Cost of Float
The difference between what Berkshire pays in claims and expenses versus the premiums it collects — determining whether insurance float is a free asset or a costly liability.
“An insurance business has value if its cost of float over time is less than the cost the company would otherwise incur to obtain funds. But the business is a lemon if its cost of float is higher than market rates for money.”
Concept Analysis
Definition & Origins
The cost of float is the price an insurer pays for the money it holds but does not own. Buffett defines it with arithmetic simplicity: the cost of float is measured by the underwriting loss — the gap between the premiums an insurer takes in and the losses and expenses it eventually must pay. When premiums exceed losses and expenses, the cost turns negative: the insurer earns an underwriting profit and is effectively paid to hold other people's money. Everything else in insurance economics — growth, scale, investment returns — sits downstream of this single number.
The concept entered the letters formally in the early 1990s, when Buffett began publishing an annual table of Berkshire's cost of float reaching back to 1967, the year Berkshire bought National Indemnity from Jack Ringwalt and entered the insurance business. What counts, he wrote in 1991, is the "cost of funds developed from insurance," or in the vernacular, "the cost of float." Float itself is defined on the liability side of the balance sheet: the total of loss reserves, loss adjustment expense reserves, and unearned premium reserves, minus agents' balances, prepaid acquisition costs, and deferred charges applicable to assumed reinsurance. The metric reframed the entire operation. An insurance company was no longer to be read as a premium-writing business with an investment sideline; it was a funding business, and like any funding business it had to be judged by what its funding cost.
Core Ideas
The underwriting result is the price of the funding. Buffett's test is stark: an insurance business has value if its cost of float over time is less than the cost the company would otherwise incur to obtain funds — and it is a lemon if the cost runs higher than market rates for money. Underwriting discipline is therefore not an ethical ornament but the mechanism that decides whether float is a free asset or an expensive loan. Berkshire's own record shows the range: a cost near zero across most decades, a staggering 19% in 1984, and 6% in 2000 when General Re's mispriced book produced a $1.6 billion underwriting loss. The combined ratio and the cost of float are the same variable seen from two sides — one measures operations, the other expresses the outcome as an interest rate.
Negative cost is the goal, and it is achievable. In the years when Berkshire has run an underwriting profit, its cost of float has been negative, and its insurance earnings have been calculated by adding the underwriting profit to the income the float produces. That is the double-barreled result no borrower at a bank can obtain: the lender pays you interest for the privilege of holding your deposit, and you keep the investment returns besides. Buffett's expectation for GEICO, stated in the 2000 letter, is that float over time will be free — the standard he applies to every insurance operation Berkshire runs.
Judge the cycle, not the year. Catastrophe years push the cost up; quiet years pull it below zero. Looking back in 2003, Buffett counted five terrible years in which float had cost Berkshire more than 10%, set against 18 of the 37 years in the insurance business in which the company operated at an underwriting profit — meaning it was actually paid for holding money. A single year's combined ratio says little; the long-run average across a full cycle says nearly everything.
Reported cost can deceive. Because loss costs must be estimated, insurers have enormous latitude in figuring their underwriting results, and estimating errors — usually innocent, sometimes not — flow directly into earnings. An experienced observer can usually detect large-scale reserving errors; the general public can typically do no more than accept what is presented. There is also a structural surcharge most investors miss: owning stocks indirectly through an insurance company carries a tax penalty that Buffett estimated at roughly one percentage point added to the cost of float.
Cost of float is the scoreboard of underwriting-discipline. Every element of discipline — refusing underpriced business, shrinking when pricing is inadequate, walking away while competitors chase share — ultimately reports its result here, expressed as a single percentage. The shrinkage years at National Indemnity and the loss-absorbing remediation years at General Re were the price; cost-free float was the product.
Practical Application
Berkshire's own ledger. The 1984 disaster pushed the cost of float to a staggering 19% — proof that float without discipline is expensive money. In 2000 the General Re cleanup produced a $1.6 billion underwriting loss and a float cost of 6%, a figure Buffett reported without varnish. The payoff showed up once the remediation held. By 2003, all of Berkshire's major insurance segments contributed to a $1.7 billion pre-tax underwriting profit, and float reached a record $44.2 billion at no cost at all — the enterprise was being paid to hold money it could invest. The sequence matters: the price was paid in the shrinkage and loss-absorbing years; the cost-free float was harvested afterward.
GEICO versus State Farm, 2000. The same year produced two very different costs. GEICO's underwriting loss of 4% of premiums produced a float cost of 6.1% — a result Buffett explicitly called unsatisfactory — while State Farm, by far the largest personal auto insurer, tolerated a float cost he estimated at about 23%. The willingness of the industry's largest player to pay 23% for its funding sets the pricing environment every other participant must compete in, which is why industry-wide float so often costs more than market money.
Float as an equity substitute. The 1995 letter runs the thought experiment directly: because Berkshire's float had cost virtually nothing over the years, it had in effect served as equity. Had the $3.4 billion of float at yearend 1994 been replaced with $3.4 billion of new equity, Berkshire would have owned no more assets, earned somewhat less, and issued many new shares — more shares, equal assets, lower earnings, materially reduced per-share value. Costless float does the work of equity without dilution.
For outside investors. The transferable discipline is to compare an insurer's long-run cost of float with market rates for money, and to treat the reported figure with skepticism, since reserving latitude lets management smooth the very number that matters most. An insurer whose float persistently costs more than market money is destroying value no matter what its investment arm earns.
Common Misconceptions
Misconception 1: One cheap year proves cheap float. Because loss reserves are estimates, a single year's low combined ratio can be manufactured by optimistic reserving, with the correction flowing into later earnings. Only the cycle average — terrible years included — reveals the true cost of float.
Misconception 2: Cost of float matters only inside insurance. The metric is the general test of any funding advantage: money obtained at a cost above what it can earn destroys value, whether the funding is insurance float, bank deposits, or acquisition debt. It is also the essential input for anyone valuing an insurer from the outside — the difference between a float franchise and a float trap.
Misconception 3: Negative cost, once achieved, can be assumed to persist. Every insurer wants an underwriting profit, and that collective wish creates competition so intense that the property-casualty industry as a whole regularly operates at a substantial underwriting loss — the price the industry pays to hold its float. Berkshire itself slipped to 19% in 1984 and 6% in 2000. Negative cost is re-earned every year or it vanishes.