Imported from JiayiV/question_mixing (skills/public-02-claude-skills-main-finance-skills-stock-analysis-references-sectors-nbfc-7b88991a/SKILL.md). Install upstream with npx skills add JiayiV/question_mixing --skill public-02-claude-skills-main-finance-skills-stock-analysis-references-sectors-nbfc-7b88991a. Copyright stays with the author.

NBFCs, housing finance and non-bank lenders — sector playbook

Use this when: the company earns most of its income by lending its own balance sheet — NBFCs, HFCs, gold-loan and microfinance lenders, vehicle and CV financiers, LAP/SME lenders, consumer and fintech lenders, US/EU specialty finance, consumer and mortgage originators, BDCs and thrifts.

A lender's balance sheet is its product line. Money is the raw material, interest expense is cost of goods, and the loan book is inventory that can silently rot for 12–24 months before it shows up in any reported ratio. That single fact breaks the generic industrial ratio set at the definitional level — not at the "different benchmark" level — and it means the two things that actually kill lenders (funding and credit) are both largely invisible in the P&L until they are terminal. Analyse this sector as: margin (NIM) minus credit cost, on a funding base that must not run, on capital that must fund growth.

All ranges below are indicative only. They shift with the rate cycle, product mix, regulatory regime and country. A lender's own 5–10 year history and its closest sub-sector peers override every absolute band in this file.

Contents

• Why the generic ratio set fails here
• The metrics that actually matter
• How to value companies in this sector
• Peer set construction
• Sector-specific red flags
• Cycle and structural context
• India vs global notes
• Checklist

Why the generic ratio set fails here

Do not compute the following for a lender. If a screener or data provider hands them to you, discard them explicitly and say why.

OPM / EBITDA margin / EV-EBITDA / EV-Sales — undefined in economic substance. EBITDA adds back interest expense. For a lender, interest expense is the cost of raw material: money bought wholesale, sold retail. Adding it back produces a number with no meaning, and any "operating margin" computed on gross interest income is an artefact of leverage rather than efficiency. Equally, Enterprise Value is undefined: debt is operating float, not a claim to be netted against equity value. Never compute EV, net debt or EV/Sales for a lender. The correct analogue of gross margin is Net Interest Margin; of operating margin, pre-provision operating profit / assets.

Debt/Equity — a regulatory input, not a risk signal, and often inverted. NBFCs typically run 4–7x, HFCs 7–11x, banks higher still. A D/E screen mechanically rejects the entire sector. Worse, the sign is unreliable: 9x leverage on a well-provisioned prime mortgage book can be far safer than 3x on an unsecured book with no collateral and 100% LGD. The real solvency measures are risk-weighted and regulator-defined — CRAR, Tier 1, CET1, plus a simple leverage backstop where the regulator imposes one.

Free cash flow and DCF-on-FCFF — structurally negative and sign-inverted. Under Ind AS 109 / IFRS 9 presentation, loan disbursements sit in operating cash flow. A fast-growing, perfectly healthy lender therefore reports hugely negative CFO and negative FCF every single year, while a shrinking, dying lender reports strongly positive FCF as the book runs off. Screening for positive FCF in this sector selects for terminal decline. Equity value must be built from FCFE, dividend discount, or residual-income/excess-return models, where "investment" is the regulatory capital consumed by growth.

ROCE — a trap, especially on Indian screeners. Screener-style ROCE = EBIT / (equity + borrowings) collapses, for a lender, to approximately the gross yield on assets. A gold-loan or microfinance NBFC yielding 20% shows a spectacular "ROCE" that says nothing about profitability after funding cost, opex and credit cost. It systematically ranks the highest-risk lenders highest. Use ROA and DuPont-decomposed ROE only.

Working-capital metrics — no counterparties exist. Current ratio, cash conversion cycle, inventory turnover, debtor days, asset turnover and capex/sales have no meaning on a lender's balance sheet. Interest coverage is also meaningless — interest is revenue-side, not a fixed charge to be covered. What replaces working capital entirely is the ALM maturity bucket table.

"Sales" / revenue growth — the wrong top line. Reported revenue is gross interest income, which rises with both book size and portfolio risk. A lender can double revenue by migrating from 9% prime home loans to 24% unsecured personal loans while destroying value. Use NII, NIM, and risk-adjusted NIM (NIM minus credit cost) as the real revenue lines.

P/E — defined, but pro-cyclical and dangerous. Reported PAT for a lender is a policy variable: it is whatever remains after a discretionary provisioning decision (ECL staging, PD/LGD assumptions, management overlays). Credit cost is near zero at the top of a cycle, so lenders look statistically cheapest on trailing P/E precisely when the loss cycle is about to turn. Trailing-P/E screens systematically buy the top. Anchor on book value and sustainable ROE instead.

Depreciation, capex intensity, gross block, fixed-asset turnover — irrelevant. This is an asset-light but capital-consuming business. The true "capex" is regulatory capital absorbed per unit of AUM growth, and it never appears in the capex line.

The metrics that actually matter

Ranges are indicative and product-specific; judge intra-segment and against the company's own history.

How to value companies in this sector

Primary method — Price to Adjusted Book, anchored to sustainable ROE.

The canonical relationship is P/B = (ROE − g) / (COE − g). Book value here is the productive asset base — unlike an industrial, where book is sunk cost — and returns are earned directly on it, so the multiple is a function of how far sustainable ROE exceeds cost of equity. A lender earning 12% ROE against a 13% COE should trade below 1x book no matter how fast it grows, because growth at sub-COE returns destroys value and forces dilutive issuance. A 22% ROE franchise with a genuine low-cost funding moat can rationally sustain 4–6x book.

Use adjusted book (P/ABV), not reported. Deduct: net NPAs not covered by provisions; the uncovered portion of restructured and ARC-sold exposures (including security receipts carried at inflated values); goodwill and intangibles; deferred tax assets that depend on future profits materialising; capitalised sourcing costs. In DM this is framed as Price/Tangible Book Value paired with ROTE — the standard for US and European consumer and mortgage lenders.

Indicative 1-year-forward P/ABV bands (India): high-ROE, high-growth, low-credit-cost franchises 3.5–6x; solid secured retail lenders 2–3x; average diversified NBFCs 1.5–2.5x; prime HFCs 1.5–3x; gold-loan lenders 2–3x; sub-scale, high-credit-cost or governance-impaired names below 1x. DM specialty consumer and near-prime auto typically 0.8–1.6x TBV; prime mortgage originators often below book because ROTE sits near COE.

Secondary methods.

• P/E on normalised credit cost. Never use trailing P/E raw. Rebuild EPS with a through-cycle credit-cost assumption drawn from at least one full downturn, then apply a multiple. This corrects the classic error of buying at 8x peak-cycle earnings just before losses normalise. Indicative Indian bands: 12–18x for average quality, 25–40x for compounders with cycle-tested underwriting.
• Residual income / excess return. Theoretically cleanest: value = current book + PV of (ROE − COE) × book, projected with explicit capital consumption for growth. It sidesteps the FCFF problem entirely and makes the ROE-vs-COE spread the explicit driver.
• FCFE / dividend discount. FCFE = earnings − equity capital absorbed by RWA growth. This is the correct DCF for a lender. DDM works well for mature, low-growth, high-payout DM lenders.
• P/AUM (or EV/AUM in DM deal contexts). Cross-check and the standard M&A benchmark, especially for microfinance, gold-loan and affordable-housing platforms where the franchise is the distribution network. Indian precedent transactions have spanned roughly 1.5–3.5x book or 20–45% of AUM depending on ROA. Useful for loss-making or early-stage lenders where P/E is undefined, but it ignores asset quality completely — never standalone.
• Sum-of-the-parts for holding structures. Many Indian NBFC groups hold stakes in AMCs, insurance, broking or housing subsidiaries valued on entirely different conventions (AUM multiples for AMCs, embedded-value multiples for life insurance). Value each on its own basis and apply a holding-company discount — historically 20–50% in India.

Do not use: DCF on FCFF, EV/EBITDA, EV/Sales, EV/EBIT, ROCE-based screens, or any multiple with EV in the numerator. Debt is an operating input, so EV is undefined; and CFO is structurally negative for a growing lender. Any EV/EBITDA quoted for an NBFC is a data-provider artefact — say so and discard it.

Sensitivity discipline. At 6–10x leverage, a 25bp move in NIM or a 50bp move in credit cost swings ROE by 200–400bps, which can move justified P/B by a full turn. Always present valuation as a NIM × credit-cost grid, never a point estimate. State the COE you assumed (India: 12–14% is the usual working range; DM lenders 9–11%) because the whole framework hinges on it.

Peer set construction

A valid comparable in this sector shares asset class, funding profile and regulatory regime — not merely the "NBFC" label. Get this wrong and every conclusion inverts, because yield, credit cost, opex and leverage all differ by 3–5x across sub-sectors.

Do not mix these:

• Prime housing finance (2.5–3.5% NIM, 10–40bps credit cost, 8–11x leverage, 15–25% C/I) with affordable housing (5–7% NIM, higher opex, different customer). They are different businesses sharing a regulator.
• Secured retail (vehicle, gold, LAP, mortgage) with unsecured consumer / personal / digital lending. LGD differs by an order of magnitude, so identical GNPA means completely different capital consumption.
• Retail granular books with wholesale / developer / infrastructure lenders. Retail fails gradually and statistically; wholesale fails in discrete, lumpy, correlated events. Loss distributions are not comparable and neither are the valuation multiples they deserve.
• Microfinance with anything else. Unsecured, joint-liability, politically exposed, geographically clustered, with regulatory rate and multiple-lending caps. It has its own cycle.
• Gold loans with other secured lenders. Short tenor, liquid collateral, auction-driven recovery, LTV capped by regulation, and an embedded gold-price sensitivity that no other product carries.
• Deposit-taking (NBFC-D / thrifts / banks) with wholesale-funded NBFCs. Deposit franchises have structurally cheaper, stickier funding and heavier regulation. Comparing their cost of funds and their multiples is meaningless.
• Captive / OEM-linked financiers with independent lenders — their sourcing cost, credit selection and growth are governed by the parent's product cycle.
• Balance-sheet lenders with originate-to-distribute / fee-based platforms and fintech marketplaces. The latter carry little credit risk and should be valued on earnings or revenue multiples, not book. Be alert to hybrids that claim to be asset-light while retaining FLDG.

Also align: stage of growth (a 40%-growing lender consuming capital vs a mature payer), country and rate cycle, accounting regime (Ind AS/IFRS 9 ECL vs older incurred-loss regimes vs US CECL), and — in India — listed vs unlisted and bank-promoted vs standalone, since parentage materially changes cost of funds.

Sector-specific red flags

• Growth far above system — AUM compounding 40–50%+ for multiple years, especially in a new product or geography. Rapid growth mechanically suppresses GNPA through denominator inflation and defers loss recognition 12–24 months. Cross-check with vintage curves: if each successive cohort shows worse 12-month-on-book losses while headline GNPA falls, the ratio is lying.
• Evergreening and disguised restructuring — top-up loans to borrowers approaching delinquency, refinancing a borrower to clear their own arrears, LAP taken to repay another loan, or interest capitalised into principal on developer exposures so a non-paying loan never stamps a DPD. Tell-tales: a large "interest accrued but not due" balance, non-cash income rising as a share of interest income, bullet/balloon structures on real-estate loans.
• Related-party, promoter-group or shell-entity lending — inter-corporate deposits, unsecured loans to entities with no operating business, exposures to the promoter's other ventures. This was the mechanism in both the IL&FS and DHFL failures. Read the related-party note and the largest-exposures list, not the press release.
• Aggressive write-off policy masking asset quality — write-offs above 30–40% of opening GNPA, or falling GNPA ratio alongside rising absolute write-offs. Recompute GNPA adding back three years of cumulative write-offs. Likewise, ARC sales where consideration is mostly security receipts retained on book simply defer recognition.
• Upfront income via direct assignment / securitisation — booking the entire future excess interest spread as a gain on derecognition. Front-loads years of margin into one quarter and, once started, must be repeated at increasing scale. Above 10–15% of PBT, restate earnings without it before valuing.
• ECL model discretion — sudden cuts to PD/LGD assumptions, release of COVID-era or other management overlays into profit, unexplained Stage 2 → Stage 1 migration, or PCR falling while Stage 2 rises. Visible in the ECL note of the annual report, not the investor deck.
• Funding fragility — CP above ~15% of borrowings, negative cumulative ALM gap in sub-one-year buckets, dependence on a single bank or instrument, or rollover reliance to fund long-tenor assets. Watch the traded spread on the company's own NCDs versus similarly rated peers: the bond market prices distress before the equity market does.
• Rating downgrade, negative outlook, or promoter share pledge — here a downgrade is not a lagging indicator, it is a causal event that raises funding cost and can shut market access outright. High promoter pledge adds forced-selling risk that feeds back into funding confidence.
• Governance and reporting signals — auditor resignation or qualification, repeated CFO or CRO departures, delayed or restated results, whistle-blower complaints, or a widening gap between standalone and consolidated performance. In lending, governance failure and credit failure are the same event.
• Concentration and single-event exposure — one state above ~30% of AUM, wholesale top-20 borrowers above 15–20% of net worth, or a single-product model with no counter-cyclical ballast.
• Off-balance-sheet risk retention — co-lending or partnership structures where the NBFC provides first-loss default guarantees (FLDG/DLG) or holds the junior tranche of its own securitisations. AUM growth looks capital-light while economic risk is fully retained. RBI caps DLG at 5% of the portfolio in digital-lending arrangements; check the disclosed retained first loss.
• Collection-efficiency definition games — including recoveries of prior-period arrears in the numerator, which can print above 100% while the current book deteriorates. Insist on current-month-billing-only efficiency and cross-check against bounce rates and 30+ DPD.
• Risk migration hidden inside stable ratios — rising average ticket size, rising LTV, rising self-employed or new-to-credit share, lengthening tenors to keep EMIs affordable, or a shift toward balance-transfer and top-up volume. Each raises loss frequency or severity while leaving current-period ratios untouched.
• Borrowing or deposit rates materially above peers — a lender paying up for retail deposits or wholesale funds is either being priced for risk by the market or funding an asset book whose yield cannot be sustained. A classic late-stage signal in both Indian NBFC-D and DM thrift/specialty failures.
• Regulatory overhang and supervisory action — scale-based-regulation reclassification, risk-weight increases (e.g. the November 2023 move to 125% on unsecured consumer credit and on bank lending to NBFCs), gold-loan LTV and auction-norm tightening, digital-lending guidelines, or business-restriction orders against a specific entity. Regulation changes unit economics overnight and is a first-order valuation input, not a footnote.
• Low effective tax rate or building deferred tax assets — often loss carry-forwards or aggressive provisioning timing differences. Reduce adjusted book by DTAs that depend on future profits materialising.

Cycle and structural context

Where you are in the credit cycle dominates everything else. The cycle runs: cheap and abundant funding → competition for growth → underwriting standards loosen → yields compress or lenders migrate to riskier products → a rate or liquidity shock → funding cost spikes and rollover tightens → losses surface 12–24 months after origination → capital erodes → forced deleveraging and consolidation. Reported earnings look best just before the turn, because credit cost is at its trough and growth at its peak. Ask explicitly: is this book seasoned, and did this management team run this book through a full downturn?

Rate cycle mechanics. Asset repricing lags liability repricing in most Indian NBFC structures (fixed-rate vehicle, gold, MFI and personal loans funded by floating or short-tenor borrowings), so rising rates compress NIM first and only later flow into loan pricing. Falling rates do the reverse and can flatter NIM for 2–3 quarters. Separately, floating-rate mortgage books repriced to an external benchmark (India: repo-linked) transmit rate changes faster on the asset side than fixed-book lenders. Model the repricing gap, not just the direction of rates.

Liquidity events, not slow decay, are how lenders die. The 2018 IL&FS default froze the Indian NBFC wholesale funding market and killed institutions that were reporting healthy capital days earlier. Northern Rock and Greensill are the DM analogues. Any lender whose survival depends on continuous access to wholesale markets carries an option written against a tail event that is not in its cost of funds.

Structural and competitive threats. Bank competition in prime segments compresses HFC and vehicle-finance spreads structurally — banks fund cheaper and can always take the prime customer. Balance-transfer/refinance activity in mortgages erodes portfolio yield and shortens effective duration. Digital and fintech origination has collapsed customer acquisition cost in unsecured lending but also compressed underwriting time and made adverse selection faster. Account aggregators, credit bureaus and UPI-linked data are steadily commoditising the informational edge that used to justify high yields. Ask what the durable advantage is: cost of funds, distribution reach, collections infrastructure, or proprietary underwriting data — and whether it survives digitisation.

Regulatory direction. In India, RBI's scale-based regulation (Base/Middle/Upper/Top layers) progressively imposes bank-like norms — LCR, CRAR, board composition, NPA recognition, listing requirements for upper-layer entities — on larger NBFCs. Direction of travel is convergence toward bank regulation, which raises compliance cost and lowers steady-state ROE for the largest players. Globally, regulatory arbitrage between banks and non-banks is the sector's founding rationale and it periodically closes: assume any regulatory gap that a lender's economics depend on will eventually narrow.

Counter-cyclical positioning. The best entry points historically follow forced deleveraging, when survivors with capital and funding access buy books cheaply and market share consolidates. The worst are at cycle peaks when trailing P/E looks lowest. Let the credit-cost gap versus through-cycle norms, not the P/E, tell you where you are.

India vs global notes

India (NSE/BSE, Ind AS).

• Regulators split by entity: RBI for NBFCs and (since 2019) HFCs, with NHB retaining supervision and refinance functions for housing finance. Deposit-taking entities (NBFC-D) face tighter rules than NBFC-ND.
• Scale-based regulation classifies NBFCs into Base, Middle, Upper and Top layers, with rising capital, governance, disclosure and listing obligations. Check which layer the company sits in — it determines the regulatory trajectory.
• RBI November 2021 daily-stamping circular: NPA upgrade only after full clearance of all arrears. GNPA reported before and after this change is not comparable; do not draw trend conclusions across that boundary.
• Ind AS 109 ECL with three-stage classification. RBI additionally requires disclosure of the gap between Ind AS provisions and IRACP norms; if IRACP exceeds ECL, the shortfall goes to an impairment reserve. Read that reconciliation — it is a direct check on ECL aggressiveness.
• Numbers in crore/lakh; AUM often quoted including off-book. Always confirm whether AUM is on-book, on+off-book, or includes co-lending partner share.
• Promoter holding and pledge are first-order: check pledge percentage in the shareholding pattern. Bank- or corporate-promoted NBFCs enjoy materially lower cost of funds than standalone peers.
• CARO reporting includes specific clauses on loans granted, related-party transactions, defaults in repayment of borrowings, and (for NBFCs) whether the entity conducted registered activities without a valid RBI registration. Read the CARO annexure and the auditor's key audit matters — ECL is almost always a KAM for a lender.
• Concalls and investor decks disclose collection efficiency, bounce rates, Stage 2, product-wise AUM and cost of funds that are not in the financials. Use them, but restate collection efficiency to a current-month-billing basis.
• Priority sector lending creates a structural market: banks buy PSL-eligible portfolios (MFI, small-ticket housing, agri) via securitisation/assignment, which is why assignment income is so prevalent in Indian NBFC P&Ls.
• Rating agencies (CRISIL, ICRA, CARE, India Ratings) publish detailed rationales with ALM and liquidity commentary — often the most informative public document on a mid-size NBFC.

US / global (10-K, EDGAR, GAAP/IFRS).

• US GAAP uses CECL (lifetime expected loss on day one) rather than IFRS 9's three-stage model, so allowance levels and the timing of provisioning differ materially. Do not compare coverage ratios across the two regimes without adjustment.
• Terminology maps: GNPA → non-performing / non-accrual loans; credit cost → provision for credit losses; write-offs → net charge-offs (NCOs); restructured → TDRs / modified loans; PCR → allowance coverage; ROE → ROTE; P/ABV → P/TBV.
• Delinquency and net charge-off disclosure is far richer in the US: 10-Ks and 10-Qs give delinquency buckets, NCO rates, and often vintage tables (ASU 2016-13 requires gross write-offs by origination year) — the vintage analysis you must reconstruct manually in India is frequently disclosed directly.
• Funding conventions differ: warehouse lines, ABS term deals, revolving conduits and deposit funding replace CP/NCD/bank-term-loan mixes. Watch advance rates and covenants on warehouse facilities — a covenant breach can be as fatal as a downgrade in India.
• Capital rules: Basel III CET1 stacks for bank-like lenders; BDCs are governed by a simple 2:1 asset-coverage leverage cap rather than risk weights; mortgage REITs are valued on book and dividend, with entirely different accounting.
• Prime US mortgage lenders often trade below book because ROTE sits near COE — that is rational, not a bargain. Judge it by the ROE-vs-COE spread, not the absolute multiple.
• Consumer-protection regulators (CFPB in the US, FCA in the UK) can impose product-level pricing and collection restrictions that change unit economics as abruptly as RBI action does in India.

Checklist

• Reject EV/EBITDA, EV/Sales, ROCE, FCF, current ratio and D/E outright; state that they are undefined or inverted for a lender.
• Compute NIM, spread and risk-adjusted NIM (NIM − credit cost); decompose any NIM change into yield vs cost of funds.
• Build the DuPont bridge from ROA to ROE and identify which driver moved and whether it is durable.
• Compare incremental cost of funds with the blended average; check CP share, single-source concentration, rating, and rating outlook.
• Read the ALM bucket table; flag any negative cumulative gap inside one year and check the liquidity buffer against 3–6 months of debt servicing.
• Check CRAR/Tier 1 (or CET1) against the minimum, then compute the growth runway to the next equity raise and the P/B at which prior raises were done.
• Reconstruct the true loss rate: GNPA + 3-year cumulative write-offs + ARC/SR exposure + restructured pool; compare to headline GNPA.
• Check PCR against product type and compute NNPA as a % of net worth.
• Pull Stage 2 assets, bounce rates, collection efficiency (current-month billing only) and 30+ DPD roll rates as the leading indicators.
• Build or demand vintage/static-pool loss curves by origination cohort; never judge a fast-growing book on headline GNPA.
• Separate on-book from off-book AUM and identify who holds first loss (FLDG/DLG, junior tranches, co-lending share).
• Strip upfront assignment/derecognition gains and fair-value gains from PBT; recompute ROA and ROE without them.
• Test for risk migration: ticket size, LTV, tenor, self-employed and new-to-credit share, top-up and balance-transfer volume.
• Check concentration by state, product, top-20 borrowers and funding counterparty.
• Read the related-party note, the ECL note, CARO/KAM (India) or the credit-quality and vintage tables (10-K), not just the investor deck.
• Value on P/adjusted book vs sustainable ROE using P/B = (ROE − g)/(COE − g); cross-check with residual income and normalised-credit-cost P/E; use P/AUM only as a sanity check.
• Present the valuation as a NIM × credit-cost sensitivity grid and state the assumed cost of equity.
• Locate the position in the credit cycle and ask whether this management has been tested through a full downturn.
• Confirm the peer set shares asset class, funding profile, regulatory regime and accounting basis; never blend HFC, MFI, gold, wholesale and unsecured lenders in one table.