Fairfax Financial Holdings

Summary of findings

A Canadian holding company that owns property and casualty insurers trades at 1.25 times book value and 7.4 times earnings at the September 18 close, while earning 16.9% on its equity. It made money on underwriting in the first half of 2026, is shrinking its share count, and earns four times the interest and dividends it did in 2021. It also lost US$4.4 billion between 2010 and 2016 betting on a crash, and the insurance pricing cycle has started to turn against it.

For an industrial company, the app’s instruments are the argument. For an insurer, most of them misfire, and the app says so before a reader sees a number: its company-type banner tells you to set aside the Graham verdict, the Altman Z-Score, the current ratio and the debt flags, and to trust four things instead — the price-to-earnings ratio, price-to-book paired with return on equity, the dividend record, and regulatory capital. On the first two, Fairfax is inexpensive. The dividend record is intact, despite a flag that says otherwise (section 6). Regulatory capital is not in the app’s data and is not assessed here.

1 · What the company is

Fairfax is a holding company. It owns insurance and reinsurance companies, and it invests the money their customers pay in premiums before claims come due — the float. At the end of 2025 that float was US$40.8 billion. In the first half of 2026 its companies wrote US$18.2 billion of gross premiums, up 4.1% on the year. The portfolios held by its insurers came to US$71.7 billion at June 30, 2026: US$8.2 billion in cash and short-term investments and US$37.6 billion in government and high-quality corporate bonds.

That 93.0% produced a record underwriting profit of US$1.82 billion in 2025. Measured against the shares, the float came to US$1,956 for each one at the end of the year — more than a share cost at the date of value, about US$1,600.

Prem Watsa has run it since 1985, when book value per share was US$1.52. At June 30, 2026 it was US$1,304.39, on 19,969,895 shares effectively outstanding. Common shareholders’ equity was US$26.0 billion; the holding company itself held US$2.3 billion of cash and investments, with its US$2.0 billion credit line undrawn.

2 · What happened last time

Fairfax is unusual among insurers because its chief executive makes large, public calls on the economy. The record of those calls is the most useful thing a buyer of these shares can study, because the next one is already being made with your money.

3 · What rising rates actually do to an insurer

This week’s letter makes the simple case: an insurer’s float is invested, so higher rates mean more income. The full case has a second half, and it is set out here before anyone recalculates after the September 16 rate decision.

The income re-prices slowly

At June 30, 2026 the fixed-income portfolio was 77% government bonds, 12% high-quality corporates (mostly short-dated) and 11% first mortgage loans, with an average term to maturity of 3.2 years. A portfolio that short rolls into new yields within a few years, so higher rates reach the income statement steadily rather than all at once. Interest and dividends were US$1,399.3 million in the first half of 2026, against US$1,272.8 million a year earlier.

The values re-price at once

Rising rates mark down bonds already held. Under IFRS 17, the accounting standard for insurance contracts, they also lower the present value of claims still to be paid, which offsets part of the loss. Fairfax reports both halves each quarter.

On September 16 the Federal Reserve raised its target range for the federal funds rate by a quarter point, to 3.75%–4%, on a 12–0 vote; on July 29 it had held the range at 3.50%–3.75%. The 10‑year US Treasury yield was 5.01% that day, from 4.96% on September 11 and 4.44% at June 30.

Fairfax’s bonds sit much shorter than the 10‑year. The Treasury nearest their 3.2‑year average term, the 3‑year, went from 4.15% at June 30 to 4.82% on September 16: a rise of 0.67 points in the quarter to September 16, against 0.26 in the first quarter and 0.34 in the second. The table above warns against scaling a markdown from that. The second quarter’s rise was the larger, and its bond loss was about a third of the first quarter’s.

The income side moves the other way, and slowly. Two- to five-year Treasuries yielded 4.74% to 4.86% on September 16, close to the 5.0% Fairfax’s fixed-income portfolio yielded in 2025, cash included. At the start of the year the 3-year yielded 3.55%, well below it.

4 · Every tool, on one company

This is the whole toolkit, run on the research page at the September 18 close. The methods are not blended into one value here; the app’s own average and composite score appear as rows, as it reports them. Most of the rows below do not apply to an insurance company, and each one says why.

Reading the disagreement

There is less disagreement than usual, because most of the instruments decline to speak. Of the four that return a value, the three built on what the company earns and owns agree: the Graham Number, Earnings Power Value and the justified price-to-book each put the price about a third below their value. The fourth, the dividend model, reads far below the price, which says only that Fairfax pays out about 7% of what it earns; the value is in the book it compounds, not in the dividend. Price-to-book of 1.25 means little on its own; next to a 16.9% return on equity it means paying just under seven and a half years of earnings for a book that has compounded at 18.7% a year since 1985 (section 8).

5 · Graham’s eight tests

As the research page reads them at the September 18 close.

Five of eight: borderline. The banner discounts the verdict for a financial company, and rightly. Two of the passes measure things that mean nothing on an insurance balance sheet, and the one failure is a loss everyone can name and date.

6 · The one flag the app raises, and what it points to

The app fires one of its eight flags on Fairfax. All eight are printed here.

7 · What is actually wrong: the cycle, and a one-off

Fairfax’s underwriting is profitable: a combined ratio of 93.6% in the first half of 2026 means claims and expenses took 93.6 cents of each premium dollar, before a dollar of investment income. The problem is that the prices behind that margin are falling.

Fairfax says so itself. Watsa’s 2025 letter describes a hard market “that began in 2019 but is now beginning to soften,” its net premiums grew 2.4% in the second quarter “despite a more competitive pricing environment,” and its Zenith workers’ compensation unit ran a 102% combined ratio in 2025 on poor pricing.

And this year’s earnings leaned on a one-off

Net earnings were US$2,088.4 million in the first half of 2026, against US$2,382.4 million a year earlier. The second quarter included an US$838.4 million realized gain on selling 23.1% of Poseidon, alongside a US$91.7 million impairment on non-insurance associates and a US$45.6 million loss from Waterous Energy Fund III. In 2025 the RiverStone run-off unit added US$299 million to reserves, mainly for asbestos and other latent claims. Debt to total capital, excluding the non-insurance companies, rose to 28.0% from 26.2% at the end of 2025.

Poseidon is the holding company that owns Seaspan, which describes itself as the world’s largest independent owner and operator of container ships. Fairfax sold about 23.1% of it for about US$1.91 billion and kept about 22.2%. It is a real gain, but it came from a sale, and only selling more of the remaining stake could repeat it.

8 · Prem Watsa, and the shareholders beside him

Watsa and his holding company own the 1,548,000 multiple voting shares, each carrying 50 votes and fixed together at 41.8% of all votes. With his subordinate shares he controls 43.3%. Two of his children, Christine McLean and Benjamin Watsa, are directors. He is paid a salary of C$600,000, with no bonus, no equity awards and no pension. The management circular describes executive officers’ trading as almost nil, apart from charitable donations and Fairfax’s 2024 purchase of 275,000 of Watsa’s shares for cancellation at US$1,106.48 each, for estate planning, after which he said he was not contemplating further sales.

The record the buybacks are betting on

Selected periods from the table in Fairfax’s 2025 letter. Its stated objective is to compound book value per share at 15% a year over the long term; 2025 came in at 21%, adjusted for the dividend. The forty-year figure is extraordinary; the decade in the middle of it, the hedge years, is the part a buyer should remember.

9 · What would make us wrong

Four named risks, in the order we weight them.

The pricing cycle

Section 7. If prices keep falling into 2027, the combined ratio drifts toward 100% and the underwriting profit that sits beneath the investment income disappears.

A jump in bond yields

Section 3. The income benefit arrives over years; the mark-to-market loss arrives in a quarter, and so far the accounting offset has covered only part of it.

A catastrophe year

Catastrophe losses were US$781.3 million in the first quarter of 2025, mainly the California wildfires, and US$119.3 million in the first quarter of 2026. Falling reinsurance prices arrive exactly when a large loss would test them.

The next macro call

Section 2 lists five since 2003, and the swaps on its own shares in section 8 are one already on the books.

10 · Considered and rejected

Figures for the other insurers are from the app’s screen of 82 US and TSX insurers, dated September 9–14, 2026 (Chubb’s on shareholders’ equity); the last row is at the September 18 close.

11 · Position and disclosure

At the date of value the author owns no shares of Fairfax, and will not buy any until after this file and the letter it accompanies have been published. He owns Berkshire Hathaway, which appears among the insurers in section 10, and Lennar, the subject of Workfile No. 5.