相关性与隐藏敞口：当三笔交易实际上是同一笔

Here is a portfolio that looks diversified: long EUR/USD, long GBP/USD, short USD/JPY. Three pairs, three charts, three separate trade ideas — and, underneath, largely one position: short the US dollar, at triple the size you intended. This sheet teaches the audit that reveals it. The method is currency-leg accounting, the tool is a small table, and the habit takes about a minute per portfolio. It is the difference between risking what you decided to risk and risking three times that without noticing.

The audit: three trades, decomposed

Start with what a pair trade actually is. Buying EUR/USD is two simultaneous positions: long euros, short dollars. Buying GBP/USD is long pounds, short dollars. Selling USD/JPY is short dollars, long yen. Write the three trades that way and the pattern is already visible — the dollar appears on the short side of every one. The euro, pound, and yen legs are each unique; the dollar leg repeats three times.

This is hidden ▲exposure: risk that exists in the portfolio but on no single chart. Each trade can be perfectly reasonable in isolation. The problem only appears in the sum, which is why traders who never sum never see it — until a single piece of US news moves all three positions at once and the day's loss is triple the plan.

The trap is built into how platforms present positions: a list of tickets, each with its own chart, its own profit or loss, its own stop. Nothing on the screen sums the legs. The interface shows three trades because you clicked three times — whether you made three decisions, or one decision three times, is a question only the table below can answer.

Net exposure accounting: the table that tells the truth

The audit itself is one small table. For each open position, write the long leg and the short leg with rough sizes — one standard lot of EUR/USD at 1.0850 is long €100,000 and short about $108,500; one lot of GBP/USD near 1.2700 is long £100,000 and short about $127,000; one lot short USD/JPY is short $100,000 against yen. Then net each currency across the rows.

The net row is the honest description: this trader is short roughly a third of a million dollars. Suppose each trade was sized to risk 1% of a $10,000 account — $100 behind its stop. A broad dollar rally pushes all three positions toward their stops together, and the realistic bad day is −$300, not the −$100 the per-trade rule promised. The symmetry holds in full: a broad dollar decline pushes all three toward their targets together, and the good day is similarly tripled. The cluster wins as one and loses as one. Nothing about the stacking is secretly unfair — what failed was only the label “three independent trades.”

Correlation in plain terms

▲Correlation is the statistician's name for what the table just showed. It runs from +1 (two prices move together) through 0 (no relationship) to −1 (they move opposite). EUR/USD and GBP/USD commonly show daily correlations around +0.8 to +0.9 — most days, they close in the same direction, because the shared dollar leg dominates both.

What does +0.9 do to your risk? If three $100 risks were truly independent, statistics would soften their sum — the typical combined swing is near $173, not $300, because losers and winners partly cancel. At +0.9 that softening almost disappears: you carry close to the full $300, while the position list still looks like ▲diversification. The spread of tickets across pairs changed nothing about the concentration of risk. Two caveats keep this honest: correlation is measured over a window and drifts as conditions change, and high correlation does not mean identical — GBP/USD can fall on UK news while EUR/USD stands still. The audit above needs no statistics at all; correlation simply explains why the legs you summed behave as one.

Where clusters hide

The dollar is the most common shared leg, because it sits inside every major pair — but it is not the only repeat offender. Currency families move on shared drivers, and a portfolio can stack a theme without the dollar appearing once. The habit to build is reading every new trade as its two legs and asking which family each leg belongs to:

• The dollar bloc: every major shares the USD leg — the cluster in this sheet's worked example.
• Commodity currencies: AUD, NZD, and CAD lean on raw-material prices and global demand; long all three against anything is one bet on the same weather.
• Safe havens: JPY and CHF tend to strengthen together when risk appetite breaks — short both is a single quiet bet that calm continues.
• Crosses with a shared leg: long EUR/JPY plus long GBP/JPY is twice short the yen, with no dollar in sight.

Risk per theme, not just risk per trade

The fix is a second budget. Keep your per-trade rule — say 1% behind every stop — and add a cap per theme: the total risk across all positions sharing a driver (short-dollar, long-yen, risk-appetite) gets no more than what you would grant a single conviction, perhaps 2%. Under that cap, our three trades cannot all run at $100; either two of them shrink, or one idea is simply not taken because it is already on in another costume.

In practice: before adding any position, run the leg table including the new trade. If the new trade deepens an existing net leg, size it as an addition to that theme, not as a fresh idea — take the theme's remaining risk budget and derive the size from it with the ▲position size calculator, exactly as you would for one trade. Some platforms show a correlation matrix; useful, but the five-row table is the tool you will actually use, because it needs nothing but the positions themselves.

Two practical notes. First, the theme cap is a budget, not a target — most weeks, most themes should sit far below it. Second, offsetting combinations need the same audit in reverse: long EUR/USD and short GBP/USD partly cancel through the shared dollar leg, so the pair of trades is smaller than its tickets suggest, and a full-size stop on each is pricing risk the combination does not actually carry. The table works in both directions; hidden concentration and accidental cancellation are the same bookkeeping error with opposite signs.

The stress clause: correlations converge when it matters

One more property, and it is the one that bites. Correlations are not stable, and their drift is not random: in calm markets they relax, and in stressed markets they tighten. When risk appetite collapses — the risk-off regime this survey's sheet on risk-on and risk-off describes — money moves by theme, not by pair, and positions that coexisted peacefully for months start moving as a single block. Measured diversification is at its weakest precisely on the days you bought it for. History keeps receipts: in late 2008, and again in March 2020, pairs that had spent years loosely related fell into lockstep within days as positioning unwound everywhere at once. Nobody needs the dates memorized — only the pattern. Stress does not create new exposures; it reveals the ones the ticket list was hiding.

The practical conclusion is conservative: treat measured correlation as a fair-weather reading and assume it strengthens under stress. A theme cap that looks slightly paranoid in a quiet month is correctly sized for the loud one. ▲Hedging — deliberately adding an offsetting leg — has its own sheet and its own costs; the cheaper discipline is simply refusing to stack the same bet three times by accident.