Investing Classics: Financial StocksBanks, Float, and the Discipline of Underwriting

Banks borrow short and lend long: liabilities are deposits that can be withdrawn at any moment, assets are loans and bonds that can't be liquidated instantly, and equity is just a thin sliver (often only 8–10% of assets — roughly 10x leverage). The deadliest combination is maturity mismatch meeting confidence: deposits can flee in an instant, but assets can't be sold in one. So what truly determines a bank's fate is not the balance sheet you can see, but the risk-control culture you can't — which is why Buffett buys only the best-run banks, and only at a fair price.

In March 2023, Silicon Valley Bank (about $209B in assets) collapsed within roughly 48 hours. The root cause: it had parked a surge of deposits in long-duration Treasuries and mortgage bonds; as rates spiked in 2022–23 those bonds racked up huge unrealized losses. When it announced a capital raise to plug the hole, its over-90% uninsured depositors took fright at once — about $42B was withdrawn on March 9 alone. Maturity mismatch plus evaporating confidence produced what was then the second-largest bank failure in U.S. history.

This doesn't mean banks are uninvestable: well-managed banks can compound for years, and Wells Fargo was a Buffett holding for decades. But the reversal lives here too — its 2016 "fake accounts" scandal exposed cultural rot, the risk hardest to detect from outside, and Buffett exited thereafter. Misuse: judging a bank only by price-to-book, while ignoring loan quality and deposit structure.

• Are deposits diversified and sticky, or concentrated in large, flighty clients?
• Is there duration/concentration risk on the asset side that rate or credit cycles could amplify?
• Is management's culture prudent, or chasing returns through aggressive growth?
• Does the price I'm paying leave a margin of safety for "confidence can vanish instantly"?

Float's value depends on its cost = underwriting profit/loss ÷ float. If underwriting is profitable (combined ratio below 100%), the cost of float is negative — others pay you to hold the money, and you still earn investment returns on it. But the precondition is underwriting discipline: the moment you cut prices to chase volume and run an underwriting loss, float becomes interest-bearing debt, with the rate hidden in future claims. It does not automatically equal "free leverage."

In 1967 Buffett acquired National Indemnity for about $8.6M, igniting the engine. Berkshire's float grew from roughly $39M in 1970 to over $160B in 2023, with underwriting profits in most of the past ~20 years — i.e., negative-cost permanent capital that funded large positions like Coca-Cola and Apple. The miracle isn't the float itself, but the combination: "near-free permanent capital + superb capital allocation."

Float can also be a trap. In long-tail lines (asbestos, pollution, liability), claims may surface decades later; if reserves are badly understated, yesterday's "cheap" float turns into ruinous liability. The classic misuse is treating reported float as free money: float that swells from under-pricing is, in essence, borrowing at a hidden high rate.

• Is this insurer's underwriting profitable (combined ratio durably below 100%)?
• Is the cost of float negative, near zero, or above its investment yield?
• Is the business short-tail (fast claims) or long-tail (high reserve-estimation risk)?
• Does float growth come from rationally priced underwriting, or from price-cutting to grab volume?

GEICO (founded 1936) uses a direct-sales model — bypassing agents to sell policies straight to low-risk drivers. Its structural low cost lets it offer lower prices, low prices attract low-risk good customers, good customers bring better loss ratios, which reinforces the cost advantage — a self-reinforcing cost flywheel. Rivals don't fail to imitate for lack of trying; their agent-channel cost structure simply can't copy it.

This company runs through Buffett's whole life. 1951: at 20, he put over half his net worth — about $10,300 — into GEICO, then sold a year later for roughly $15,300; had he held to 1995 it would have been worth over a million dollars, a lesson he calls "selling too early." 1976: GEICO, after reckless expansion and inadequate reserving, neared bankruptcy — the stock fell from about $61 in 1972 to about $2 (−97%); new CEO Jack Byrne turned it around, and Buffett invested about $45.7M. 1996: Berkshire bought the remaining ~half it didn't own for $2.3B, taking 100% ownership.

A moat ≠ immortality. GEICO's brush with death came from within — underwriting gone wrong — not from competitors. The low-cost moat preserved the franchise and bought time to recover, but shareholders who held from 1972–76 still lost about 97%. A moat buys "time," not "exemption."

• Is the cost advantage structural, or just a temporary scale dividend?
• Can the moat withstand internal risks (management, underwriting, culture)?
• Am I mistaking "has a moat" for "no need to worry about operational mistakes"?
• If it suffered a self-inflicted crisis, could the moat hold long enough for it to recover?

The core measure of underwriting is the combined ratio (claims + expenses ÷ premiums): below 100% means an underwriting profit. The temptation: insurance is a commodity, and cutting prices easily grows premiums, while losses surface only much later — so aggressive expansion "looks great" for years, then suddenly blows up. Discipline means daring to let your premium volume collapse in a soft market.

Positive example: National Indemnity's premiums shrank from about $366M in 1986 to about $55M in 1999 (−85%) in a soft market — Buffett refused to write loss-making policies and refused to cut the team, holding out until the hard market let premiums rebound. Counter-example: Lloyd's of London, in 1988–92, saw long-tail asbestos and pollution liability policies written decades earlier erupt at once, with cumulative losses of about £8B; thousands of unlimited-liability "Names" were ruined and many went bankrupt.

Discipline has a price too: prolonged shrinkage loses channels and talent, and excessive caution can miss reasonably priced growth. The biggest misuse is treating fast premium growth as a mark of success — in insurance, premium growth that's too fast is often not a moat, but the fuse on a time bomb.

• Is the combined ratio durably below 100%, or are investment gains masking underwriting losses?
• Does it dare let premium volume shrink in a soft market? Has it ever?
• Does its high premium growth come from rational pricing, or from undercutting to grab volume?
• Has it set limits on catastrophe/correlation risk (one event piercing many policies)?