Types of NBFC

Five businesses share a funding model and almost nothing else, plus a few hybrids. Each gets one paragraph: the economics, the normal numbers, the characteristic death. The table comes first. All its figures are illustrative through-cycle ranges in % of AUM unless stated, so calibrate them to current filings:

How does housing finance work?

Housing finance. This one's a strange beast. It attaches the safest asset in Indian lending to the most dangerous balance sheet. Individual prime mortgages are so safe that borrowers die before they default, and through-cycle losses run 20 to 50 bp. But they're the longest assets in the sector, so they produce the widest structural ALM gap and the greatest sensitivity to funding freezes. Its spread is brutally thin. You lend at 9.5% against banks lending at 8.5%, because mortgages are their favorite asset too, and you borrow at 7.5 to 8%. This model only produces a respectable ROE with 8 to 10× gearing and sub-0.7% opex. It's a scale-and-funding-cost game, where a 30 bp CoF disadvantage erases a third of the ROE. The growth economics are glacial and beautiful: 15 to 20 year relationships, near-zero credit cost, NHB refinance as a stabilizer. HDFC Ltd compounded this model for 45 years into India's most admired financial. Study its pre-merger annual reports, because that disclosure standard is still the sector's textbook.

HFCs go wrong in two ways. One is the funding side, which is everything above. The other is yield-stretching. When the individual-mortgage spread feels too thin, HFCs drift into developer construction finance and loan-against-property at 12 to 15%. Those are a different business, lumpy and cyclical with 30 to 50% LGD, wearing the mortgage book's leverage. DHFL was the terminal case, and every HFC deck since then discloses its "individual : developer mix" in the first five slides. Watch the spread, where 1.8 to 2.5% is the game, and the individual-book share. Watch prepayment and balance-transfer rates too. RBI bars foreclosure charges on floating-rate individual loans, so mortgages reprice with a click when rates fall. And watch gearing against the actual asset mix. For context, India's mortgage/GDP is only ~11 to 12%, against 50%+ in the US. The demand runway is real, so the segment keeps attracting more capital than its spread can feed.

How does vehicle finance work?

Vehicle finance. This is lending against depreciating, mobile, income-producing collateral (trucks, tractors, two-wheelers, used cars) to borrowers whose income is the vehicle itself. The craft is asset knowledge, meaning resale values by model-year-district, plus a collections culture of field officers who know the borrower's route. Those two skills are the moat of the forty-year franchises: Shriram in used CVs, Sundaram, Chola, M&M Financial in rural markets. Yields run 13 to 18% (used > new, CV > car). Opex is heavy at 2.5 to 3.5%, because collections are labor. Credit cost moves with rural cash flows: monsoons, freight rates, infrastructure spending.

GNPA runs structurally higher, at 3 to 6%, and it means less. Delinquency here is a rhythm. Don't read it as a verdict. The borrower misses monsoon-month EMIs and catches up at harvest, and ultimate losses after repossession run well below peak GNPA. So for vehicle lenders read credit cost and LGD, and look past the GNPA optics. Watch the repossession market as well. In a real downturn everyone repossesses at once, and used-vehicle prices, the collateral floor, fall just when you need them most, as in 2019's CV cycle. Its ALM is blessedly boring, with 3 to 4 year assets against 2 to 4 year liabilities and gaps near zero. Vehicle financiers sailed through 2018 on the liability side while suffering on the asset side. The segment's cycle test is FY19-21. The leaders' credit cost peaked at ~3% and was absorbed within the NIM. That's what "priced for the cycle" looks like.

How do gold loan NBFCs work?

Gold loans. Here's the best balance sheet in Indian lending, wrapped in headline risk. The product is ₹40,000 against pledged ornaments, disbursed in minutes, on a 6 to 12 month tenor. RBI caps the LTV, tiered under the 2025 gold-lending directions from up to ~85% for the smallest tickets down to 75%. Yields run 17 to 22%. Think of it as convenience pricing. It isn't desperation pricing, because the alternative is the pawnbroker at 30%+. Opex is heavy (branch density, vault security, assayers), while credit cost is near zero because the collateral is liquid, portable, and auctionable. A defaulted gold loan is an auction with a small recovery lag. Nobody has to run a workout. GNPA can spike optically, since borrowers rationally let marginal loans slide when gold prices soften, and it barely matters. Watch LGD instead of GNPA, the reverse of every other segment.

Its ALM is the sector's dream. Assets of 4 to 6 months against liabilities of 1 to 3 years produce positive short-bucket gaps. Muthoot and Manappuram barely noticed September 2018. Practice 9's sign-flip drill shows you why. What can hurt a gold lender comes from outside. A deep gold-price crash can compress LTV cushions, as in the 2013 episode. Regulators strike with LTV caps and cash-disbursement limits. And banks compete with agri-classified gold loans at 9%, periodically invading and then retreating. Muthoot holds collateral approaching 200 tonnes of household gold, an illustrative ~FY25 scale, so pull the filing. The franchise is trust plus branch adjacency. It earns 4.5 to 5.5% ROA at modest gearing, arithmetic most banks would kill for, and it's valued warily because every few years a regulator or the gold price coughs.

How does microfinance work?

Microfinance. These are uncollateralized ~₹30-60,000 loans to low-income borrowers (mostly rural, mostly women), underwritten by the joint-liability group, where social collateral replaces physical. RBI's 2022 framework sets the rules. Qualifying microfinance must be ≥75% of assets for NBFC-MFIs. There are household income ceilings, and total repayment obligations are capped at ~50% of household income. Pricing was deregulated from the old margin caps into board policies under supervisory watch. Verify the current rules, because this framework moves. In a normal year the economics are lovely: 21 to 26% yields, 10 to 13% NIM, 5 to 6% opex (weekly village meetings are expensive), 2 to 3% credit cost, 3.5 to 5% ROA.

The catch is the tail. Microfinance losses are correlated and event-driven. Think of a state ordinance (Andhra 2010, the founding trauma), demonetization (2016) and the COVID waves. Think too of borrower-stress cycles amplified by several lenders stacking loans on the same household, as in the 2024-25 guardrails episode. In tail years credit cost hits 6 to 10%, and the sector's equity absorbs it along with the earnings. CreditAccess Grameen, the segment's best operator, saw ROA swing from ~5%+ to low single digits in the FY25 cycle. Those figures are illustrative, so pull the filings. Price MFIs on through-cycle ROE. Four good years and one bad one average to ~14 to 16%, well short of the 20% a good year annualizes to (Practice 10). And treat geographic concentration as the primary risk metric. A Karnataka-heavy book is a Karnataka bet, even when it's presented as a diversified book with a Karnataka flavor.

How does unsecured consumer lending work?

Unsecured consumer. Personal loans, consumer durables, credit-card-adjacent credit. This is the segment technology transformed. Bureau data did much of it once CIBIL coverage exploded after 2010. So did account-aggregator cash-flow underwriting, UPI-era transaction trails, and point-of-sale distribution. Bajaj Finance is the archetype, and one of the world's great lending franchises. It turned the zero-cost-EMI checkout counter into a customer-acquisition machine and cross-sold ~100 million customers into an ecosystem, with AUM around ₹4 lakh crore. That's an illustrative ~FY25 scale, so pull the filing. It has consistently earned ~4%+ ROA and ~20% ROE with a liability stack deliberately built as the opposite of DHFL's: deposits, long NCDs, minimal CP, LCR held far above floor. The segment's economics are 16 to 24% yields, credit cost of 1.5 to 3% for prime books and ROA of 3 to 4.5%. That's the highest-return lending model in India, when underwriting holds.

Two caveats keep you employed. First, the through-cycle number is unobserved. India's unsecured-retail boom has never met a real unemployment cycle at today's scale. The FY24-25 stress in small-ticket personal loans (fintech-sourced, sub-₹50,000), with credit costs in that sliver running 6 to 10%, was a preview. It wasn't the exam. Second, the regulator leans on this segment deliberately, and the November 2023 risk-weight hike was aimed right at it. Assume the rules tighten pro-cyclically. So underwrite the underwriter. Demand vintage curves by cohort in the deck, track the bureau-score migration of the book, and check what share of growth comes from new-to-credit borrowers.