What has changed, and what was wrong
Figures move for two different reasons: new data arrives, or the old figure was wrong. Both are recorded here, and the second kind quotes what the superseded claim actually said rather than rewriting it quietly.
A valuation updated silently is indistinguishable from one that is never updated, which is why every change to this analysis is recorded here — including the ones that were mistakes.
11 corrections · 2 figure updates · 5 method changes · 11 additions. The figures currently published are as of December 31, 2025 from the 2025 Form 10-K, with book value per share from the Q2 2026 Form 10-Q.
-
An introduction, and the valuation essay moves to /berkshire/
The site now opens with the question readers were actually arriving with: whether Berkshire is worth owning when the price sits above the estimate of intrinsic value. It answers that a premium lowers the return a buyer should expect without removing it, works the arithmetic through a rental house and then through Berkshire, and sends the reader on into the rest of the site. The sum-of-the-parts essay that used to occupy the homepage is unchanged and now lives at /berkshire/.
The feedback that prompted it was consistent: readers were reading "price above intrinsic value" as "do not buy", which skips the step where the premium is priced. Nothing in the valuation changed. /berkshire.html and the old redirect to the root now point at /berkshire/ instead.
-
Berkshire’s own repurchase prices are now on the site
The second-quarter 2026 repurchase table from the 10-Q is published on the introduction, with the policy language it was executed under. Berkshire bought its own stock at an average of $486.08 per B-share equivalent in the quarter, 1.8% below the $495 base case this site reaches independently, and bought nothing at all in April.
The repurchase authorisation is the only place Berkshire applies a test against its own intrinsic value, so the prices it was willing to pay are a revealed floor under management’s conservative estimate. The policy language also changed with the handover: through 2025 the judgement was Warren Buffett’s as Chairman and Chief Executive Officer, and from the 2026 filings it is the chief executive’s after consultation with the chairman of the board. Both versions are quoted.
-
Warren Buffett’s title was wrong, and the page said so emphatically
PreviouslyWarren Buffett remains Chairman of the Board. … He is not retired and he is not emeritus, and the distinction matters for anyone reading the proxy.
He has been Chairman Emeritus since 18 September 2026, when Howard Buffett was elected chairman. The claim above was published here for several hours on the day the Buffett and Munger page launched.
The error came from checking the 2026 proxy statement, which says he "continues to be the Chairman of the Board of Directors" and was accurate when filed in May. September’s announcement superseded it. Reading a primary document is not the same as reading the current one, and the entry is here because the wrong version was stated more confidently than the right one would have been.
-
An em dash meaning “no figure” rendered as a comma in 29 table cells
Previously— (a blank cell)
Cells with no figure printed ", " instead across six pages, including the sum-of-the-parts table, the book value series and the BH Energy dividend record. No number changed; the placeholder marking a number’s absence did.
A scripted copyedit replacing em dashes in prose matched the em dashes used as empty-cell placeholders in templates. It was live for about an hour. The pass now excludes quoted punctuation and table placeholders.
-
Abel’s own figure now cross-checks the BNSF efficiency gap
Greg Abel’s 2025 letter prices each percentage point of operating margin at about $230 million of incremental operating cash flow. That is one percent of BNSF’s $23.4 billion of revenue, which is the same multiplication this site performs, so the two reconcile: $1.45 billion against $1.47 billion.
His gap is 5.7 points rather than the 6.3 used here, because he compares BNSF with Union Pacific’s reported operating ratio of 59.8% while this site recomputes both railroads from revenue and operating income to hold them on one basis, putting UNP at 59.3%. Both bases are defensible and the page now prints both, so the prize is stated as a range of $1.3 to $1.5 billion a year instead of a single figure resting on an undisclosed choice.
-
The two book value pages are now one
/berkshire/book-value-per-share/ redirects to /berkshire/book-value/. The derivation, the series since 1999, the price-to-book record and the float-adjusted multiple sit on a single page. No figure was dropped in the merge; the two pages had each told most of the story separately.
The split existed so that two search queries would land on two pages. In practice each page answered both questions, which cost about 900 words of repetition and gave a reader no reason to visit the second.
-
A page on Buffett, Munger and the structure they left
Sixty years of control, the fifteen elements of what Munger called the Berkshire system, and what changes now that Greg Abel is chief executive. It is the part of the analysis that is not arithmetic, and it states plainly which assumption the valuation is actually underwriting.
-
The comparison of third-party book value per share figures was removed
The book value page carried a table of what data providers published for Berkshire’s book value per share, with a diagnosis of each error. It has been taken down. The job of this page is to be the figure that is right, not a running commentary on other people’s.
Nothing in it was found to be wrong. The derivation it was attached to, which is the part that matters, is unchanged and still set out in full.
-
BNSF’s efficiency gap to Union Pacific was understated
PreviouslyThe 5.5-point gap in operating ratio is worth about $1.3 billion of pre-tax profit a year at current revenue.
The gap is 6.3 points and worth about $1.5 billion. Union Pacific’s operating ratio for 2024 and 2025 was carried as exactly 60.0% in both years, which was a placeholder in the source workbook rather than a figure — UNP earned 59.7% and 59.3% on Capital IQ revenue and operating income. BNSF’s own 2025 ratio was also a tenth light, at 65.5% against the 65.6% its segment accounts show. Both errors ran in the same direction, which is the uncomfortable part: they flattered the business being valued.
Found by tying the segment series to the source workbook tab by tab, which had never been done — the file carried no source statement at all. It now names the authoritative tab for each block, and says plainly which arrays remain unverified: BNSF’s return on equity column and the Union Pacific valuation multiples. The same pass confirmed that BNSF revenue runs above BNSF Railway Company’s own filing because the series is the segment Berkshire consolidates, of which the Railway is a subsidiary — that was an open question and is now closed.
-
The BHE valuation record is now equity value, not price per share
PreviouslyBerkshire valued BHE at $883 a share in 2024, having paid $1,174 in 2022 — a markdown of 25%.
Those prices are not comparable, because the share count behind them is not the same. The $35.05 paid in 1999 bought MidAmerican’s 62.8 million public shares; BHE’s own count was 9.3 million until 2005 and 74–77 million afterwards, and it has been falling since as Berkshire buys the minority out. A per-share series across those years compares three different units. The chapter now runs on equity value — price multiplied by the shares outstanding in that year, which is the same quantity throughout and the one that appears in the sum of the parts — with the compound return and the implied earnings multiple alongside it.
The markdown moves with the change: 29% on equity value against the 24.8% the per-share figures showed, because the share count fell as well as the price. The 1999 acquisition is now the starting point of the series, so the record reads $2.2 billion to $88.8 billion in 2022 — 40.4× at 17.4% a year — and back to $62.9 billion in 2024, or 14.4% a year since the purchase. The 2004 mark of $100 a share is excluded: at 9.081 million shares it values the company below its own 1999 price and under a third of book equity, and the source workbook computes no multiple against it.
-
BH Energy added to the return-on-equity comparison, adjusted for its BYD holding
The chart comparing returns on carrying equity showed MSR against BNSF. It now carries BH Energy as well, which earns more than either because Berkshire built it by subscribing for new shares rather than buying it from anyone, so no acquisition premium sits in its equity.
BHE is shown on an adjusted basis and labelled as such. It holds a listed stake in BYD which is marked to market through earnings, and the swings are larger than the utility result: the mark added $4,797 million to pre-tax profit in 2020 and removed $2,002 million in 2022, making the reported return read 17.6% and then 5.7%. Neither MSR nor BNSF holds anything equivalent, so the reported figure is not on their basis. The adjustment uses the 25% effective rate implied by the gross and net BYD figures BHE itself publishes, rather than an assumed rate. On the adjusted basis the low point of the record is 2023, the wildfire year, not 2022. Both series are shown on the BH Energy page. One figure is left unadjusted and flagged instead: 2008 includes a large one-off gain from the terminated Constellation Energy acquisition, which sits inside a line that also carries recurring items and cannot be cleanly separated.
-
Berkshire Hathaway Energy rebuilt as a full deep dive
The BHE page explained how a regulated utility earns and little else. It now carries twenty-seven years of assets, equity, earnings and return on equity; every repurchase Berkshire has made in BHE and the valuation each one implies; the capital expenditure and depreciation behind the zero-dividend policy; the earnings split by business; generation capacity by fuel; and the data centre load the company now quantifies.
The finding that drove the page: Berkshire repurchased BHE stock at $1,174 a share in 2022 and at $883 in 2024 — a markdown of 25% — while making its largest ever repurchase, $2.9 billion, at the lower price. BHE is not listed, so these transactions are the only direct read on what its owner thinks it is worth. Subsidiary figures are BHE’s segment view rather than each subsidiary’s own balance sheet; the two differ by a fixed amount per subsidiary, which is goodwill pushed down at the BHE level, and only the segment view reconciles to the consolidated total. Where BHE stopped breaking a business out around 2019, the series continues from Capital IQ for those that still file their own accounts and stops for those that do not.
-
The book value chart said it plotted average prices. It plots year-end closes.
PreviouslyThe chart of book value, book value plus float and price carried the note “Price is the average for the year.”
It does not. That series is the year-end closing price, and always has been — the 2025 point is $502.65, the close on 31 December 2025. The description belonged to the price-to-book chart below it, which does use the average price, against the prior year’s book value. The caption now states the basis the chart actually uses. No data changed; the label was describing the wrong chart.
-
The book value chart now runs to the latest close, not to last December
It stopped at the 2025 year end, which made a chart about what the market pays for this company ten months out of date. It now continues through the two quarters filed since — book value per share of $337.15 at 31 March 2026 and $348.26 at 30 June, float of $176.9 and $177.5 billion — and the price line runs on to the last close of $546.97 on 22 October 2026.
Two things are deliberate. Book value and float stop at the last reported quarter rather than being carried forward to meet the price: they are only known at a balance sheet date. And the horizontal axis is now decimal dates rather than whole years, so three months of 2026 takes up a quarter of the width of a year instead of a whole one — plotted the old way, the three new points would have been spread across nearly two years of the axis. The year-end series is kept separate from the part-years in the data, because the long-run growth rate, the median multiple and the below-book counts are all year-end statistics.
-
The underwriting average was described as fourteen years. It is seventeen.
PreviouslyThe homepage, the valuation page, the insurance page and the glossary all stated that insurance underwriting is carried at its "fourteen-year average".
The published $2.0 billion is the average pre-tax underwriting profit from 2009 to 2025 — seventeen years — taxed at 23%. A genuine fourteen-year average would be $2.22 billion. The description was wrong in five places and is now derived from the data rather than written into the prose, so it cannot drift again. No valuation figure changed; the label did.
One of the five was a literal 14 wrapped in a template expression, which is why it survived earlier checks for hardcoded figures.
-
The liquidity reserve, explained correctly at the third attempt
PreviouslyThe $30 billion liquidity reserve was annotated "Per the 10-K: capital that must stay put to pay claims." During the day it was twice re-explained and twice wrongly: first as a judgement with no basis in the filing, anchored to statutory surplus of $333 billion; then as the forecast claim payments on prior-year occurrences, which happen also to be about $30 billion.
Neither was right. The figure is Berkshire’s own liquidity floor: the 10-K states that Berkshire will not repurchase stock if doing so would reduce consolidated cash, cash equivalents and U.S. Treasury Bills below $30 billion, adding that "financial strength and redundant liquidity will always be of paramount importance". It is a theoretical maximum call on liquidity the company has committed to keep covered, not a claims forecast and not a regulatory requirement.
The floor was $20 billion through fiscal 2020 and was raised to $30 billion from fiscal 2021. Its size follows from the other limit Berkshire publishes: it avoids writing policies from which a single catastrophe might aggregate more than $15 billion of pre-tax loss, so the floor covers two maximum-exposure events landing together. The insurance page now sets out that reasoning. The deduction has been $30 billion throughout and no valuation figure moved at any point; the error was entirely in the explanation, and the searches behind the first two attempts looked for claims and regulatory language rather than the buyback condition where the commitment actually appears.
-
The underwriting average is now shown year by year
The insurance page now shows the pre-tax underwriting result for every year from 2000, split by unit, marking which years fall inside the window the normalised average is struck over. That window is 2009-2025, seventeen years, producing $2.60 billion pre-tax and $2.00 billion after tax at 23%. The three loss years before it are excluded deliberately: they cost $6.1 billion pre-tax, 92% of it from the reinsurance group as General Re’s reserves were strengthened and the September 2001 attacks were paid out, while GEICO stayed profitable. The excluded years are shown rather than dropped, so the choice can be argued with.
-
Corrected GEICO’s policy counts, and removed an invented figure
PreviouslyThe GEICO page stated that policies in force "grew 3.0% in 2025, to 15.5 million", that the count peaked at 18.7 million in 2020, and that 3.6 million policies — 19.4% of the book — were lost.
The 3.0% was not a disclosed figure. Berkshire’s 2025 Form 10-K attributes the rise in premiums written to "an increase in policies-in-force over the past year" and gives no magnitude anywhere in the filing. The number had been estimated and then presented as fact. The policy series now ends at 2024, and the page states the direction of the 2025 change in words, because that is what is disclosed.
Rebuilding the series on disclosed figures alone also moved the earlier years. The anchor is the 2020 filing — "an increase of approximately 820,000 (4.6%)" — carried by the percentage stated for each year since. On that basis the count held level across 2020 and 2021 rather than peaking in 2020 (Berkshire called 2021 "slightly higher" and "flat"), and the fall to 2024 is 3.4 million, or 18.2%, not 3.6 million and 19.4%. Two independent 2022 disclosures — a 8.9% decline and "a reduction of 1.7 million" — agree to 0.2%, which is the best check available on the series. Figures before 2018 remain interpolated from five-year cumulative disclosures and are labelled as estimates in the data.
-
The cost of float, and the underwriting split behind it
The insurance page asserted that float is worth more than the underwriting profit it comes with, without ever showing the figure that settles it. It now carries the net cost of float by year: negative in 21 of the 26 years where both figures are published, averaging -1.3%, meaning Berkshire has generally been paid to hold the money rather than paying for it.
The page also splits the underwriting result five ways, which a single combined ratio hides. GEICO, the reinsurance group and the primary insurers have earned $70.4 billion pre-tax since 1999; retroactive reinsurance and the periodic payment annuities have cost $19.1 billion, deliberately, because what they buy is long-dated float. Float by originating business is shown for 2016, the last year Berkshire published the split, and is not extrapolated past it. Float growth is struck from 1998 rather than 1997: General Re closed in 1998 and tripled the float in one step, which is a purchase rather than growth.
-
GEICO rebuilt as a full deep dive, with Progressive alongside
The GEICO page previously explained only why GEICO has no line of its own in the valuation, and said the comparison against Progressive was "the page worth building, and it is not built yet". It is now built: 28 years of GEICO’s underwriting record from the annual reports, against Progressive’s personal-vehicle book, with policies in force, the combined ratio split into its loss and expense halves, and insurance-group headcount.
The finding that drove the page: GEICO wrote 1.33 times Progressive’s personal-vehicle premium in 2017, crossed below parity in 2022, and writes 0.65 times today — having given up roughly a fifth of its policies to recover an underwriting margin. GEICO’s policy counts after 2020 are reconstructed from the percentage changes Berkshire discloses in words, and are labelled as estimates on the page.
-
Book value per share moved to the second quarter of 2026
Price-to-book multiples were being struck against the 31 December 2025 balance sheet while the share price was live. Ten months of retained earnings had accumulated in between, so the multiple read roughly a point too high. Book value per share is now taken from the Q2 2026 Form 10-Q at $348.26, derived as Berkshire shareholders’ equity over equivalent Class B shares outstanding at the balance sheet date.
Berkshire stopped publishing book value per share, so the figure has to be derived. The share count comes from the common stock note rather than the filing’s cover page: the cover page is dated a month after the balance sheet and reflects buybacks made in between, which would put a smaller denominator under an earlier numerator.
-
Scenarios are carried forward when compared with a current price
Refreshing the book value alone created the opposite error: a live share price against an intrinsic value struck at the 2025 year end made the shares screen above even the upside case. That was an artefact of comparing two dates. Where a current price is compared with the scenarios, they are now carried forward at the assumed rate of compounding for the elapsed period, and labelled as rolled forward. The published sum of the parts is unchanged.
-
The compounding case is stated explicitly
The site estimated what a share was worth without ever saying plainly what that implied about returns. It now does: buy at a multiple of book value, sell at the same multiple, and the return is whatever book value compounded at in between. The base case is therefore not a forecast of the share price but the price at which a buyer should expect roughly ten per cent a year from the business alone.
-
Glossary and this revision log added
Twelve terms defined in the sense this analysis uses them, with the part a dictionary omits — what each one does in a valuation of Berkshire.
-
Corrected a claim about how often Berkshire has traded below book plus float
PreviouslyThe book value page stated that Berkshire had traded below one times book value plus float “for most of its history”.
It has not. On year-end prices it is five years of twenty-six. The sentence was rewritten and the underlying counts are now derived from the price history rather than asserted in prose, so the claim cannot drift from the data again.
-
Corrected the cheapest year on record
PreviouslyThe page named 2008 as the cheapest the shares had been, at 1.02 times book plus float.
The cheapest year-end on the record carried here is 2020, at 0.93 times. Both the cheapest and dearest years are now computed from the series instead of being written into the text.
-
Reconciled the Manufacturing, Service & Retail revenue figure
PreviouslyThe valuation page and the MSR deep dive disagreed about the segment’s revenue — $213 billion against $214.3 billion.
The deep dive’s figure was right. The valuation page carried a separate copy that had not been updated. Revenue there is now derived from the same series the deep dive charts, so the two cannot disagree. No valuation figure moved: revenue is displayed on that page but the value is a multiple applied to profit.
-
All nine segment deep dives rewritten
Each segment page was a table of numbers. They are now editorial pages with charts, takeaways and the historical evidence behind each assumption in the sum of the parts. Historical series moved out of the page templates into a shared data layer, which is what surfaced the MSR disagreement above.
-
Valuation updated to fiscal 2025
The sum of the parts, the scenario range and every per-share figure were updated from the 2025 Form 10-K. This is the annual revision; the figures reflect the 31 December 2025 balance sheet unless a page says otherwise.
Found something else wrong? That is the most useful thing you can send me — intrinsicvaluepershare@gmail.com. Corrections are published here with the original claim intact.
Follow the research
New valuations and research notes, a few times a year. No promotional email, no sharing of your address, unsubscribe in one click.
Delivered by Buttondown. You can also subscribe by RSS.