + 4,900 − 1,000 − 900 = 5,160. Both tie.

Read the deterioration the way an expert does. (1) The tell is FY2, and FY3 only confirms it. Slippage rose 75% while credit cost rose 9 bp. The provisioning line refused to follow the asset-quality line, so PCR fell from 70 to 60 to keep ROE at 15.8%. Flag #3 fired here, two years before the ROE collapsed. (2) NIM fell ~45 bp across two years of 25 to 35% growth. Axiom was buying share on price, which is flag #2. (3) The CD ratio went from 88 to 97, so growth was funded by confidence-sensitive money (flag #4). (4) The GNPA ratio understated everything throughout, because the denominator grew 62% in two years. Run the lagged check: FY3 GNPA ÷ FY1 advances = 5,160 ÷ 1,00,000 = 5.2%. That's the number the vintage actually earned.

Now normalise, as the market eventually will. Restate FY3 to a 70% PCR. Required provisions = 0.70 × 5,160 = 3,612, the shortfall = 3,612 − 2,096 = 1,516, and the post-tax hit = 1,137. Reported net worth of 13,500 normalises to 12,363. Adjusted book value = 12,363 − normalised NNPA (5,160 − 3,612 = 1,548) = 10,815. At a market cap of 16,200, the headline P/B of 1.20× looks reasonable. But a P/ABV of 1.50× is expensive for a bank with 3.7% slippage, 97% CD and compressed margins. And the FY2 to FY3 vintage (₹62,000 cr, 38% of the book) hasn't finished seasoning. At even a 4% eventual loss rate, that's ~2,500 more of losses still to be recognised, ~20% of adjusted net worth. The "cheap" bank is dear, and the workup's job is to say so before the write-offs do.

Worked example 3: SVB by the numbers: duration risk in four steps

(Figures are approximate and come from public filings and the Fed's April 2023 Barr report. Check them there. Work in $bn.)

Step 1. The asset bet. At end-2021, deposits were ~189, up from ~62 in 2019. SVB parked the flood in ~91 of held-to-maturity MBS and Treasuries at a ~1.6% yield, with duration ~5.7 years, unhedged. Modified duration is a promise: ΔMV ≈ −duration × Δrate. For +200 bp, −5.7 × 2.0% = −11.4%, or ≈ −$10.4bn on the HTM book alone.

Step 2. The rate shock. 2022 delivered ~+425 bp of policy tightening across seven FOMC moves (25, 50, 75, 75, 75, 75, 50). Convexity and portfolio runoff damp the linear estimate. SVB's filed number is ~$15.2bn of unrealised HTM losses at end-2022, against ~$16bn of total equity. Economic CET1 ≈ zero. Regulatory CET1 was ~12%, because HTM is carried at amortised cost and Category IV banks excluded AOCI marks from capital. So the insolvency was disclosed in a footnote and invisible in every ratio.

Step 3. The liability structure. Roughly 90%+ of deposits were uninsured, corporate treasuries far above the $250k FDIC cap, and they were concentrated in one networked industry. Now the run maths. 9 March 2023 saw $42bn of attempted withdrawals in one day (~22% of deposits), with ~$100bn queued for the next morning. Resources that were liquid at par, cash plus whatever the AFS sale raised, covered a fraction of that. Selling HTM would have crystallised the $15bn hole. There was no way out.

Step 4. The two questions to ask of every bank, every quarter. (i) What happens to book value if rates move 200 bp? Read the AFS/HTM split and the disclosed fair-value gap. For Indian banks that's in the investment schedule, and for US banks in the securities note. (ii) Who are the depositors? Check insured share, granularity, concentration. Then go back over the KPI panel's logic. SVB scored clean on every credit metric (GNPA ≈ 0). A panel is only as good as the risks it remembers to include, and for banks that always means credit and duration and funding.

Worked example 4: Sarayu Bank: the ROA-tree-to-ROE bridge, and what it is worth (India)

(Synthetic, in ₹ crore. Worked Example 1 built the panel line by line. This one runs the "so what?" chain the app grades: whether the ROE is real, whether it creates value, whether capital can sustain it and what book is worth.)

The numbers. Disclosed averages: total assets ₹4,00,000 · interest-earning assets ₹3,70,000 · advances ₹2,60,000 · net worth ₹34,000. P&L: NII ₹13,320 · other income ₹3,600 · opex ₹7,200 · provisions ₹1,850 · tax 25%. Period-end: gross advances ₹2,64,000 · GNPA ₹6,600 · specific provisions held ₹4,620 · deposits ₹3,10,000 · CASA ₹1,27,100. Capital: RWA ₹2,75,000 · CET1 ₹30,000 · AT1 ₹1,500 · Tier 2 ₹5,000. Cost of equity r = 13%.

Step 1. Build ROA from the tree, not from the bottom line. Express each P&L line as a percentage of average assets. NII is 13,320/4,00,000 = 3.33%, other income 0.90%, opex 1.80% and provisions 0.46%. Pre-tax ROA = 3.33 + 0.90 − 1.80 − 0.46 = 1.97%, and after 25% tax ROA = 1.48% (PAT = ₹5,902.5). Now sanity-check the branches. NIM = 13,320/3,70,000 = 3.60%, which is healthy. Credit cost = 1,850/2,60,000 = 0.71%, which is normal. Cost-to-income = 7,200/(13,320 + 3,600) = 42.6%, which is good. The tree tells you which branch earns the ROA. Here NII does the work, fees are modest, credit is quiet.

Step 2. Bridge to ROE, and interrogate the leverage. Leverage = 4,00,000/34,000 = 11.76×, so ROE = 1.48% × 11.76 = 17.36% (= 5,902.5/34,000). That clears the 13% cost of equity by ~4.4 points, so it creates value. It's also built on sane leverage (~12×, ROA 1.48%), so this is no thin-capital coiled spring. Asset quality confirms the ROE is real and isn't being borrowed from the future. PCR = 4,620/6,600 = 70%, and NNPA = (6,600 − 4,620)/(2,64,000 − 4,620) = 0.76%. The recognised book is largely paid for.

Step 3. Can capital sustain it? CET1 = 30,000/2,75,000 = 10.91%, and CRAR = 36,500/2,75,000 = 13.27%. Distance to the 8.0% CET1 floor (incl. CCB) is 2.91 pp, which is ₹8,000 of loss absorption (30,000 − 0.08 × 2,75,000). For growth headroom, retaining 75% of PAT (₹4,427) funds RWA growth of 4,427/0.1091 ≈ ₹40,580 ≈ 14.8% without dilution. That's above system credit growth, so the franchise can compound its own book without going to the market, cap in hand, for capital.

Step 4. What is it worth? On the one-stage anchor, P/B = (ROE − g)/(r − g) = (17.36 − 5)/(13 − 5) = 1.55× book. That's the floor a stable 17% ROE justifies at a 13% cost of equity. A credible decade of 17%+ ROE with reinvestment supports more, and Step 3 says the capital is there. But pay above ~2× only with evidence that this ROA survives one full credit cycle. The whole workup compresses to four sentences: real ROE, sane leverage, capital to grow, ~1.5× floor.

Worked example 5: Cascade Republic Bancorp: the same panel in US clothing (United States)

(A synthetic US regional bank; work in $bn. It runs the India-to-US translation table end to end, through the efficiency ratio, net charge-offs, ROTCE, CET1 against the stress capital buffer and P/TBV, so an India-trained eye can read a 10-K without tripping on the labels.)

The numbers. Averages: total assets 180 · earning assets 165 · loans 120 · tangible common equity 15 (common equity 17, incl. ~2 of goodwill/intangibles). Income statement: net interest income 5.28 · noninterest income 2.70 · noninterest expense 4.60 · provision for credit losses 0.55 · net charge-offs 0.45 · tax 21%. Credit: nonperforming loans 0.72 · allowance for credit losses (ACL) 1.50. Deposits: 150 total, of which 54 noninterest-bearing. Capital: CET1 16.5 · RWA 130 · regulatory CET1 requirement 4.5% + a 3.0% stress capital buffer = 7.5%.

Step 1. Margin and efficiency, in US labels. NIM = 5.28/165 = 3.20%, which is strong for a US bank. In the US the band is 2.5 to 3.5%, and India's 3 to 4.5% belongs to a different rate planet. The cost-to-income analogue is the efficiency ratio = noninterest expense / (NII + noninterest income) = 4.60/7.98 = 57.6%. That's under the 60% "good" line but not yet at the 55% "strong" line.

Step 2. Credit, in charge-off language. US banks lead with the net charge-off (NCO) rate = 0.45/120 = 0.38%. A mortgage-heavy book runs at ~0.1% and cards at 3 to 4%, so judge it against the mix. NPL ratio = 0.72/120 = 0.60% (<1% is good). ACL coverage of NPLs = 1.50/0.72 = 208%, against a US norm of >100%. That's heavier than India's PCR >70% convention because CECL front-loads lifetime expected loss.

Step 3. ROA to ROTCE. Pre-provision net revenue = 7.98 − 4.60 = 3.38. Pre-tax = 3.38 − 0.55 = 2.83, and net income = ×0.79 = 2.24. ROA = 2.24/180 = 1.24% (>1% is good). US banks quote ROTCE, return on tangible common equity, where India would quote ROE. Here 2.24/15 = 14.9%, just shy of the 15% "strong" bar. It's above the plain-equity ROE of 13.1% because goodwill is stripped out. Compare US banks on ROTCE and Indian banks on ROE, and never cross the two without adjusting for intangibles.

Step 4. Capital and price. CET1 = 16.5/130 = 12.7% against the 7.5% bespoke requirement, so there's 5.2 pp of headroom. In the US, "distance to minimum" uses the requirement including the stress capital buffer, the analogue of India's CCB-inclusive floor. Noninterest-bearing deposits = 54/150 = 36%. That's the US CASA analogue, and it's above the 25 to 30% marker, a real funding edge. Value it on P/TBV against ROTCE. Justified = (ROTCE − g)/(r − g) = (14.9 − 4)/(10 − 4) = 1.82× tangible book, using a US-regional cost of equity of ~10% and g ~4%. At a $27bn market cap the stock trades at 27/15 = 1.80× P/TBV, fairly priced for what it earns. It's the same excess-return engine as Sarayu. Only the vocabulary changed.

Practice set

Work in ₹ crore unless stated. Problems 1 to 5 are guided, 6 to 11 are independent and 12 is timed.

Problem 1 (guided). Ravi Bank has average advances of 50,000 yielding 9.2% and average investments of 20,000 yielding 7.1%. Its average deposits are 60,000 costing 5.3%, and its average borrowings 5,000 costing 7.8%. Compute NII and NIM, and explain why NIM ≠ (yield on advances − cost of deposits).

Solution. Interest income = 50,000 × 9.2% + 20,000 × 7.1% = 4,600 + 1,420 = 6,020. Interest expense = 60,000 × 5.3% + 5,000 × 7.8% = 3,180 + 390 = 3,570. So NII = 2,450. Earning assets are 70,000, which gives NIM = 3.50%. Advances alone give a spread of 9.2 − 5.3 = 3.9%. NIM comes in lower because 29% of earning assets are low-yield investments at 7.1%, and that's only partly offset by equity funding a slice of assets at zero cost. NIM is the blended margin of the whole earning balance sheet. The loan spread is only one piece of it.

Problem 2 (guided). Two banks each have deposits of 1,00,000, average assets of 1,20,000, equity of 10,000 and a 25% tax rate. Bank A: CASA 45% (CA 15% at 0%, SA 30% at 3%), term 55% at 7.5%. Bank B: CASA 25% (CA 8%, SA 17%), term 75% at 7.5%. Quantify A's funding advantage in bp, in ₹ and in ROE points.

Solution. Cost A = (0×15 + 3×30 + 7.5×55)/100 = 5.025%. Cost B = (0×8 + 3×17 + 7.5×75)/100 = 6.135%. A's advantage = 111 bp = ₹1,110 pre-tax on 1,00,000 of deposits = 833 post-tax = 8.3 ROE points (833/10,000). A CASA franchise 20 points wider is worth more ROE than most banks' entire fee businesses. So deposit franchises are the moat. It's also why "we'll fix funding later" growth stories usually can't.

Problem 3 (guided). In the year, Surya Bank had opening gross advances of 1,00,000, of which GNPA was 6,000. Slippages were 2,800, recoveries + upgrades 1,200 and write-offs 1,600. Closing gross advances were 1,05,000. Opening specific provisions were 4,200, and P&L provisions 1,900 (all specific). Compute closing GNPA (and its ratio), the slippage ratio, credit cost, closing PCR and NNPA. Then explain why the "improving" GNPA ratio is misleading.

Solution. Closing GNPA = 6,000 + 2,800 − 1,200 − 1,600 = 6,000, a ratio of 6,000/1,05,000 = 5.71% against 6.00% at the open. Slippage = 2,800 ÷ opening standard advances (1,00,000 − 6,000 = 94,000) = 2.98%. Credit cost = 1,900 ÷ average advances (≈1,02,500) = 1.85%. Closing provisions = 4,200 + 1,900 − 1,600 = 4,500, so PCR = 75%. NNPA = (6,000 − 4,500)/(1,05,000 − 4,500) = 1,500/1,00,500 = 1.49%. The ratio "improved" for two reasons only: (i) advances grew 5%, and (ii) 1,600 of write-offs left the numerator. Meanwhile the inflow, slippage at ~3%, stayed at stress levels. Stock optics look better, but the flow tells the truth. This bank is still bleeding.

Problem 4 (guided). As % of average assets, Bank P reports NII 3.2, other income 1.1, opex 2.1 and provisions 0.5. Tax is 25% and leverage 11×. (a) Find ROA and ROE. (b) Bank Q reports the same 14.0% ROE from ROA of 0.70% at 20× leverage. A shock costs each bank 0.75% of assets, post-tax. Compare the damage.

Solution. (a) ROA = (3.2 + 1.1 − 2.1 − 0.5) × 0.75 = 1.7 × 0.75 = 1.275%, and ROE = 1.275 × 11 = 14.0%. (b) Per 100 of assets, P's equity = 100/11 = 9.09 and Q's = 100/20 = 5.00. A 0.75 shock is 8.3% of P's equity but 15% of Q's. It wipes out 59% of P's annual ROA but 107% of Q's, so Q's year goes negative. Same ROE, wildly different fragility. Always decompose ROE into ROA × leverage before you compare banks.

Problem 5 (guided). This year Nalanda Bank has total income (NII + other) of 100 and opex of 48. (a) Next year income grows 12% and opex 8%. What are the new cost-to-income ratio and PPOP growth? (b) From the same starting point, income grows 6% and opex 12%. (c) State the "jaws" rule.

Solution. (a) Income is 112 and opex 51.84, so C/I = 46.3%, down from 48%. PPOP goes from 52 to 60.16, up 15.7%. (b) Income is 106 and opex 53.76, so C/I = 50.7%. PPOP goes from 52 to 52.24, up just 0.5%, which is flat profit despite growing revenue. (c) Jaws = income growth − opex growth. Because opex is ~half of income, every point of positive jaws adds roughly two points to PPOP growth. Negative jaws strangle it, and slowly enough that nobody notices. Efficiency is operating leverage, both ways.

Problem 6. Vindhya Bank holds G-secs of 30,000 at a risk weight of 0% and housing loans of 40,000 at 50%. Its AAA corporates are 20,000 at 20% and its BBB corporates 30,000 at 100%. Unsecured personal loans are 10,000 at 125%, and operational plus market RWA add 8,000. Capital is CET1 8,200 plus AT1 500 plus Tier 2 1,700. NNPA is 2.1%. Compute RWA, the CET1 ratio, CRAR and the distance to PCA on both capital and asset quality. Then find the loss the bank can absorb before it breaches the CET1 floor.

Solution. Credit RWA = 0 + 20,000 + 4,000 + 30,000 + 12,500 = 66,500, so total RWA = 74,500. CET1 = 8,200/74,500 = 11.0%. Tier 1 = 8,700/74,500 = 11.7%. CRAR = 10,400/74,500 = 13.96%. Including CCB, the floors are CET1 8.0% and CRAR 11.5%, which leaves headroom of 300 bp and 246 bp. Loss absorption before a CET1 breach ≈ 3.0% × 74,500 = ₹2,235. On asset quality, NNPA of 2.1% against the PCA risk-threshold-1 of 6% is 390 bp away, the safer margin. So capital is the binding constraint. One loss year at 1.5% of advances (≈ 1,500 on ~1,00,000 of loans) eats two-thirds of the headroom. Notice the composition, too. The 10,000 of unsecured personal loans uses up 12,500 of RWA. That's 17% of total RWA from 7.7% of the exposures shown.

Problem 7. Trishul Bank starts with advances of 80,000 (GNPA 1,920, 2.4%). In Year 1 advances grow 45% to 1,16,000. Slippages are 1.4% of the opening standard book, and recoveries + write-offs come to 653. In Year 3, two years later, advances are 1,21,800. That year's slippages are 2,900 and recoveries + write-offs 800, with GNPA of 2,360 entering the year. Compute the GNPA ratio after Year 1 and after Year 3. Then compute the lagged ratio (GNPA ÷ advances two years earlier) for each, and state the lesson.

Solution. Year 1 slippages = 1.4% × (80,000 − 1,920) = 1,093. Closing GNPA = 1,920 + 1,093 − 653 = 2,360, a ratio of 2,360/1,16,000 = 2.03%. That's "improved" from 2.4%, while the stock rose 23%. In Year 3, GNPA = 2,360 + 2,900 − 800 = 4,460, and 4,460/1,21,800 = 3.66%. That reads as "deterioration," though underwriting never changed. Year 1's 36,000 vintage simply seasoned. Now the lagged ratios. For Year 1, 2,360/80,000 = 2.95%, against the headline 2.03%. For Year 3, 4,460/1,16,000 = 3.84%, close to the headline because growth stopped. So growth flatters the ratio at the very moment risk is being added, and the lagged denominator takes away the anaesthetic. For any bank growing >25%, headline GNPA is marketing.

Problem 8. Work in $bn. Pinewood Bancorp has a securities book of 84, made up of HTM 54 and AFS 30, which is already marked through equity. Portfolio duration is 6.2 years and CET1 capital is 11. Deposits are 130, of which 60% are uninsured, and liquid assets at par are 18. Rates rise 150 bp. (a) Find the total MTM hit and the hidden (HTM) portion, against CET1. (b) If 20% of uninsured deposits try to leave within 48 hours, does liquidity hold? (c) Name the two disclosures you'd have read in advance.

Solution. (a) ΔMV ≈ −6.2 × 1.5% = −9.3% × 84 = −7.81 in total. Of that, the hidden HTM share = −9.3% × 54 = −5.02 = 46% of CET1, since the AFS 2.79 is already in equity. If HTM were marked, economic CET1 would be ≈ 11 − 5.0 ≈ 6.0. (b) Outflow = 20% × (0.60 × 130) = 15.6 in two days, against 18 of liquid assets. It survives 48 hours with 2.4 to spare. But at this run rate (~7.8/day), day 3 forces HTM sales that crystallise the 5.0 loss and probably speed up the run. This bank is 3 days from failure, and calling it "adequately liquid" would be wrong. (c) The securities note, with its AFS/HTM split and fair-value gap, and the uninsured-deposit disclosure. It's SVB's grammar again, a hole (duration) meeting a run (uninsured concentration).

Problem 9. Kanan Bank has book value of 100 and a cost of equity of 13%. In years 1 to 5 it earns ROE of 15% and pays out one-third, so g = 10%. From year 6 competition arrives. In case (a), ROE falls to 13% forever. In case (b), ROE holds at 15% with g = 5% forever. Value the equity both ways and state the moral.

Solution. Stage 1 excess returns = (15% − 13%) × opening book: 2.00, 2.20, 2.42, 2.662, 2.9282. Their PVs at 13% are 1.770, 1.723, 1.677, 1.633 and 1.589, which sum to 8.39. Book₅ = 100 × 1.10⁵ = 161.05. In case (a), terminal ROE = r, so terminal P/B = 1 and the terminal excess value = 0. That gives V = 108.4, or 1.08× book. In case (b), terminal P/B = (15 − 5)/(13 − 5) = 1.25. Excess value at t=5 = 0.25 × 161.05 = 40.26, its PV = 40.26/1.8424 = 21.85, and V = 130.2, or 1.30× book. So the duration of the spread, not this year's ROE, is most of a bank's value above book. That's why the whole Phase 4 moat apparatus decides bank multiples, which here means the CASA franchise and the underwriting culture.

Problem 10. Five banks trade at (ROE %, P/B): A (17, 2.9), B (15, 2.4), C (12, 1.5), D (10, 1.0), E (8, 0.7). Bank F earns ROE 14% and trades at 1.1× book. (a) Fit the least-squares line on A to E. (b) Where "should" F trade? (c) List, in order, the three things you check before calling F cheap.

Solution. (a) x̄ = 12.4 and ȳ = 1.7. Σ(x−x̄)(y−ȳ) = 5.52 + 1.82 + 0.08 + 1.68 + 4.40 = 13.50, and Σ(x−x̄)² = 21.16 + 6.76 + 0.16 + 5.76 + 19.36 = 53.20. Slope = 13.50/53.20 = 0.254, and intercept = 1.7 − 0.254 × 12.4 = −1.45. So the market pays book only above ROE ≈ (1 + 1.45)/0.254 ≈ 9.6%, its rough read of the sector's cost of equity minus growth. (b) F's predicted P/B = 0.254 × 14 − 1.45 = 2.11×, and its actual 1.1× is ~48% below the line. (c) Check three things, in this order. (1) Is the book real? Look at NNPA, the PCR trend, restructured stock, any divergence. Then recompute P/ABV. (2) Is the ROE real and durable? Look at leverage against peers and one-off gains, then compare credit cost with slippage. (3) Is there a governance or ownership overhang? Promoter exposure or a pending capital raise would count. If all three come back clean, F is a candidate. Your variant question is then why the market disagrees.

Problem 11. Godavari Bank reports gross advances of 1,20,000 and GNPA of 3,000 (2.5%), with provisions held of 2,100 (PCR 70%). PAT is 1,500, net worth 12,000 and market cap 18,000 (P/B 1.5×). RBI's inspection report, disclosed under the divergence rules, finds additional GNPA of 1,800 and additional provisions required of 720, at a tax rate of 25%. Restate GNPA, PCR, NNPA, PAT, net worth, ABV and P/ABV, then write the two-sentence governance read.

Solution. Restated GNPA = 4,800, or 4.0%. Provisions = 2,820, so PCR = 58.8%, down from 70%. NNPA before = (3,000 − 2,100)/(1,20,000 − 2,100) = 0.76%, and after = (4,800 − 2,820)/(1,20,000 − 2,820) = 1,980/1,17,180 = 1.69%. PAT = 1,500 − 720 × 0.75 = 960, down 36%. Net worth = 12,000 − 540 = 11,460. ABV = 11,460 − 1,980 = 9,480. P/ABV = 18,000/9,480 = 1.90×, against a headline restated P/B of 1.57×. As a governance read, the referee found 60% more bad loans than management reported. Treat every management estimate in this bank's accounts, from PCR adequacy to the restructured book to fee accruals, as an opening bid. A second consecutive divergence would make the accounts themselves the thesis.

Problem 12 (timed at 25 minutes, the exam-condition workup). This is Kaveri Bank's FY summary, in ₹ cr. Disclosed averages are total assets 5,00,000, earning assets 4,60,000, advances 3,20,000 and net worth 36,000. P&L: NII 14,720 · other income 4,000 · opex 9,360 · provisions 3,000 · tax 25%. Period-end: gross advances 3,25,000 · specific provisions held 7,605 · GNPA 12,675 · deposits 3,42,000 · CASA 1,06,020 · net worth 38,000 · CET1 ratio 9.4% · CRAR 13.1%. Flow: opening standard advances 3,00,000 · slippages 9,750 · SMA-2 6,175. Market cap is 34,600. Compute the twelve-KPI panel, classify each KPI, and deliver a five-line verdict that names the two deadliest issues and a fair-multiple range.

Solution. NIM = 14,720/4,60,000 = 3.20%, healthy but at the low end. C/I = 9,360/18,720 = 50.0%, which is weak. PPOP = 9,360. Credit cost = 3,000/3,20,000 = 0.94%, elevated. PAT = (9,360 − 3,000) × 0.75 = 4,770, so ROA = 4,770/5,00,000 = 0.95%, below the 1% bar, and ROE = 4,770/36,000 = 13.3%, also below the bar. GNPA = 12,675/3,25,000 = 3.9%, above 3%. PCR = 7,605/12,675 = 60%, below 70%. NNPA = (12,675 − 7,605)/(3,25,000 − 7,605) = 5,070/3,17,395 = 1.60%, above 1%. CASA = 1,06,020/3,42,000 = 31%, weak. CD = 3,25,000/3,42,000 = 95%, a red flag. Slippage = 9,750/3,00,000 = 3.25%, stress. SMA-2 = 6,175/3,25,000 = 1.9%, on watch and near 2%. On capital, CET1 of 9.4% is only 140 bp above the 8% floor (≈ ₹5,700 of absorption on implied RWA ≈ 4,07,000), and CRAR headroom is 160 bp.

Verdict: this is a below-hurdle franchise. ROA is 0.95% and ROE 13.3% against a ~13% cost of equity, which is value-neutral at best. Stress is flowing in (slippage 3.25%, SMA-2 1.9%), and provisioning is already behind it (PCR 60%). Funding is stretched (CASA 31%, CD 95%), so growth can't rescue earnings without expensive money. And capital headroom of 140 bp can't absorb one bad year of this slippage at 50% LGD. Deadliest of all is the pair CD 95% + CET1 headroom 140 bp, because a funding squeeze and a dilutive raise are both live. ABV = 38,000 − 5,070 = 32,930. At market, P/B = 34,600/38,000 = 0.91×, but P/ABV = 1.05×. A fair range is ~0.7 to 0.9× ABV until slippage turns and PCR rebuilds. The current price offers no margin of safety for a bank this close to its constraints. (Score yourself: 1 point per correct KPI + classification, 12 max. Add 4 for a verdict that names CD + capital. Deduct 2 if you called P/B 0.91× "cheap" without the ABV and flow adjustments.)

Problem 13 (guided). As % of average assets, Tungabhadra Bank reports NII 3.1, other income 1.0, opex 1.9 and provisions 0.6. Tax is 25%, leverage 12.5× and the cost of equity 13%. (a) Compute ROA and ROE and say whether the bank creates value. (b) A push into unsecured retail lifts NII to 3.4% of assets but raises provisions to 1.1%. Recompute ROA and ROE. (c) NIM rose. Is the bank better off? State the rule.

Solution. (a) ROA = (3.1 + 1.0 − 1.9 − 0.6) × 0.75 = 1.6 × 0.75 = 1.20%, and ROE = 1.20 × 12.5 = 15.0%. That beats the 13% cost of equity by 2 points, so the bank creates value. (b) ROA = (3.4 + 1.0 − 1.9 − 1.1) × 0.75 = 1.4 × 0.75 = 1.05%, and ROE = 1.05 × 12.5 = 13.13%. (c) No. The higher margin was bought with credit cost that rose more than the margin gained. ROA fell, and the value spread collapsed from 2.0 points to ~0.1. It's the KPI 1 trap again. A rising NIM driven by a mix shift into unsecured lending is deferred credit cost dressed up as alpha. The yield books now, and the losses season in 12 to 24 months. Decompose every NIM move into rate, mix and funding before you applaud it.

Problem 14 (guided). Chambal Bank has gross advances of 1,50,000 and GNPA of 6,000 (4.0%), with specific provisions held of 3,000 (PCR 50%). Net worth is 14,000, market cap 16,800 (P/B 1.2×) and tax 25%. (a) Find the current NNPA ratio. (b) Restate provisions to a 70% PCR. Find the shortfall and the post-tax hit, then normalised net worth. (c) Find adjusted book value and P/ABV. (d) Give a one-line read.

Solution. (a) NNPA = (6,000 − 3,000)/(1,50,000 − 3,000) = 3,000/1,47,000 = 2.04%. (b) Required provisions at 70% = 0.70 × 6,000 = 4,200. So the shortfall = 4,200 − 3,000 = 1,200, the post-tax hit = 1,200 × 0.75 = 900, and normalised net worth = 14,000 − 900 = 13,100. (c) Normalised NNPA = 6,000 − 4,200 = 1,800. ABV = 13,100 − 1,800 = 11,300, and P/ABV = 16,800/11,300 = 1.49×, against a headline P/B of 1.20×. (d) The "cheap" 1.2× book is 1.49× once provisioning is done properly. That discount came from weak provisioning (PCR 50%), and there was never an opportunity in it. Adjust book before you ever call a bank cheap (Common Mistake #4).

Problem 15. Bhima Bank has CET1 capital of 18,000 and RWA of 1,50,000, a CET1 ratio of 12.0%. PAT is 2,400 and dividend payout 20%, and the internal CET1 target floor is 11.0%. It plans 18% RWA growth next year and raises no external capital. (a) Find the year-end CET1 ratio. (b) Does it clear the 11% internal target? (c) What RWA growth could retained earnings fund while holding CET1 at 12.0%? State the trap for fast growers.

Solution. (a) Retained earnings = 0.80 × 2,400 = 1,920, so new CET1 = 19,920. New RWA = 1,50,000 × 1.18 = 1,77,000, and the CET1 ratio = 19,920/1,77,000 = 11.25%. (b) Yes, but by only 25 bp, and it fell from 12.0%, so the bank ate its buffer to grow. (c) Holding CET1 at 12.0%, max RWA = 19,920/0.12 = 1,66,000, so internally funded growth = 10.67%. Above ~10.7% growth this bank uses up capital faster than it makes it. Keep growing at 18% and it ends in a dilutive raise. That's the trap from KPI 12: too little capital to grow out of problems, too little profit to build capital. Growth is capital-hungry even when it looks free.

Problem 16. Cardinal Financial is a US regional, so work in $bn. Average earning assets are 90 and total assets 100, with tangible common equity of 8.5. Income: NII 2.79 · noninterest income 1.35 · noninterest expense 2.55 · provision 0.30 · net charge-offs 0.24 on average loans of 62 · tax 21%. Credit: NPLs 0.40 · ACL 0.90. Capital: CET1 capital 9.0 · RWA 72 · regulatory CET1 requirement 7.4% (4.5% + 2.9% SCB). Compute, in US terms, NIM, the efficiency ratio, NCO rate, NPL ratio, ACL coverage, ROA, ROTCE and CET1 headroom, and give a one-line verdict.

Solution. NIM = 2.79/90 = 3.10%. Efficiency ratio = 2.55/(2.79 + 1.35) = 2.55/4.14 = 61.6%, which is weak (>60). PPNR = 1.59 and pre-tax = 1.29. Net income = ×0.79 = 1.019. ROA = 1.019/100 = 1.02%, good. ROTCE = 1.019/8.5 = 12.0%, below the 15% strong bar. NCO rate = 0.24/62 = 0.39%. NPL ratio = 0.40/62 = 0.65% (<1% is good). ACL coverage = 0.90/0.40 = 225%, which is >100% and healthy under CECL. CET1 = 9.0/72 = 12.5%, so headroom = 5.1 pp = $3.67bn over the 7.4% requirement. Verdict: clean asset quality and ample capital, but a fair-not-premium earner. Efficiency >60% and ROTCE of only ~12% say the franchise is adequate. It isn't elite. Only the translation labels changed. The KPI logic didn't.

Problem 17. Kosi Bank trades at 2.2× book, with a cost of equity of 13% and sustainable growth of 6%. (a) Using the one-stage identity P/B = (ROE − g)/(r − g), what constant ROE does the price imply? (b) The bank currently earns 16% ROE. What does the gap tell you? (c) At a 16% ROE, what one-stage P/B is justified, how far above it is the price, and why should you not simply shout "overvalued"?

Solution. (a) 2.2 = (ROE − 6)/(13 − 6), so ROE − 6 = 2.2 × 7 = 15.4 and the implied ROE = 21.4%. (b) The market is pricing a sustained ~21% ROE while the bank earns 16%. So the price embeds either a large step-up in ROE or a high-return duration longer than one stage captures. (c) Justified one-stage P/B = (16 − 6)/(13 − 6) = 10/7 = 1.43×, and the 2.2× price is ~54% above it. But the warning from the valuation section holds, because the one-stage model understates multi-decade compounders. So the useful output is a testable sentence: "this price requires ~21% ROE, or roughly two decades of 16% ROE compounding book at 6%." Name the plank you'd bet against, and don't pronounce a verdict from the multiple alone.

Problem 18 (timed at 25 minutes, the stretch synthesis). Damodar Bank (₹ cr) has grown fast, from advances of 1,15,000 two years ago to 2,20,000 now. This year: gross advances 2,20,000 · GNPA 5,500 · PCR 52% · opening standard advances 1,57,000 · slippages 6,600 · SMA-2 4,800 · deposits 2,36,000 · net worth 20,000 · market cap 30,000 · tax 25%. NIM fell from 3.5% to 3.0% across the two growth years. Compute the headline GNPA ratio, the lagged GNPA ratio, slippage, SMA-2, the CD ratio and NNPA. Normalise book to a 70% PCR and compute ABV and P/ABV. Then name the red flags that are firing and deliver a three-line verdict.

Solution. Headline GNPA = 5,500/2,20,000 = 2.50%, which looks pristine. Lagged GNPA = 5,500/1,15,000 = 4.78%, the number the two-year-old book actually earned. That's the denominator fix from the credit-cycle section, with growth of ~38%/yr, or +91% over two years. Slippage = 6,600/1,57,000 = 4.20%, stress. SMA-2 = 4,800/2,20,000 = 2.18%, above the 2% line, the pipe behind a rising reservoir. CD ratio = 2,20,000/2,36,000 = 93.2%, red. Provisions held = 0.52 × 5,500 = 2,860, so NNPA = (5,500 − 2,860)/(2,20,000 − 2,860) = 1.22%. Now normalise to a 70% PCR. Required provisions are 3,850, so the shortfall is 990. The post-tax hit is 742.5, and normalised net worth = 19,257.5. Normalised NNPA = 1,650 and ABV = 17,607.5. Headline P/B = 30,000/20,000 = 1.50×, but P/ABV = 30,000/17,607.5 = 1.70×. Four red flags are firing: #1 (SMA-2 2.2% + slippage 4.2%, both high), #2 (NIM down 50 bp through ~38%/yr growth), #3 (PCR 52%, under 70 into a rising GNPA stock) and #4 (CD 93%).

Verdict: the 2.5% headline GNPA is a growth mirage, with lagged GNPA at 4.8%, slippage at 4.2% and a huge unseasoned vintage still to ripen. Provisioning already trails the deterioration, and the "cheap" 1.5× book is 1.70× P/ABV on a franchise that funds growth with confidence-sensitive money. Underweight, and the next question is the loss rate of the last two years' vintage as it seasons. (Score: 1 point per correct figure, 11 max, plus 3 for naming ≥3 correct flags and 2 for a verdict that refuses to call 1.5× P/B "cheap".)

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

Task. Run the full playbook on one Indian bank and one US bank. Deliver a two-page workup for each, plus a one-paragraph cross-market comparison. For India, choose from SBI, ICICI Bank, HDFC Bank, Axis Bank, Kotak Mahindra Bank or Federal Bank. For the US, choose JPMorgan, Bank of America, Wells Fargo or a large regional (e.g., PNC, USB). Pick at least one bank not used as an example above.

Data to pull (all free). For India, get the latest annual report and the last four quarterly investor presentations and press releases from the bank's website or BSE/NSE. Those presentations carry the asset-quality movement and the SMA/restructured and divergence disclosures. Add the Basel III Pillar 3 disclosure for capital detail, and screener.in for the 5-year ratio history. For the US, the sources are the 10-K and latest 10-Q from SEC EDGAR, the quarterly earnings supplement (NPLs, NCOs, ACL walk) and the bank's UBPR from ffiec.gov for instant peer percentiles. Cross-check any policy threshold against rbi.org.in, whether it's a CRAR minimum or the PCA matrix.

Deliverables. (1) The 12-KPI panel for 5 years, with each cell classified healthy/watch/stress. For the US bank, use the India-to-US translation table. (2) The NPA/NPL roll-forward for the last 2 years, rebuilt from disclosures, with slippage (or NCO) trends. (3) A funding-quality paragraph: CASA/noninterest-bearing share, CD/LDR, depositor concentration, insured share. (4) Capital: current ratios, distance to regulatory floors (PCA thresholds or the SCB requirement) and growth headroom. (5) Valuation. Place the bank on a P/B-vs-ROE scatter of ≥8 peers, run a two-stage excess-return model, and reverse the current price into its implied ROE-duration claim. (6) The six-flag red-flag sweep with evidence for each (including ABV and any divergence). (7) The verdict: an adapted 8-line teardown that ends with the one question you'd answer next.

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. Pass this and a mastery-check form at ≥85%, and the NBFC playbook unlocks.

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

Case Lab: run case CL-5.1 (The Two Balance Sheets of Ashokan Bank, synthetic) in the app's Cases tab. It shows one bank at the peak of a credit boom and again two years later, so you can feel the denominator effect, vintage seasoning and the PCR-versus-slippage tell move the twelve-KPI panel in real time. It's the lag from the credit-cycle section, made tactile.

Reading & resources

• Damodaran, "Valuing Financial Service Firms" (free paper, pages 1 to 25) + the eqexret.xls spreadsheet, at stern.nyu.edu. The theory behind the valuation section, from the source. [Free] [Intermediate]
• Golin & Delhaise, The Bank Credit Analysis Handbook (2e): chs. 1–6 (the credit analyst's view of banks, asset quality, earnings) and ch. 13 (capital). The deepest single book on reading banks. [Paid] [Advanced]
• Damodaran, Investment Valuation (3e), ch. 21: valuing financial service firms; pairs with M3.10. [Paid] [Intermediate]
• McKinsey, Valuation (8e), the banks chapter: the practitioner's excess-return treatment. [Paid] [Advanced]
• Zerodha Varsity, Sector Analysis module, banking chapters: an India-first walkthrough of bank financials at beginner level, and good revision. [Free] [Beginner]
• RBI, Financial Stability Report (semi-annual) and Report on Trend and Progress of Banking in India (annual), at rbi.org.in. System GNPA, credit growth, stress tests; the free macro layer for every Indian bank workup. [Free] [Intermediate]
• RBI Master Circular, Prudential Norms on IRAC and the Basel III Capital Regulations Master Direction: the primary law behind the KPI panel and the regulatory map. Skim them for structure and cite them for thresholds. They're where you check every threshold quoted above. [Free] [Advanced, reference]
• Tamal Bandyopadhyay, Pandemonium: The Great Indian Banking Tragedy. The IL&FS/Yes Bank era narrated by India's best banking journalist. It makes the credit cycle and the blow-ups visceral. [Paid] [Beginner–Intermediate]
• Federal Reserve, Review of the Federal Reserve's Supervision and Regulation of Silicon Valley Bank (the Barr report, April 2023), at federalreserve.gov. Read the executive summary and ch. 2; the primary source for Worked Example 3. [Free] [Intermediate]
• JPMorgan annual report + Jamie Dimon's shareholder letter (jpmorganchase.com; 10-K on EDGAR): the best free education in how a fortress bank describes its own risks; read one full letter per year. [Free] [Intermediate]
• FFIEC UBPR (ffiec.gov/ubpr): free peer-percentile ratios for any US bank; the US workup shortcut. [Free] [Intermediate]
• FDIC Quarterly Banking Profile: system-level US banking data each quarter. [Free] [Intermediate]
• CFA Program curriculum, "Analysis of Financial Institutions" (Level II FSA): a rigorous, exam-grade treatment of bank analysis incl. CAMELS. [Paid] [Advanced]
• screener.in / Trendlyne (India) and stockanalysis.com (US): 10-year KPI histories for the scatter plots and panels. [Free] [Beginner]
• Going deeper: Marc Rubinstein, Net Interest (netinterest.co): a weekly newsletter dissecting banks, fintechs and financial-sector blow-ups (SVB, Credit Suisse, and more) in the language used here. It's the best free ongoing education in reading lenders as they happen. [Free] [Intermediate]
• Going deeper: Moorad Choudhry, The Principles of Banking (2e): the asset-liability-management view that the opening section and the SVB autopsy only sketch. It covers maturity transformation, the liquidity gap, the funding curve and how a balance sheet is actually run day to day. Read it once you can compute the panel and want to understand the machine behind it. [Paid] [Advanced]

The Modern Analyst's Addendum

Everything above teaches this skill from first principles, by hand. 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 full bank playbook: the ROA tree, the twelve-KPI panel with its thresholds, the NPA roll-forward and the leading stack (slippage, SMA-2, restructured, divergence), the RBI/US regulatory maps, P/B-against-ROE and the excess-return model, and the six-flag red-flag sweep",
  "use": "This is the module where a general-purpose tool is most likely to be confidently, catastrophically wrong, because banking is where the ordinary ratio vocabulary means the opposite of what it says. Use it on the four jobs it is genuinely good at. (1) EXTRACTION at scale: pull the asset-quality movement table, the SMA and restructured disclosures, the top-20 borrower and depositor concentration lines, and the AFS/HTM split out of forty quarterly investor presentations and 10-Qs into one tidy frame — this is transcription with judgement, and it is hours of your life. (2) DIVERGENCE and disclosure archaeology: ask it to find every RBI divergence disclosure, auditor change, and senior-risk-officer exit in five years of filings, with the page cite. (3) TRANSLATION between the two vocabularies of section 4.6 — hand it a US call-report line and ask which Indian line it maps to, then check it against the module's own table, because this is exactly where it will invent a plausible mapping. (4) The DEFINITION audit: banks restate what 'PCR' or 'restructured' means between decks; ask it to diff the definition footnotes quarter by quarter. What you must never do is ask it for a verdict on a bank, or for a threshold. Thresholds are the module's, calibrated to segment and cycle; a model asked 'is 2.4% GNPA good?' will answer, fluently, without knowing the growth rate, the vintage mix, the write-off policy or the denominator.",
  "tools": ["Chat assistants — Claude, ChatGPT for the India-US vocabulary translation and the definition-footnote diff", "Document/PDF AI — for extracting the asset-quality movement, SMA, concentration and AFS/HTM tables out of investor presentations and 10-Qs at scale", "Coding assistants — for the sector-gated panel engine, the cross-sectional P/B-ROE fit and the credit-cost normalisation below", "XBRL and filing APIs — SEC EDGAR companyfacts and the FFIEC/FDIC bulk endpoints, driven from your notebook rather than by hand"],
  "prompt": "From the attached quarterly investor presentations for <BANK>, extract into one table, per quarter: gross advances, GNPA, specific provisions held, slippages, recoveries, upgrades, write-offs, opening standard advances, SMA-1, SMA-2, restructured standard book, deposits, CASA split, CET1, CRAR, RWA. Cite the page for every cell. Where a figure is not disclosed, write NOT DISCLOSED — never interpolate, never carry forward. Then, separately: list every place the definition footnote for PCR, restructured or SMA changed between quarters, quoting both versions. Do not compute any ratio and do not offer a view on asset quality; I am doing that.",
  "verify": "Two verifications, and the first one is the whole module. First, the ROLL-FORWARD must tie: closing GNPA = opening + slippages - recoveries - upgrades - write-offs, and closing provisions = opening + P&L specific provisions - write-offs. Section 8's Meridian panel ties on both (3,796 + 3,000 - 1,500 - 400 = 4,896; 2,724 + 1,250 - 400 = 3,574) and so does Axiom's FY2 and FY3. An extraction that does not tie has a wrong cell in it, and you will not find it by re-reading the answer — you find it by running the identity. Second, the SECTOR GATE: before any generic ratio is computed on a lender, it must be refused. Run the engine below on a bank you already know is healthy and confirm it declines to produce interest coverage, EV/EBITDA and net debt rather than producing them wrongly. And hold the line on thresholds: every band in section 4.4 is a rule of thumb for Indian commercial banks mid-cycle, and the module tells you to calibrate; a tool will not.",
  "diy": "The gate is unaided. You build the twelve-KPI panel from raw statements, tie the NPA roll-forward, run the ROA tree to ROE, compute distance-to-PCA in rupees, value a bank on the one-stage identity and on a multi-stage excess-return model, restate book to ABV, and run the six-flag sweep — all on paper, in the exam, with a calculator. Nothing in this callout is admissible in that room, and the reason is not ceremony: the analyst who cannot tie a roll-forward by hand cannot tell a tool's extraction error from a bank's."
}

Modern Data Analysis

By hand first. You built Meridian's panel line by line and tied both roll-forwards. You ran the ROA tree to a 15.4% ROE, restated Axiom's book to a 70% PCR and computed distance-to-PCA in rupees. Keep every one of those in your hands. Code won't add much arithmetic speed here, since the panel is only thirteen divisions. What it adds is that the sector classification stops being something you remember and becomes something the engine can't proceed without. It also means a single bank stops being the unit of analysis.

Today's workflow. Write the panel as a function whose first statement is a refusal. Elsewhere in the course, a separate workstream had to rebuild its scoring engine. It had scored HDFC Bank 0 out of 100 and flagged "thin interest coverage" on a bank, because generic ratio machinery had been pointed at a business where interest expense is cost of goods sold. A formula bug didn't cause that. A missing assertion did, and "why banks are different", the argument you started with, is the specification for it:

• assert sector not in LENDERS guards every EV-based metric, every FCFF, every coverage ratio, every net-debt calculation. For a bank, debt is raw material. Cash is inventory, and EBIT doesn't exist.
• The lender branch computes NIM, GNPA, NNPA, PCR, credit cost, slippage, ROA, ROE, cost-to-income, CASA, CD and CRAR. Its first statement is the roll-forward identity, and it raises an error on any bank whose published movement table doesn't tie.
• Thresholds go in a table keyed by market and segment, never as literals in the code. That's what the India-to-US translation table is for. A 3.5% NIM is healthy in India and remarkable in the US, and 60% ACL coverage is a disaster in India and impossible in the US.

Written that way, the engine has one useful property. It fails loudly on an entity it doesn't understand, instead of producing a number without a word of warning. For the modern analyst, that's all there is to "sector analysis." Everything else (the panel, the scatter, the excess-return model) is arithmetic you already know.

Tools & sources (IN + US). Use pandas for the multi-quarter panel, numpy and scipy.stats.linregress for the cross-sectional fit and its residuals, matplotlib for the P/B-against-ROE scatter, and plain assert statements for the gate.

For India, start with the bank's own quarterly investor presentation and press release on its website and on BSE/NSE. They carry the asset-quality movement, SMA and restructured tables that no aggregator carries. The Basel III Pillar 3 disclosure gives the RWA split, capital composition and leverage ratio. RBI's Database on Indian Economy (DBIE) has the series for system credit, deposits and sectoral deployment. RBI's semi-annual Financial Stability Report, its annual Report on Trend and Progress of Banking in India and its Statistical Tables Relating to Banks in India hold the system GNPA and provisioning history behind the credit cycle. RBI's Basel III Capital Regulations Master Direction and Master Circular on IRAC norms cover every threshold quoted above. And screener.in has the ten-year ratio spine that seeds the scatter.

For the US, SEC EDGAR XBRL companyfacts gives the ten-year P&L and balance-sheet history. For peer percentiles no Indian source matches, use the FFIEC Call Report series (forms FFIEC 031/041) and the UBPR. System aggregates come from the FDIC BankFind Suite API and the Quarterly Banking Profile. FRED (DGS10, DGS2, FEDFUNDS) has the rate path behind any AFS/HTM mark, and the Fed's Barr report on SVB backs the autopsy in worked example 3.

# The sector gate: an assertion, not a comment. Recomputed in-session — R3.
LENDERS = {"bank", "nbfc", "hfc", "insurer"}

def generic_credit_block(f):                     # what a sector-blind engine does
    assert f["sector"] not in LENDERS, (
        f"{f['name']}: interest expense is cost of goods for a {f['sector']}; "
        "EBIT, FCFF, net debt and EV are undefined (§4.1). Route to bank_panel().")
    ebit = f["pbt"] + f["interest_expense"]
    return {"interest_cover": ebit / f["interest_expense"],
            "net_debt_ebitda": (f["debt"] - f["cash"]) / ebit}

meridian = dict(name="Meridian Bank", sector="bank", pbt=6160, interest_expense=13740,
                debt=245000 + 25000, cash=24000)                  # §8, worked example 1
try:
    generic_credit_block(meridian)
except AssertionError as e:
    print("REFUSED:", e)

# What the refusal is worth: run it anyway and read the damage.
ebit = 6160 + 13740                                                # 19,900
print(f"'interest coverage' {ebit/13740:.2f}x      -> a '<2x = thin' rule FIRES")
print(f"'net debt/EBITDA'   {(270000-24000)/ebit:.2f}x     -> a '>4x = distressed' rule FIRES")
print(f"'debt/equity'       {270000/32000:.2f}x      -> a '>2x = levered' rule FIRES")
print("truth: ROA 1.54%, ROE 15.4%, CET1 12.9%, PCR 73% — an excellent bank, scored zero")

Verify. Prove the engine against the worked figures above before it touches a real bank. Meridian's thirteen panel cells should reproduce to the digit: NIM 3.52%, GNPA 2.40%, NNPA 0.66%, PCR 73.0%, credit cost 0.65%, slippage 1.60%, ROA 1.54%, ROE 15.4%, cost-to-income 42.0%, CASA 40.0%, CD 83.3%, CET1 12.9% and CRAR 15.0%. Both roll-forwards should tie (3,796 + 3,000 − 1,500 − 400 = 4,896, and 2,724 + 1,250 − 400 = 3,574). Ravi Bank's NII of ₹2,450 cr and NIM of 3.50% should reproduce too. Problem 10's least-squares fit on the five-bank scatter should come out at slope 0.2538 and intercept −1.4466, which puts the P/B = 1 crossing at ROE 9.64% and predicts 2.106× for bank F. If any of those disagrees, it's the engine that's wrong. The arithmetic here isn't.

Verify, at full precision. Every worked figure on this page, restated unrounded, so the next reader can catch a rounding drift the way GR-214 was caught. Meridian (worked example 1): NIM 3.521429%, GNPA 2.400000%, NNPA 0.659595%, PCR 72.998366% against an opening 71.759747%, credit cost 0.650000%, slippage 1.595745%, ROA 1.540000%, ROE 15.400000%, cost-to-income 41.990669%, CASA 40.000000%, CD 83.265306%, CET1 12.857143%, CRAR 15.000000%, SMA-2 0.882353%, restructured 0.784314%, stressed book 3.184314%. Meridian's ROA tree: 3.286667 + 1.000000 − 1.800000 − 0.433333 = 2.053333 pre-tax, times 0.75 = 1.540000, times a leverage of 10.0 = 15.400000, and the one-stage anchor is 1.300000 times book. Meridian's capital headroom is 485.714 bp and its absorption is ₹10,200.00 cr. Running the same two off the displayed 12.9% returns 490 bp and ₹10,290.00 cr, which is the ₹90 cr of display rounding the capital KPI now flags. Axiom (worked example 2): FY3 PCR 40.620155%, lagged GNPA 5.160000%, required provisions at 70% coverage 3,612.00, shortfall 1,516.00, post-tax hit 1,137.00. Axiom restated: normalised net worth 12,363.00, normalised net NPA 1,548.00, adjusted book 10,815.00, headline P/B 1.200000 and P/ABV 1.497920, with a 4% loss on the ₹62,000 cr vintage at 2,480.00, or 22.93% of adjusted book. SVB (worked example 3): the 2022 policy path is 25 + 50 + 75 + 75 + 75 + 75 + 50 = 425 bp, and the linear duration estimate is −5.7 × 2.0% = −11.400000%, or −$10.374bn on a $91bn book. Sarayu (worked example 4): tree 3.330000 + 0.900000 − 1.800000 − 0.462500 = 1.967500 pre-tax and 1.475625 post-tax, PAT ₹5,902.50, NIM 3.600000%, credit cost 0.711538%, cost-to-income 42.553191%, leverage 11.764706, ROE 17.360294%, PCR 70.000000%, NNPA 0.763359%, CET1 10.909091%, CRAR 13.272727%, absorption ₹8,000.00, retained ₹4,426.875 funding ₹40,579.69 of new risk-weighted assets or 14.756250% growth, justified P/B 1.545000. Cascade (worked example 5, $bn): NIM 3.200000%, efficiency ratio 57.644110%, net charge-offs 0.375000%, non-performing loans 0.600000%, allowance cover 208.333333%, pre-provision net revenue 3.380000, pre-tax 2.830000, net income 2.235700, ROA 1.242056%, ROTCE 14.904667%, plain ROE 13.151176%, CET1 12.692308% for 5.192308 points of headroom, non-interest-bearing deposits 36.000000%, justified P/TBV 1.816667 against an actual 1.800000. Practice 6: risk-weighted assets 74,500, CET1 11.006711%, Tier 1 11.677852%, CRAR 13.959732%, headroom 300.671 bp and 245.973 bp, and absorption ₹2,240.00, where the solution's rounded "≈ 3.0% × 74,500" route gives ₹2,235. Practice 7: year-1 slippages 1,093.12, year-1 GNPA ratio 2.034483% against a lagged 2.950000%, and year-3 GNPA of 4,460 at 3.661741% headline against a lagged 3.844828%. Practice 8: total mark-to-market −7.812, the hidden held-to-maturity share −5.022 or 45.65% of CET1, and a two-day outflow of 15.60 against 18 of liquid assets.

Quantitative lens

Once the panel is a function and the peer set is a frame, four questions raised above become computable, ones you can't settle on paper. The figures in these lenses are all recomputed from the stated figures above. They're properties of those numbers, and they don't measure any listed bank.

A sector-blind engine scores a healthy bank at zero. Take Meridian (ROA 1.54%, ROE 15.4%, CET1 12.9%, PCR 73%, every KPI healthy or better) and point ordinary corporate machinery at it. Interest coverage computes to 1.45×, against the "below 2× is thin" rule every generic screener carries. Net debt to "EBITDA" computes to 12.36×, against an "above 4× is distressed" rule. Debt to equity computes to 8.44×, against "above 2× is over-levered". That's three solvency tests and three failures, on a bank in the top decile of everything the sector's own panel measures. So the gate has to be an assertion. A note in the documentation won't do. A wrong number that looks wrong would be easy to catch. What you get here is three confident, mutually reinforcing wrong numbers that look like a thesis. A generic engine doesn't know it's confused. And that's the case for refusing to compute at all, where the alternative is computing first and caveating afterwards.

Inverted, the cross-sectional P/B line tells you the market's required return, and it isn't your CAPM number. Problem 10's five-bank scatter fits at slope 0.2538 and intercept −1.4466, with R² of 0.990. Now invert it through the identity above. P/B = (ROE − g)/(r − g) rearranges to P/B = ROE × [1/(r − g)] − g/(r − g). So a linear fit of P/B on ROE has slope 1/(r − g) and intercept −g/(r − g). That makes the fitted line a solved pair, and much more than a descriptive curiosity: r − g = 1/0.2538 = 3.94 points, g = 1.4466 × 3.94 = 5.70%, and therefore r = 9.64%. The P/B = 1 crossing is the implied cost of equity, because at ROE = r the identity gives P/B = 1 at any growth rate. Set that against the CAPM build of 13% above and the gap is 3.36 points.

There are two readings, and you have to pick one before you use the line. Either this cross-section is priced off a required return well below a textbook Indian bank CAPM, or the linear form is wrong. It is wrong. A single line assumes every bank in the sample shares one r and one g, and the valuation section argues throughout that duration and franchise quality differ. Against the theoretical line at r = 13%, g = 5%, which has slope 0.125, the fitted line is 2.03× steeper. A fitted P/B-against-ROE line is a pricing rule. Inverting it is how you find out whether you're being handed a pricing rule or a valuation.

The residual is the output, but a book-quality adjustment moves banks further than the residual does. Bank F is at ROE 14% and P/B 1.1× against a predicted 2.106×. That's a residual of −1.006, or 9.4 residual standard deviations on a fit whose residual standard deviation is only 0.107. It looks decisive. It isn't, and the second number is the reason. Across the five specimens restated above (Axiom, Godavari, Chambal, Damodar and Kaveri), the move from headline P/B to P/ABV averages ×1.209 and ranges from ×1.133 to ×1.267. On a 1.30× bank that average correction is +0.272 of P/B, which is 2.5 residual standard deviations of the fit. So the adjustment the screen doesn't make is worth more than twice the noise the screen does model. And the correction isn't uniform. A 13-point spread in it across five banks re-ranks the sample. A P/B-against-ROE screen therefore ranks banks on a number that one mandatory, publicly computable adjustment moves by 13 to 27 percent, and unevenly. Run the fit on P/ABV or don't run it.

Leverage means the credit line is the valuation. Meridian's assets are 10.0× its equity. Across the five fully worked specimens above (Meridian, Sarayu, Kaveri, Tungabhadra, Cascade Republic), leverage averages 11.75×, so dROE/dROA is 11.75 and 10 bp of ROA is 117 bp of ROE. Now push that through the one line where a bank's estimate lives. One basis point of credit cost on Meridian's ₹2,00,000 cr of average advances is ₹20 cr pre-tax and ₹15 cr after tax, on ₹30,000 cr of equity. That's 5.0 bp of ROE per 1 bp of credit cost. So walking from the "normal" 0.65% to the "cycle turn" threshold of 1.5% costs 425 bp of ROE. That's 85 basis points, one bad year, with no change in underwriting policy. ROE goes from 15.4% to 11.15%, below the 13% cost of equity. And on the one-stage identity with g = 5%, justified P/B falls from 1.30× to 0.77×. That's a 41% fall in justified value from an 85 bp move in one estimated line item. It's the arithmetic behind the whole credit-cycle argument. You aren't being asked to forecast a bank's revenue. You're being asked to forecast the one number management gets to estimate, and it's levered five to one into ROE and eleven to one into the multiple.

NIM also decomposes cleanly, which ends the "why isn't NIM the loan spread?" confusion for good. For any lender, NIM = (Y − C) + C × (free funds ÷ earning assets), where Y is the yield on earning assets and C the cost of interest-bearing liabilities. It's the spread plus a credit for the slice of assets that equity funds at zero. For Ravi Bank (Problem 1), Y = 8.60% and C = 5.4923%, so the spread = 3.108% and the free-funds benefit = 5.4923% × (5,000/70,000) = 0.392%. Those sum to the stated 3.50% with nothing left over. Starting from the loan spread, advances add 3.900, the asset-mix drag from low-yield SLR investments takes off 0.600, the funding-mix drag from borrowings costing more than deposits takes off 0.179, and free funds at the deposit rate add back 0.379, for a total of 3.500. Every NIM move a bank reports is one of those four terms. A decomposition that doesn't sum to the reported NIM has a missing term, usually the free-funds credit. Analysts who leave it out end up crediting capital-raise effects to "margin improvement", again and again.

Now the limits, said plainly, because a fitted line is the easiest thing here to over-trust. The P/B-against-ROE regression is estimated on five points with three degrees of freedom. Its R² of 0.990 is a property of five hand-picked numbers in a practice problem. It tells you nothing about Indian banking. As for the "9.4 standard deviations" for bank F, that's arithmetic on a residual standard deviation that itself carries almost no information. Run the fit on a real peer set of ten to fifteen banks and expect a far lower R² and a far wider residual band. The method transfers, but the numbers don't. Inverting to r = 9.64% and g = 5.70% holds only if a single (r, g) pair describes every bank in the sample, and the valuation section tells you to distrust that assumption. Treat the inversion as a sign that the linear form is being asked to do too much. It doesn't measure the sector's cost of equity.

The leverage and credit-cost sensitivities hold ROA's other branches fixed, and no real cycle does that. A credit-cost year is usually also a NIM year and a growth year, and that interaction isn't modelled here. Those ABV correction statistics come from five synthetic specimens built to teach the adjustment, so their mean of ×1.209 describes that teaching set and nothing else. Build your own distribution from real disclosures before you quote a number. And the sector gate is itself a classification problem wearing an assertion's clothes. It works only as well as the label on the entity. A diversified conglomerate with a captive finance arm, an insurer with a large investment book, or a fintech that's a lender in everything but its licence will each defeat a single sector field. A gate like this isn't there to be right about every entity. It's there to make you supply the classification consciously, and to stop the engine when you haven't.

Do it in code: put the sector assertion above every EV-based and coverage metric. Make the roll-forward identity the first line of the lender branch, and raise an error on a book that doesn't tie. Keep thresholds in a market-and-segment table, never as literals. Fit P/B against ROE on ABV instead of book, and report residuals with their standard error. And run the credit-cost normalisation as a range before you quote a justified multiple, never as a point. See G2/CS2.06, G1/QM1.02 and G2/DA1.02.

Where this goes next: galaxy cross-links

• G2 · CS2.06 Software Engineering & AI-Assisted Development: the sector gate as an assertion the engine can't run without, the roll-forward identity as a test that fails on a bad extraction, and thresholds as configuration instead of magic numbers. This banking playbook writes the specification, and that course supplies the engineering habits.
• G1 · QM1.02 Hypothesis Testing & Regression: how to fit the P/B-against-ROE line properly, what a standard error on the slope means, why R² on five points is not evidence, and how to read a residual as a hypothesis and not yet a verdict.
• G2 · DA1.02 Pandas Data Wrangling: assembling the multi-quarter, multi-bank panel from investor presentations, Pillar 3 disclosures, EDGAR companyfacts and FFIEC call reports into one frame with consistent definitions. That's nine-tenths of the work in the bank-workup mini-project.
• G1 · FI1.02 Duration & Convexity: the SVB arithmetic done in full, without the linear approximation of worked example 3. It covers why convexity and portfolio runoff damp the −11.4% estimate, and what a 200 bp shock really does to an AFS/HTM book.
• G1 · MS1.03 Central Bank Operations & Repo: CRR, SLR, LCR and the lender of last resort as monetary plumbing, which is a long way from compliance trivia. It's the layer under the regulatory map here and under Diamond-Dybvig's patches.

Teach it back & journal

Feynman prompt. Write one page for a smart 15-year-old, titled "How a bank can report record profits and be dead two years later." It has to do three things. Use a lemonade-stand-scale example of 10× leverage, with ₹10 of your money and ₹90 borrowed. Explain why the cost of a bad loan shows up years after the loan is made. And say what a line of people outside a branch (or a $42bn app-tap day) actually means. No jargon is allowed except the three terms you define yourself: NPA, provision, capital.

Journal reflection. Look at the six red flags. Which one would you be most likely to rationalise away in a bull market, and what would you tell yourself? Be straight with yourself here, because "growth denominators" fooled most professionals in FY07 and FY19 alike. Write the pre-commitment rule that would stop you, for example "any bank growing >25% gets valued on lagged asset quality, no exceptions", and add it to your personal checklist thread. Then note one thing you learned about your own money-and-safety instincts from imagining yourself as an uninsured SVB depositor on 9 March 2023.

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