Japan's five trading houses

What a sogo shosha is

The name translates as “general trading company”, which is accurate about the history and misleading about the business. These houses began as intermediaries for a country with no natural resources, buying raw materials abroad and selling manufactures outward, and they still run that business. But the trading is no longer where the money is. Over decades they used the margin from it to buy the assets at either end — iron ore and coking coal mines, LNG trains, power stations, grain terminals, car dealerships, convenience store chains, shipping fleets, pharmaceutical distributors, salmon farms.

What each one is today is a portfolio of operating stakes with a trading desk attached. That has two consequences for anyone trying to value one. A large share of the profit arrives as the group's share of affiliates it does not consolidate, so an earnings multiple is measuring something other than the earnings the group controls. And consolidated revenue is close to meaningless, because whether a trade is presented gross or net changes the top line by multiples without changing a yen of profit. The sum of the parts is the only frame that works, which is the same conclusion this site reaches about Berkshire — and the reason these five belong here.

They are not interchangeable, and a reader who treats them as one trade will be surprised in both directions. Mitsubishi and Mitsui are resource-weighted and move with iron ore, coking coal and LNG. Itochu deliberately is not, and its record shows it. Sumitomo is the cheapest of the five on both multiples and has the patchiest history. The individual deep dives are where that gets settled.

What is inside them

The chapter above says what these companies are and the one below says what they returned. This one is the part in between, and it is the part that only appears once all five balance sheets are built on the same basis. Three things are true of every one of the five. None of them is visible from any single house, and together they settle most of what a reader needs before deciding whether this group is cheap.

Half the profit is earned outside the consolidation

A sogo shosha reports two kinds of profit. There is what its subsidiaries earn, which runs through its own revenue and costs, and there is its share of what its affiliates earn — companies it holds a stake in, has influence over, and does not control. The second arrives as a single line near the bottom of the income statement, and at these companies that line is enormous. Over the ten years to March 31, 2026, 46.2% of the five houses' combined net income came from it. Not one of them was below 39.9%.

Profit earned outside the consolidation Equity-method income as a share of net income · fiscal years ended 31 March

The share is struck on the sums rather than as an average of the ten yearly ratios, which would weight a small year equally with a large one and would have to drop the two loss years outright. It is the ratio of two published lines, so it is exact as a ratio; it is not a claim about how much of the economic value sits outside the group, because a consolidated subsidiary's minority interests are accounted for elsewhere. What it does establish is that the earnings column in the table above is measuring something materially different from the earnings of the business the group controls — most of all at Marubeni, where the ten-year figure is 57.6%, and least at Itochu at 39.9%. Which companies those are, and what they are worth, is the question each deep dive spends most of its length on.

Every one of them has taken leverage down

The common shorthand for this group is a levered bet on Japanese industry funded with cheap debt. The accounts say the opposite happened. Measured as average total assets over average equity, leverage fell at all 5 houses between the year to March 2018 and the year to March 2026 — furthest at Marubeni, where it came down 38.1%. The five now sit in a band from 2.3× to 2.7×, which is narrow enough that borrowing is close to a constant across the group rather than a variable distinguishing them.

Leverage, first year to last Average total assets over average equity

Four of the five got more productive. The exception is the largest

Leverage coming down while returns held up means the assets themselves had to do more work, and at four of the five they did. Return on average assets rose at 4 of the 5, led by Marubeni at 79.7%. The exception is Mitsubishi, the biggest of the five, where it went 0.5% the other way — which is why it now earns the lowest return on equity of the group despite having started the period in the middle of it.

Return on equity is return on assets multiplied by leverage, exactly, so the distance between any two of these houses is the sum of two effects and nothing else. The widest gap in the group is the 6.1% between Itochu at 14.6% and Mitsubishi at 8.5%. Hold leverage still and give Mitsubishi the other's asset productivity, and 85.6% of that gap closes. Hold productivity still and move the borrowing instead, and 14.4% of it does. The difference between the best and worst of these companies is what they own, not how much they borrowed to own it. That is the single most useful thing to carry into the five pages, and it is why the question each of them asks is about assets rather than about financing.

Ten years, on fundamentals

Return on equity is struck on average common equity, which is the basis the companies use themselves — it is why these figures tie to their own presentations rather than sitting a point away from them. The loss years are left in, because they are the most informative rows in the table: Marubeni lost $1.2 B in the year to March 2020 and Sumitomo $1.0 B the year after, on resource impairments. Any series that starts after those years shows five flawless compounders.

Return on equity Per cent · fiscal years ended 31 March

The two dashed rules on that chart are the only forward-looking figures on this page, and they are quoted rather than estimated: every one of the five publishes a return target in a named medium-term plan, and the five cluster on two numbers. Mitsubishi aims at 12.0% by 2028 (Corporate Strategy 2027); Mitsui aims at 12.0% by 2029 (MTMP2029); Marubeni aims at 15.0% by 2028 (GC2027); Itochu aims at 15.0% by 2027 (FY2026 Management Plan); Sumitomo aims at 12.0% by 2027 (Medium-Term Management Plan). The years are not the same, so each one travels with its figure. Neither are the distances left to run: Mitsubishi is 3.5% short of its own number, while Sumitomo is already past its target — which is worth holding against the fact that Sumitomo is also the cheapest of the five on both multiples, and set the lower of the two bars.

Market capitalisation compounded faster than earnings at every one of the five, and the share count fell at every one of them. The first gap is the multiple; the second is buybacks, which is new behaviour for these companies and the clearest evidence that the Tokyo Stock Exchange's pressure on sub-book valuations changed something real. Across the five, $29 B went on repurchases in the last 5 years alone.

Why Berkshire owns more than a tenth of each

On August 31, 2020 Berkshire disclosed that it held slightly more than five per cent of all five. Issued 8:30 a.m. JST on 31 August, the evening of 30 August in the United States — Buffett's 90th birthday. The stakes were acquired by National Indemnity over approximately twelve months of regular purchases on the Tokyo Stock Exchange.

“I am delighted to have Berkshire Hathaway participate in the future of Japan and the five companies we have chosen for investment.”

It was a news release, not a shareholder letter, and the distinction matters to anyone citing it: it went out at 8:30 in the morning Tokyo time and carried a single paragraph from Buffett.

The release also contained a limit. Buffett pledged purchases only up to a maximum of 9.9% of any of the five. That ceiling, and the fact that it has since been lifted, is the shape of everything that followed.

Only two of the five crossings carry a published percentage: Marubeni at 10.1% and Sumitomo at 10.1%, both on May 7, 2026. The other three are above 10% without a current figure, so this page prints the floor rather than inventing precision the filings do not give. A flat 10% of the five is $42 B.

Funded in yen, which is the part usually left out

Berkshire did not buy yen and then buy shares. It borrowed yen, and bought the shares with the proceeds. A yen asset financed by a yen liability has no currency exposure between the two: when the yen falls, the dollar value of the shares falls and the dollar cost of repaying the debt falls with it. There is no hedge to roll and no counterparty — the offset is structural, because both sides are denominated in the same money.

What is left is a spread. Berkshire pays a yen coupon and receives yen dividends, and the difference is cash every year the position is held. The two most recent issues priced at 2.4% (November 2025, 10-year) and 3.1% (April 2026, 2036); the April 2026 deal was ¥272.3 B across 6 tranches maturing 2029 to 2056.

The carry argument has a condition attached, and it is worth stating because it is usually skipped. On Berkshire's cost the dividends clear the coupon with about 1.1% to spare. On today's price they do not: 2.1% against a 3.1% coupon is a negative 1.0%. The positive carry is a function of a low basis, not of Japanese dividend yields — so fresh yen borrowing to buy more of the five at these prices would not pay for itself out of dividends, and would have to be justified on the businesses instead.

What they returned, and where it came from

The window opens at the first fiscal year end after the August 2020 announcement. It is a proxy for the holding period, not Berkshire's cost basis. Berkshire's own basis is lower and spread across roughly twelve months of purchases in 2019–20 and every addition since, so read this as the experience of the five shares rather than as Berkshire's statement of account.

The CAGR columns are the compound annual growth rate: the single yearly rate that would carry the share from its price at 31 March 2021 to its value at 31 March 2026. It is struck from the two ends and the 5 years between them, so it says what the period delivered and nothing about how steady it was — which matters here, because two of the five have a loss year inside the window.

Price to book Multiple of book value · fiscal years ended 31 March

Price return is the product of three things, exactly: the growth in book value, the reduction in the share count that turns it into growth per share, and the change in the multiple the market puts on it. Dividends are added on top and currency is applied to the result. The first four columns below are percentage points of the yen total return and they sum to it.

Book value is the denominator here rather than earnings, and that is a choice with a reason. The earnings version of the same identity reconciles just as exactly, but it cannot be computed for Sumitomo, which lost money in the year to 31 March 2021 and so has no positive base to grow from. It also inverts Mitsubishi: measured off that depressed year, earnings growth looks like 4.6× and the multiple looks as though it contracted. Equity is positive in every year for every house.

Buybacks have become material and one house dominates them: Mitsubishi spent $13.1 B and retired 17.3% of its shares, against $3.5 B and 5.9% at Itochu.

Where the return came from Percentage points of the 5-year yen total return

Book value growth Shares retired Dividends Multiple re-rating

The currency, and the half of it that is usually missed

The yen went from 110.8 to 158.7 to the dollar between 31 March 2021 and 31 March 2026. For a dollar investor who simply bought the five shares, that is a factor of 0.7× on the whole return — it takes 30.2% off every one of them identically, because it is a property of the currency and not of any company. Quoting the yen figure to a dollar audience without that line is misleading them.

Berkshire is not that investor. Against the yen assets sits yen debt, and a falling yen shrinks the dollar cost of repaying it. The loss on the asset and the gain on the liability are the same move in opposite directions, so the currency washes out to the extent the two match. That is what the funding structure buys: it converts a currency bet into a spread.

How much it washes out depends on how large the yen debt is against the position. The last balance Berkshire has confirmed is ¥1,029 B of parent-company yen senior notes at December 31, 2022 (2022 Form 10-K), and it has issued in each of the three years since, including ¥272.3 B in April 2026. Against a position with a floor value of $42 B, a yen liability of that order offsets a minority of the exposure rather than all of it. The hedge is real and it is partial: the precise fraction is a function of a debt balance Berkshire reports in aggregate rather than currency by currency.

The five, one at a time

Each house gets its own page: the sum of its parts, its segments, how much of its profit arrives from affiliates it does not consolidate, the commodity sensitivities it publishes, and what its own management is committing to. All 5 are written.