A brokerage statement is a document about one currency. It reports the change in a position, denominated in whatever the exchange lists it in, and stops there. Everything that determines what the investor actually keeps happens afterwards, in a different currency, and in an order that matters.
Three subtractions, in this order
The kept return can be written as one line:
kept ≈ (1 + r) × (1 + e) − τ × (converted gain − allowance) − π
Here r is the asset's return in its own currency, e is the move of that currency against the one you spend, τ is the marginal rate on the converted gain, and π is inflation in your spending currency. The first two multiply; the last two are taken out of the result, and neither is proportional to what you earned.
Multiplication is where most estimates go wrong. A 20% gain in an asset and an 11% fall in its currency against yours is not a 9% gain: it is 1.20 × 0.89 = 1.066, or 6.6% before tax. On a position of any size, tax then removes a fifth of what survived, and inflation removes a further slice of purchasing power that nobody refunds.
The detail that settles more arguments than the headline rate is this: in every system below the gain is measured in nominal terms. No cost basis is indexed to inflation, so a year in which an asset rises 3% while prices rise 3% produces a taxable gain and a real loss at the same time. The tax is charged on the first and the second is yours.
The rules that decide the number
| Jurisdiction | Rate on a share gain | Annual allowance | Losses | Paid |
|---|---|---|---|---|
| Korea, resident | 22% (20% income tax + 2% local) | ₩2,500,000 | Same calendar year only; no carry-forward | Following May |
| Japan, resident | 20.315% (15% + 0.315% reconstruction surtax + 5% local) | None | Carried forward 3 years, with a filing every year | Following February–March |
| United Kingdom, resident | 18% within the basic-rate band, 24% above | £3,000 | Carried forward indefinitely against future gains | Self assessment |
| United States, resident | 0%, 15% or 20% by taxable income, plus 3.8% net investment income tax above $200,000 single / $250,000 joint | None (graduated brackets instead) | Against gains, then $3,000 a year against other income, remainder carried forward | Following April |
| United States, non-resident holding US shares | Generally none | — | — | No US return, unless US real property or a US trade or business is involved |
Read the column that is not the rate. Korea's 22% sits close to Japan's 20.315%, a difference of 1.7 points, but the loss rules are opposites: a Korean investor whose overseas position falls in a bad year cannot carry that loss into the next one, while a Japanese investor can carry it for three years. For a volatile portfolio that difference is worth more than the rate. It also changes behaviour — a Korean account has a reason to sell losers in December and a Japanese account does not.
Britain shows a quieter version of the same point. The rate for shares went to 18% and 24%, and the annual exemption fell from £12,300 in 2022–23 to £3,000. A £30,000 gain that once left £17,700 taxable now leaves £27,000 taxable: at 24% the bill goes from £4,248 to £6,480, a 53% increase with no change in the quoted rate. Allowances are fixed sums; portfolios are not.
The exchange rate is inside the tax base
The most consequential mechanical detail sits in how a gain is converted. In Korea, when a resident sells a foreign share, the proceeds are converted to won at the first rate quoted on the day the money arrives, and the cost is converted at the rate on the day the money left to buy the share. The taxable gain is the difference between two won amounts struck on two different days. The currency's move is not an adjacent line item; it is inside the gain.
Two consequences follow. If the won strengthens while you hold a dollar asset, part of your dollar gain is taxed before you ever receive it. If the won weakens, you can owe tax on a won gain that never existed in dollars. And the symmetry is not a comfort: a dollar loss and a won gain are both possible, and it is the won figure that is charged.
Wherever the gain is computed in the tax office's currency rather than the asset's, the same structure appears. The rate on the day of the transaction — not the rate you saw on your screen, and not the average rate over the year — decides the taxable amount.
Four and a half months, priced
Take a concrete case with published rates at both ends. A Korean resident invests $100,000 in early July 2026, when the won was still trading above 1,500 to the dollar, and sells on 13 November 2026, when the base rate was 1,332.4. No other overseas gains that year, so the full deduction applies.
| Dollar return | Position at sale | Value in won | Gain in won | Korean tax | Kept |
|---|---|---|---|---|---|
| −10% | $90,000 | ₩119,916,000 | −₩30,084,000 | 0 | −20.1% |
| 0% | $100,000 | ₩133,240,000 | −₩16,760,000 | 0 | −11.2% |
| +10% | $110,000 | ₩146,564,000 | −₩3,436,000 | 0 | −2.3% |
| +20% | $120,000 | ₩159,888,000 | +₩9,888,000 | ₩1,625,360 | +5.5% |
| +25% | $125,000 | ₩166,550,000 | +₩16,550,000 | ₩3,091,000 | +9.0% |
The second row is the one worth reading twice. With no gain at all in dollars, the position lost 16.8 million won, and in Korea that loss cannot be used against next year's gains — only against other overseas gains in the same calendar year. The fourth row shows the decomposition: of a 20% dollar gain, the currency took 13.4 points and the tax took 1.1. The asset performed; the crossing cost.
The middle columns are the ones that travel. A reader under another regime keeps the same "value in local currency" column and swaps only the rule: £3,000 of exemption and 24% in Britain, no allowance and 15% for a US resident, NISA in Japan. What does not change is the point of the exercise — the currency is charged first, and the tax is charged on what is left.
Note also what the deduction is worth. ₩2,500,000 is 1.7% of a ₩150 million position and 0.5% of a ₩500 million one. It is an allowance shaped for a retail account, in a market where retail accounts now hold close to $200 billion of US shares.
Where the money is going, and why it is the same story
The flow data and the arithmetic describe one event from two sides. Korean investors held $197.5 billion of US shares in the first three weeks of September 2026, down from a May peak of $204.2 billion, then bought $5.42 billion net in October and $2.198 billion in the first five trading days of November alone — as the domestic index fell and the won strengthened from the 1,500s to the mid-1,300s, making each dollar cheaper in won.
That is rational at the point of entry and painful for the stock already invested. New money buys dollars at 1,330; money that bought at 1,500 is carrying an 11% currency loss on top of whatever the asset did. Because tax is due on realisation, the loss stays unrealised and therefore uncounted — until a year arrives when the investor wants to sell something else, and in Korea discovers that a loss from a different year cannot help.
The tax code also prices the same gain differently depending on where the exchange is. A Korean resident's domestic listed shares are not taxed at all for a minority shareholder trading on-market; the same resident's overseas shares are taxed at 22%. Japan answers the same question with a shelter instead of an exemption: NISA allows ¥3.6 million of new investment a year, about ₩30 million at 13 November rates, with gains and dividends exempt — but a loss inside NISA is treated as if it never happened, so the door only opens one way. Britain uses a small exemption and separate ISA wrappers; the United States uses graduated rates plus tax-advantaged accounts. In each case the direction of capital follows the shape of the rule, not only the return.
One portfolio, two returns
For a reader paid in dollars and spending in euros, yen or won, 2026 has been a lesson in which return counts. The dollar index stood at 103.2 on 13 November, its highest since April 2025, with the Federal Reserve expected to raise rates in December after US consumer prices rose 0.4% in October — the third consecutive month at or above that pace, and with no year-on-year figure published because the 2025 shutdown left the bureau without a base. The 10-year Treasury yielded 5.12%.
The cross rates from the same day's Seoul quotations tell the rest: with the dollar at ₩1,332.4, the euro at ₩1,472.97 and the pound at ₩1,763.5, the implied EUR/USD is about 1.105 and GBP/USD about 1.324, while 100 yen at ₩833.45 puts the dollar near ¥160. A euro-based investor holding the same dollar portfolio as the Korean investor above has a different kept return, because their second factor is EUR/USD rather than USD/KRW. The identity is unchanged; only e moves.
And e is decided after the fact. A reader whose tax residence is in one currency and whose assets are in another is, in effect, taxed on a number that will not be final until the year ends and the rate is struck.
What the arithmetic asks of you
Write the kept return, not the reported one. Four lines — fees, currency, tax, your own inflation — turn a statement into a decision, and if the total is not positive, no headline figure rescues it.
Know which currency your liabilities are in. A gain measured in a currency you never spend is a fact about a currency, not about your life. A deposit, school fees, a parent's care: those are the units the return has to be measured in.
Read the loss rule before the rate. The rate tells you what a good year costs. The loss rule tells you what a bad year costs, and bad years are when the decisions get made.
Treat an allowance as a fixed sum. It does not grow with your portfolio or with prices, and it is re-set by legislation, usually downward.
What this does not say
It is not advice, and it is not a forecast of the won, the euro or the Fed. Rules move more often than the arithmetic suggests: in January 2027 Japan's 2.1% reconstruction surtax becomes a 1.1% surtax plus a new 1% defence surtax, which changes the label on the 20.315% and leaves the number where it is. Korea's filing window shifts around the calendar each May. The rates above carry the date they were checked; a figure quoted from an older page may already be wrong.