BRK-B

We rate Berkshire a Hold: we value a class B share at $465 and it trades at $525.68. The price is 13% above our central value and 35% above the purchase price, but inside our bull case ($564). Roughly half of Berkshire's value is cash and listed stock that we take at market. The call rests on the other half: what the operating businesses and the insurance float are worth, and what Berkshire does with its cash.

How we got there. We first valued Berkshire blind, meaning without looking at its price: $491 ($490.74; commit b950f61, pushed before the price was read). Two reviewers then checked it: a model auditor, who reruns every script, and an advocate, who argues the opposite side. Review changed five things, each on dated filing or market evidence, or on the change to our valuation method on main (section 2), none of which depends on BRK's price. The largest is the cash: Berkshire's own record says its cash pile has grown, not been spent, so the central case now leaves today's excess cash in T-bills and deploys 55% of the cash the businesses send up, −$20.19. Net move: −$25.24, to $465 (section 10). Both reviewers kept Hold, and so do we.

The anchor risk. Before building the model we read an average buyback price of about $486, implied by H1 2026 buyback cash over the shares retired. That price level may have anchored (pulled our estimate toward itself) the work. The model auditor traced no input to it, but we cannot rule it out, so treat any closeness to $486 as a warning, not as support.

Terms used below. The central case is our best estimate. The bear and bull cases are the severe but plausible bad outcome and the matching good one; only the central case sets the call. The float is premiums collected before claims are paid; Berkshire owes it to policyholders but invests it meanwhile. GEICO is Berkshire's car insurer.

1. Berkshire is worth $465 a B share: $221 of cash and listed stock, $244 of operating businesses

Berkshire is a holding company of unlike parts. One class A share converts into 1,500 B shares; at 2026-06-30 there were 2,147.5M B-equivalent shares (10-Q Note 18). Values below are per B share at 2026-10-15 (output.txt, central case).

So nearly half the value is cash and listed stock that we take at market, and the call rests on the other half: what the operating businesses and the float are worth, and what Berkshire does with the cash.

• Insurance. Float of $177.5B (6/30/26) costs nothing in an average year. We treat it as a debt to policyholders, take the matching T-bills out of cash, and value only what shareholders get: the after-tax T-bill yield on it plus after-tax underwriting profit, about $9.5B in 2027. We call that the float stream. Central case: a 93.5% P&C combined ratio for ever, Berkshire's own 10-year average (93.0%, letter) plus GEICO's 2026 slippage (H1 combined ratio 89.3% against 81.7%, loss ratio +4.9 points; 10-Q). 2025 was 87.1%.
• BNSF (the railroad). Revenue +9.8% in H1 2026 on 4.3% more volume and fuel surcharges; margin 34.5% in 2025. Central: +7% in 2026, then +4% a year, margin to 37% by 2031. After-tax earnings $6.2B in 2027.
• BHE (Berkshire Hathaway Energy, the utilities). $4.0B after tax in 2025, $2.0B in H1 2026 (+11%). Central: +4% a year from $4.15B. It pays the parent no dividend; its subsidiaries borrowed a net ~$2.7B in H1 2026 (10-Q). The PacifiCorp wildfire accrual is $2.85B to date, $572M of it unpaid. The Oregon Supreme Court hears the class-liability appeal on 2026-11-03.
• Manufacturing, service and retailing (MSR). $13.6B after tax in 2025, $7.7B in H1 2026 (+15%, with OxyChem from 2026-01-02, tariff refunds and divestiture gains in Q2). Central: $15.0B in 2026, $15.6B in 2027 with a full year of Taylor Morrison, then +4%.

Berkshire gives no guidance, so there is no guidance record to measure.

2. The long fade values Berkshire at $465, and no fade horizon reaches the price

Our valuation method has taken the long fade as the value since commit 3e798e2ac on main (2026-10-15). Each part's earnings run 15 years past 2031, with growth fading to 3% and the return on new capital fading to its mature level, then a terminal value at 3% growth. The outside view (what similar listed companies trade at) and the value-driver formula (which links value to growth, return on new capital and the discount rate) are printed beside it as checks, never averaged in.

• The fade's year-five multiples (central) on 2032 earnings, with the formula beside each:

  • float stream 19.3x (formula 17.3x at 2.5% growth; the fade ends at 3%);
  • BNSF 14.0x (14.0x at a 10% mature return);
  • BHE 12.9x (13.0x);
  • MSR 15.3x (16.1x).

  The fade's horizon barely matters here: 10 to 20 years moves central fair value from $463 to $467, all below the price. The blind model also had a judged multiple (a multiple we chose ourselves) and took the median of three values; in every case the median was the fade, so dropping the judged multiple moves no case by more than its terminal growth change (section 10, the row on the valuation method).

• Outside view. Peer forward P/E medians (market data, 2026-10-14):

  • insurance 12.6x;
  • rail 21.3x;
  • utility 16.6x;
  • MSR 20.4x (72% industrial, 28% distribution and retail).

  It sits 7% above the fade. Rail and industrial peers trade on returns that BNSF and MSR, carried at Berkshire's purchase prices, do not earn on new capital. Street consensus EPS for industrials is usually adjusted for acquisition amortisation; our MSR earnings also leave it out (it is booked in "other"), so the basis matches. The peers' ten-year median multiples are not in the saved data, so the split of the gap between rates and the business is not measured.

• Cost of equity 8.29%:

  • 10-year Treasury 5.19% (FRED, 2026-10-14);
  • Damodaran implied equity risk premium (ERP) 4.20% (2026-10-01);
  • Blume-adjusted median beta of 23 fixed peers, 0.738. Beta measures how much a stock moves with the market; the Blume adjustment pulls a raw beta toward 1.

  Comparisons:

  • Berkshire's own beta (market data, 0.601 raw, 0.733 Blume) gives 8.27% and $467: the purchase price moves 0.2%.
  • Damodaran's Diversified industry beta (0.88, 0.92 Blume-adjusted, harsher) gives 9.05% and $429, purchase price $374.

  The fade is sensitive to the discount rate: +1 point is −$46. The blind model's median hid most of this, because a higher rate pushed the fade below the judged multiple, which does not move with the rate.

• Cash, set from Berkshire's own record. The blind model let all excess cash, and all the interim cash (the cash the businesses send up to the holding company to 2031), grow at the cost of equity. Review asked what the record says. Berkshire's filings ($B; 10-K 2024 for 2022-23, 10-K 2025 for 2024-25, 10-Q Q2 2026; output.txt):

  Period	Free cash (operating cash flow less capex)	Acquisitions	Buybacks	Net equity purchases
  2022	21.9	10.6	7.9	34.3
  2023	29.8	8.6	9.2	−24.2
  2024	11.6	0.4	2.9	−134.1
  2025	25.0	1.1	0.0	−13.8
  H1 2026	11.0	9.7	4.8	11.6
  Total	99.4	30.8	24.3	−126.2

  • Cash and T-bills went from $146.7B at 2021 year-end (market data consolidated) to $318.0B (2024), $369.0B (2025) and $359.2B (2026-06-30). The pile grew; none of the 2021 pile was spent.
  • Acquisitions and buybacks took $55.1B of $99.4B of free cash, 55.4%. The rest, and $126B of net stock sales, went into T-bills.
  • The 2025 letter (published February 2026) calls the cash "dry powder" and says Berkshire will aim to own productive businesses over Treasuries, but also that it acts patiently whatever the size of the pile. There is no 2026 letter yet.

  So the central case keeps today's excess cash in T-bills to 2031 (0% deployed, at 3.24% after tax) and deploys 55.4% of the cash the businesses send up, both computed in model.py from the table. That costs −$17.08 (pile idle) and −$3.10 (interim cash 44.6% idle), −$20.19 in all on the blind method (from $486.67 to $466.48). Against it, H1 2026 drew the pile down by $9.8B, $19.6B a year; at that pace from today the central is $471 (+$5). If Berkshire deploys everything at its cost of capital: $486, exit $735, purchase price $407, return 6.7%. One generous assumption remains: the equity portfolio is assumed to stay invested, though Berkshire sold a net $126B of stock over the period.

• Sensitivities (output.txt):

  • combined ratio 96% / 91%: $451 / $480;
  • T-bill yield 3%: $449;
  • cost of equity +1 point: $420;
  • year-five operating earnings ±10%: $445 / $486;
  • mature return on new capital ±2 points: $456 / $472;
  • wildfire $8B instead of $3.5B: $463.

3. After the price was read: the market pays 18.5x for the operating parts, we pay 14.9x

At $525.68 the market value is $1,129B (6/30 share count). Take out the $474B of market items and add back the capitalised holding cost. That leaves $661B for the float, BNSF, BHE and MSR: 18.5x their 2027 earnings of $35.6B. Our central value pays 14.9x. One lever at a time (unblinded-output.txt), the price needs one of:

• operating earnings 28.7% above our path from 2027 on;
• a P&C combined ratio of 83.2% for ever, better than 2025 (87.1%), against a 10-year average of 93.0%;
• 4.5 more turns on every part's 2032 earnings.

Price to book is 1.51x ($348 of book per B share at 6/30/26).

• Five-year return. Today's price to our central exit returns 5.8% a year (5.6% with all cash idle, 6.7% with all of it deployed): 1.2% in the bear, 9.7% in the bull. To earn the 12% hurdle from here, the 2031 exit has to be $949, above our bull ($853).

• Alpha's view. Alpha is Capfolio's statistical ranking model, which scores stocks from their reported data. alpha rank BRK-B --raw --json --allow-stale-model-cache, as of 2026-10-15: rank 221 of 713, score 69.1, "Above Average".

  • Grades: expectations B, growth B, momentum C, valuation C.
  • Integrity verdict WARN.

  Alpha's middling valuation grade agrees with us. Its momentum grade is C despite the recent run: TradingView daily bars show the stock up from $515.62 (2026-10-09 close) to $525.68 (2026-10-15 close), and headlines describe a ten-day winning streak.

• Street. TradingView shows an average target of $541 from 7 analysts (2 Buy, 4 Hold, 1 Sell). Its Q3 EPS estimate of $5.83 a B share is about 4-7% above what our segment forecasts sum to, so our near term is not far from it (the review check for a central case far from the Street does not apply).

4. The market sees a fortress and a deploying buyer; we see a good business priced for its best results

• The market's thesis. A fortress at a modest premium to book.
  • Greg Abel, Berkshire's chief executive, is now deploying: OxyChem $9.4B, Taylor Morrison $6.8B, $4.8B of buybacks in H1 2026.
  • In the news since the blind commit was pushed, unverified: a reported ~$10B in Alphabet's private placement (an AI-generated TradingView recap); $930M in Bending Spoons; Lennar purchases.
  • Insurance earned 87% combined ratios in 2025.
• Ours. The businesses are fine, but the price pays for insurance results at their best year or better, and for a cash pile that the record says mostly waits.
• Deployment. This is where the market could be right. If Abel puts the pile to work at a fair price, our value rises by up to $20 (to $486), still 8% below the price. The H1 2026 drawdown pace gives +$5.
• What settles it.
  • GEICO's loss ratio over the next four quarters.
  • The cash pile in the Q3 10-Q and the next letter: a sustained drawdown into whole businesses moves us toward the deployed case.
  • Whether Berkshire keeps buying its own stock above about $500. Its rule is to buy only below intrinsic value, so heavy buying at today's price would say management's value is well above ours.
  • The advocate estimates (from weekly bars) that the 6.83M B-equivalent shares retired between 6/30 and 7/29 cost about $3.4B at ~$495. The Q3 10-Q will settle that and our buyback falsifier (the buyback condition in section 7).
• Edge. Small. This is a well-understood company, and we are 11% from the price.

5. The advocate's best case for the other side is $502-519, still below the price

The advocate's best case was $502-519 on the blind cash treatment, still below the price, so it cannot make a Buy.

• Our insurance assumption may be too harsh.
  • H1 2026 ran at 88.2% (10-Q unit tables), the five-year average is 90.7% and the twenty-year 92.2% (letter).
  • At 90.7% the central value is $482, still below the price. Against it:
    • H1 had no significant catastrophes and about $1.3B of reserve releases;
    • GEICO's loss ratio rose 4.9 points;
    • the 2025 letter says P&C pricing is softening.
• The cash may be deployed. All of it at the cost of equity adds $20; section 2 gives the record against it.
• The float may be worth more than its stream. Counting half the float-backing T-bills at face adds $14 (all of them, $28); see section 10 for why we do not.
• Our multiples may be too low for an irreplaceable rail network. Rail peers trade at 21x. At their multiple the outside view gives $751 at exit, and the price needs 4.5 more turns on everything.
• Management's own act. The H1 2026 buybacks say management valued the stock above about $486 then. That says nothing about $526, and it is also the figure that may have anchored us (see the top).
• The case-weighted value is $466, the same as the central, so the weights do not argue for the other side.

6. The bear case is $369, 30% below the price

• Combined ratio 97%, flat float, T-bills at 3%.
• BNSF margin slips to 33%; MSR flat after a weak 2027.
• $8B of wildfire claims plus $1B at HomeServices.
• All excess and interim cash idle in T-bills at the bear's 3% yield (2.37% after tax), taken at face today.
• Mature returns on new capital below the cost of equity (BNSF 8%, BHE 8%), so growth adds nothing; the fade gives 12-17x.

That gives $369 today, 30% below the price.

7. Catalysts and falsifiers

Catalysts:

• Q3 2026 results, expected Saturday 2026-10-31 (not confirmed). TradingView lists 2026-11-02. Q3 2025 was posted on the Saturday before its Monday 10-Q filing (2025-11-03). Our forecasts are in the record. assumptions.md gave 2026-11-07 as the expected date; the date is moved here on the filing pattern, and no value changed.
• 2026-11-03. The Oregon Supreme Court hears the James Phase I appeal.
• By 2026-11-16. The Q3 13F shows the deployment.
• Late February 2027 (expected). The annual report, letter and the 2026 combined ratio.

Falsifiers (what would prove us wrong, and what we would do):

• Buybacks above $5B in a quarter at an average above $500 a share: management's value is above ours, so re-examine the returns the fade assumes.
• The P&C combined ratio above 96% for 2026 or 2027: weight the bear ($369).
• BNSF's 2027 operating margin below 33% or above 37%: each margin point is about $1 a share today.
• The Oregon Supreme Court upholds class-wide liability in James: take the bear's wildfire charge.
• An acquisition above $30B at more than 15x earnings: excess cash deployed below our cost of capital.
• Cash and T-bills below $300B by 2027-06-30 with the money in whole businesses at a fair price: move the central toward the deployed case (+$20).

8. Four automatic review checks fired

Each report is run against a fixed list of review checks. A check never changes a number; it tells the reviewers where to look first. Those that fired, each of medium severity:

• Net move from the blind value. The net move is −$25.24 (−5.1%).

  • The largest single change is the cash deployment share, −$20.19.
  • It rests on Berkshire's 10-Ks, 10-Q and letter, not on a management claim about the future or a Street figure.

• A lenient input where a harsher sourced one existed. Our peer beta (0.738) is lighter than Damodaran's Diversified beta (0.88, Blume-adjusted 0.92): −$36.1, purchase price $374. The peer set and the beta rule were fixed blind; the method allows either.

• A batch of reports all on one side of the price. This is one of a batch of reports on 2026-10-15 that all landed below the price (MSFT, ASML, NOW, CRM, VEEV and this one).

• Bridge items with no filing line. Two items in the bridge from business value to a share price have no filing line:

  • the $10B of operating cash;
  • the wildfire charge beyond the $2.85B accrual.

  Both are our assumptions, with sensitivities.

Checks that did not fire: the view changing from the blind view; fair value very close to or very far from the price (−11.4%); the outside view more than 30% from the fade, or the fade's 10- and 20-year values straddling the price (the outside view is +6.6% from the fade; the 10- and 20-year values, $463 and $467, are both below the price); a central case far from guidance or the Street; fair value moving more than 10% between our peer beta and Berkshire's own (+0.2%); an option, convertible or warrant missing from the share count (Berkshire has no options, convertibles or warrants on its stock); no filing read; and reviewers moving the central value more than 20% (−5.1%).

9. Sources and the trail

• Model: model.py → output.txt; after the price: unblinded.py → unblinded-output.txt. Every review finding's dollar effect below is printed by model.py ("review findings measured on the central case").
• Filings and company texts in data/, with price-revealing lines withheld:
  • 10-Q Q2 2026 (filed 2026-08-10) and 10-Q Q3 2025;
  • 10-K 2025 and a 10-K 2024 excerpt;
  • the 2025 letter;
  • six 8-Ks and three press releases;
  • the 13F filed 2026-08-14.
• Market data: market data (peers, holdings closes, the BRK-B profile after the blind commit was pushed, the 2021 consolidated cash), FRED rates, Damodaran betas and ERP, TradingView (Tokyo closes before the blind commit was pushed; the BRK.B price, targets, Q3 date and news after).

10. How each reviewer finding was answered

Two reviewers ran: a model auditor and an opposite-side advocate. Capfolio's coordinating agent then worked from the auditor's rerun. The auditor reran both scripts byte for byte and confirmed these inputs:

• shares;
• cash and T-bills;
• float;
• the portfolio and its deferred tax, each counted once;
• parent debt;
• rates and ERP.

It found pensions $2.05B in surplus (10-K 2025 Note 24); we leave that out, which is harsh by about $1. Effects are on the central fair value, measured in model.py on the final method unless the row says otherwise. "Accepted" means the model changed.

Net of the accepted changes: −$5.41 (BHE payout) −$0.21 (wildfire accrual) +$0.82 and +$0.73 (marks and interest) −$20.19 (cash) −$0.98 (method) = −$25.24, from $490.74 to $465.50. The two changes that raise value (marks and interest) are marks to a date and a known omission; the larger changes lower it.

11. Friction log: what went wrong in the work itself

• Market data returns 402 for Tokyo tickers; TradingView OHLCV worked for the five trading houses.

• berkshirehathaway.com's news page came back compressed to curl. WebFetch read it, then the linked PDFs downloaded directly.

• pypdf is not installed; pdftotext -layout worked.

• The secret scanner flagged a peer's numeric CUSIP in the market data profile JSON. The identifier fields were dropped from the saved peer file.

• Robinhood needed a new sign-in, so the price is TradingView's close.

• Found late: the equity statement's buyback cash divided by Note 18's share change gives an average buyback price. That is a way for the price to leak into blind work for any company that buys back stock, and it was not in our research procedure's list of such routes. It is now in tools.md.

• The reviews found three blind-model mistakes:

  • an interim payout for BHE that the filing contradicts;
  • cash grown at the cost of equity while the record shows it idle;
  • a bear that marked cash down and then grew it at the cost of equity, which charged nothing.

  No review check pointed at any of them. A check for "cash or a cash-like pile grown at the cost of equity" would have.

• The three-value median hid the fade's sensitivity to the cost of equity (+1 point was −$19 on the median, −$46 on the fade).

• Bigdata.com was not used (0 of 3 calls).