The business that is valued at almost nothing
The insurance underwriting line is one of the smallest in the valuation. The insurance operations are the reason the valuation looks the way it does. Both statements are true, and the gap between them is the most important idea on this page.
Berkshire's insurance operations are valued at close to nothing for their underwriting and almost everything for their float, the premiums they hold before claims come due are interest-free money with no maturity date.
- 1 Underwriting is carried at $20 B, 2% of the valuation. That is $2.0 B of normalised profit at 10×, and the $2.0 B is a 17-year average, not any single year’s result.
- 2 The float is about $176 B. Premiums collected and not yet paid out, which Berkshire invests and keeps the returns on. It is the reason $242 B of Treasury bills sits inside the insurance subsidiaries, 23% of the valuation, against underwriting’s 2%.
- 3 Float is not free capital, it is cheap capital. It only works while the underwriting breaks even or better, which is why the valuation values the underwriting at its long-run average, not at a good year, and why the downside case deducts $30 B of reserves that cannot leave.
How an insurer funds itself
An insurer collects premiums today and pays claims later, sometimes decades later. In between it holds the money. That holding is called float, and it behaves like a loan from policyholders: it has to be repaid, nobody knows exactly when, and in the meantime it can be invested.
The cost of that loan is the underwriting result. Pay out more in claims and expenses than you collect in premiums, and the float has cost you something. Come out even or ahead, and you have been handed $176 B to invest for nothing. Berkshire has mostly come out ahead, which is why it has $242 B of Treasuries sitting inside the insurance companies and a $298 B equity portfolio alongside them.
Why underwriting is valued at an average. Catastrophe reinsurance, one of Berkshire’s specialities, produces a run of profitable years and then a single very bad one. Valuing it on last year’s result would mean valuing it on whether a hurricane happened to make landfall. The $2.0 B used here is a 17-year average, and the slider on the main page runs from 0 to 5 billion to let you argue with it.
26 years of float
Float per share grew from $10.96 in 1999 to $81.58 in 2025, 8.0% a year, compounded, without Berkshire raising a dollar of equity to do it. This is the engine of the whole company, and it is close to invisible in the income statement.
Why Berkshire is less of an insurance company than it was
In 1999 the float was 45% of book value. Today it is 25%. The float did not shrink, it grew 7.4×, but book value grew faster, because the operating businesses Berkshire bought with the proceeds now generate earnings of their own.
This matters for anyone sizing the risk. A company whose capital is half borrowed from policyholders is a different proposition from one where it is a quarter. The leverage that made Berkshire’s early decades extraordinary has been steadily diluted by its own success.
The takeaway The underwriting line is small; the float it produces is not. Underwriting and the insurance Treasuries together are 25% of the base case. Read the two lines as one business and the insurance operation is the largest thing Berkshire owns.
What the float has actually cost
Float is only worth more than debt if it is cheaper than debt, and that is a measurable claim, not a slogan. The cost of float is the underwriting result divided by the float carried: an underwriting profit makes it negative, which means Berkshire was paid to hold other people’s money.
Over the 26 years where both figures are published, the average is -1.3%. It was negative in 21 of them. The exceptions are the years worth knowing: 2000, 2001, 2002, 2017, 2022. Three of them fall in the hard reinsurance market at the start of the record, one is the $3.0 B catastrophe year, and one the 2022 inflation shock.
That is the whole case for valuing the float the way this site does. Money borrowed at a negative rate, with no maturity and no covenant, is not a liability in the ordinary sense, and $176 B of it is a larger number than almost anything else on Berkshire’s balance sheet.
Two businesses wearing one combined ratio
Berkshire’s group underwriting result is the sum of five units, and reading it as one number hides the most interesting thing about it. Three of them underwrite for profit. Two of them lose money almost every year, on purpose.
Since 1999, GEICO, the reinsurance group and the primary insurers have earned $70.4 B of pre-tax underwriting profit between them. Retroactive reinsurance and the periodic payment annuities have cost $19.1 B. The net is $51.4 B.
The two loss-makers are not mistakes. A retroactive reinsurance contract takes over claims that have already happened: Berkshire is paid a premium now and pays claims out over decades, booking an accounting loss as the deferred charge unwinds. What it buys is the longest-dated float available anywhere. The annuities work the same way. Both are deliberate purchases of time, and the $19.1 B is the price.
The takeaway Judge the underwriting units on $70.4 B and the float purchases on the float they bought, not both on one combined ratio. The $19.1 B spent on retroactive reinsurance and annuities is the cost of carry on the longest-dated money in the group.
Why underwriting is carried at an average
Reported underwriting profit is useless as a valuation input: it swung from ($4.1 B) to $11.4 B inside this record. The line is therefore carried at an average taxed at 23%, struck over 17 years, 2009 to 2025. That is $2.60 B pre-tax and $2.00 B after tax, which is the $2.0 B the valuation uses. The table runs the whole record so the excluded years are visible instead of quietly dropped. An assumption a reader cannot audit is just an assertion.
| Year | GEICO | Reins. | Primary | Retro | Annuity | Total |
|---|---|---|---|---|---|---|
| 2000 | (0.22) | (1.23) | 0.03 | (0.19) | 0.00 | (1.62) |
| 2001 | 0.22 | (3.95) | 0.03 | (0.37) | 0.00 | (4.07) |
| 2002 | 0.42 | (0.41) | 0.03 | (0.45) | 0.00 | (0.41) |
| 2003 | 0.45 | 1.58 | 0.07 | (0.39) | 0.00 | 1.72 |
| 2004 | 0.97 | 0.83 | 0.16 | (0.41) | 0.00 | 1.55 |
| 2005 | 1.22 | (1.19) | 0.24 | (0.21) | 0.00 | 0.05 |
| 2006 | 1.31 | 2.33 | 0.34 | (0.17) | 0.00 | 3.81 |
| 2007 | 1.11 | 2.36 | 0.28 | (0.38) | 0.00 | 3.37 |
| 2008 | 0.92 | 2.08 | 0.21 | (0.41) | 0.00 | 2.79 |
| 2009 | 0.65 | 1.27 | 0.08 | (0.45) | 0.00 | 1.56 |
| 2010 | 1.12 | 0.72 | 0.27 | (0.09) | 0.00 | 2.01 |
| 2011 | 0.58 | (1.10) | 0.24 | 0.65 | (0.12) | 0.25 |
| 2012 | 0.68 | 1.02 | 0.29 | (0.20) | (0.16) | 1.63 |
| 2013 | 1.13 | 2.08 | 0.39 | (0.32) | (0.18) | 3.09 |
| 2014 | 1.16 | 1.71 | 0.63 | (0.63) | (0.20) | 2.67 |
| 2015 | 0.46 | 1.22 | 0.82 | (0.47) | (0.20) | 1.84 |
| 2016 | 0.46 | 1.20 | 0.66 | (0.06) | (0.13) | 2.13 |
| 2017 | (0.31) | (1.65) | 0.72 | (1.33) | (0.67) | (3.24) |
| 2018 | 2.45 | 0.01 | 0.67 | (0.78) | (0.34) | 2.01 |
| 2019 | 1.51 | 0.18 | 0.38 | (1.27) | (0.38) | 0.42 |
| 2020 | 3.43 | (0.82) | 0.11 | (1.25) | (0.64) | 0.84 |
| 2021 | 1.26 | 0.43 | 0.61 | (0.78) | (0.40) | 1.11 |
| 2022 | (1.88) | 2.29 | 0.39 | (0.67) | (0.16) | (0.02) |
| 2023 | 3.64 | 3.86 | 1.37 | (1.54) | (0.42) | 6.91 |
| 2024 | 7.81 | 4.02 | 0.86 | (0.90) | (0.44) | 11.35 |
| 2025 | 6.82 | 3.54 | 0.79 | (0.95) | (0.62) | 9.58 |
| Avg 2009–2025 · $2.00 B after tax | 2.60 |
Why 2000–2002 is excluded
Those three years cost $6.1 B pre-tax, and 92% of it came from one place: the reinsurance group. General Re’s reserves were being strengthened after Berkshire acquired it, and the September 2001 attacks produced the largest insured loss in history to that point. GEICO was profitable throughout, earning $0.4 B across the same three years. This was a single identifiable event in a reinsurance book written on terms Berkshire no longer accepts, not evidence about the underwriting the group does today. Averaging across it would be normalising for a scenario that has been specifically underwritten against since.
That is a judgement, and it is the one worth arguing with. Include those years and the after-tax average falls to $1.52 B; take only the last ten and it rises to $2.39 B. The $2.0 B used here sits 16.5% below the recent figure, so against the business as it is now run the assumption is conservative. The slider on the main page runs from 0 to 5 billion for a reader who weighs the tail differently.
$30 B is two of the worst days Berkshire will insure
The $30 B deducted in the downside case is not a claims estimate and not a regulatory capital figure. It is Berkshire’s own floor, a minimum level of cash the company has committed never to go below, stated in the 2025 Form 10-K in the plainest terms it uses anywhere:
Berkshire will not repurchase its common stock if the repurchases reduce the value of Berkshire’s consolidated cash, cash equivalents and U.S. Treasury Bills holdings to less than $30 B.
The size of it is the interesting part. Berkshire states that it writes its book to avoid groups of policies from which pre-tax losses from a single catastrophe event might aggregate above $15 B. That is the largest single hit the group is willing to underwrite by design. The floor of $30 B is 2× that number, so it is sized to absorb two maximum-exposure events at once: two separate catastrophes, each at the limit the company deliberately caps itself at, landing in the same period, with every obligation still met without selling an investment, drawing a credit line or issuing a share.
That is what makes it a theoretical maximum liquidity requirement and not a forecast. It is not what Berkshire expects to pay. It is what Berkshire has arranged to be able to pay in a scenario it does not expect at all.
The commitment is expressed as a limit on buybacks because repurchases are the one use of cash Berkshire controls completely: it cannot decline to pay a claim, but it can always decline to buy its own stock. The same passage adds that “financial strength and redundant liquidity will always be of paramount importance at Berkshire”. In practice the floor binds every use of cash, not just the buyback. It has moved once, from $20 B through fiscal 2020 to $30 B from fiscal 2021, a 50% increase as the group and its exposures grew.
The takeaway In the downside case this $30 B is not investable capital and not distributable to owners. It is the cost of the group’s own insurance against itself, and the reason Berkshire can write catastrophe cover competitors cannot price at any level. Deducting it is the conservative treatment; the alternative is to assume a buyer would run the balance sheet in a way Berkshire has said it never will.
One figure in the same filing is worth not confusing with this one. Berkshire expects to pay more than $30 B in 2026 on claims that have already occurred, against total claim liabilities of $152 B. That is the ordinary run-rate of the business, funded by premiums and investment income as it comes due. It happens to be a similar number and it is not the same thing as the floor above.
Where the float comes from
Berkshire published float by originating business until 2016 and then stopped. At that last full disclosure the reinsurance group was the largest single source, not GEICO, which is the business most readers associate with the float.
GEICO’s own float is still derivable and has continued to grow, reaching $30.5 B. The rest of the $176 B is no longer split in public, so this page shows the 2016 composition and does not extrapolate it forward.
Float growth has averaged 7.9% a year since 1998. The record starts at 1997, but compounding from there would report a far higher rate: General Re closed in 1998 and multiplied the float 3.1× in a single step. That was a purchase, not growth, so the rate above is struck from 1998.
Download the series (.xlsx) Berkshire insurance float, with sources and the date it was struck written into the file. This page’s series only; there is no combined download.