Data status · Using emergency seed prices (table 2026-08-21) — daily feed unavailable · Filings are seeded (EDGAR not yet applied). Emergency seeds dated Aug 21, 2026.

Sources →

The engine

Insurance · float and underwriting

Insurance funds column one and, when combined ratios stay under 100, produces the franchise this desk capitalizes. Ratios and premiums below are from the 10-K / 10-Q reporting groups — GEICO, BH Primary and BHRG — not invented subsidiary valuations.

Float

$177.5B

$82.92 per B · not deducted

UW after-tax run-rate

$6.9B

H1 2026 annualized

GEICO combined · H1’26

89.3%

Loss 75.3 · expense 14.0

UW / float

3.9%

Implied after-tax underwriting yield on float

How the engine works

Three moving parts · only one is capitalized

01

Premiums become float

Policyholders pay now; claims are paid later. The lag is float — money Berkshire holds and invests until losses come due. It sits as a GAAP liability.

02

Float funds column one

That liability is the cheap revolving capital behind cash, T-bills and the public equity book. Adding float on top of those marks would double-count the same dollars.

03

Underwriting is the franchise

When combined ratios stay under 100, the cost of float is negative — Berkshire is paid to hold the capital. That underwriting profit is what this desk capitalizes.

Cost of float on this run-rate is −3.9% — negative: Berkshire is being paid to hold the capital. That is the H1 2026 picture, not a law of nature. Float is 16.7% of market cap. Franchise is 4.5% of IV at 8.0×.

GEICO loss vs expense

Stacked ratios to premiums earned. Combined = loss + expense. Line at 100 is break-even underwriting. H1 2026 is six months of GEICO, not a full year and not the whole group.

2022 was the loss-ratio spike (93.1% loss, combined 104.8%) — float was expensive. 2024 was the repair (combined 81.5%, expense held at 9.7%). H1’26 the loss ratio has drifted up and the expense ratio is rising as GEICO reinvests — still profitable, no longer spectacular.

Combined ratio by group

Under 100 = underwriting profit. BHRG is property/casualty only (ex life/health). H1 2026 BHRG omitted — P/C vs life mix is not split in the 10-Q table we use.

GEICO is a low-expense machine (~10–14%). Primary is specialty commercial: expense ratio near 28%, so a mid-90s combined is still an underwriting profit. Reinsurance P/C has run the mid-80s — catastrophe years will move it.

Premiums earned by group

Calendar-year P/C earned premium, $ billions. BHRG excludes life/health (~$5B a year). Alleghany/TransRe sits inside BHRG/Primary from 2022.

In 2025 GEICO was 53% of this P/C earned book, Primary 22%, BHRG P/C 24%. Volume is not the franchise — combined ratio is. GEICO’s 2024 profit came from a better loss ratio on a book that barely grew in 2022–23.

H1 2026 premiums earned

Premiums, not profit. Six months ended June 30, 2026, all three groups. GEICO’s $22.5B slice is the same H1 2026 row in the ledger — the $39–44B ledger rows are full calendar years for GEICO only.

$44.5B

H1 premium

  • GEICO$22.5B · 51% of H1
  • BH Primary$9.3B · 21% of H1
  • BHRG$12.7B · 29% of H1

GEICO underwriting ledger

GEICO only. Premiums earned are the volume of the auto book; underwriting profit is what’s left after losses and expenses (1 − combined ratio). H1 2026 is six months — do not line it up against a full year. That $22.5B of GEICO premium is the large slice of the mix above.

PeriodPremiums earnedLoss ratioExpenseCombinedUW profit
2022full year$39.0B93.1%11.7%104.8%−$1.9B
2023full year$39.3B81.0%9.7%90.7%$3.6B
2024full year$42.3B71.8%9.7%81.5%$7.8B
2025full year$44.5B72.3%12.4%84.7%$6.8B
H1 2026six months$22.5B75.3%14.0%89.3%$2.4B

10-K years 2022–2025 and 10-Q for the six months ended June 30, 2026. Ratios are as reported (or loss + expense). H1’26 is a half-year, not annualized.

Desk treatment

Capitalize underwriting · do not add float

GEICO, the reinsurance group, and the primary-commercial cluster. Capitalized on after-tax underwriting profit only. Float funds column-one investments and is not added as equity; investment income is the yield on those same assets and is not capitalized. Slider is the same control as Overview — it writes to one store.

$55.2B

$6.9B × 8.0×

After-tax underwriting multiple
In IV
$55.2B franchise
Not in IV
$177.5B float
Also not in IV
$11.5B investment income (yield on column one)

Float through time

Year-end figures reconstructed from 10-Ks and shareholder letters; Q2 2026 is the desk seed ($177.5B). Not XBRL-tagged.

The insurance book

2025 calendar-year P/C earned premium and combined ratio at the reporting-group grain Berkshire actually publishes.

  • GEICO

    Private-passenger auto. Direct writer; the scale engine of primary underwriting.

    $44.5B earned

    Auto · CR 84.7%

  • Berkshire Hathaway Reinsurance Group

    National Indemnity, GenRe, TransRe — retrocessional, catastrophe, structured and treaty. P/C earned shown on the mix chart; life/health sits here too.

    $20.4B earned

    Reinsurance · CR 84.5%

  • Berkshire Hathaway Primary Group

    Specialty commercial: MedPro, Guard, USLI, BH Specialty, RSUI. Higher expense ratio than GEICO; mid-90s combined is still underwriting profit.

    $18.7B earned

    Commercial · CR 95.8%

Letter-grounded

What the letters ask owners to notice

Paraphrase only. Originals on berkshirehathaway.com. Also Letters & Lessons.

Float is revolving capital, not a loan to repay tomorrow

The letters treat float as a rolling fund that is replaced as old policies run off and new ones are written. This desk therefore does not subtract it like parent bonds. That is a modelling choice — read the originals.

Cost of float is the underwriting result

When underwriting is profitable, Berkshire has described the cost of that capital as negative. When it is not, float can be expensive. The 2022 GEICO combined ratio above 100 is the recent reminder.

Do not capitalize investment income twice

Insurance investment income is the yield on cash, T-bills and stocks already marked in column one. Capitalizing it as a third earnings stream would count the same assets twice.

10-K years 2022–2025 and 10-Q for the six months ended June 30, 2026. Ratios are as reported (or loss + expense). H1’26 is a half-year, not annualized. Float is a judgement input — see Sources. Unofficial estimate, not investment advice.