交易中的认知偏差：为什么大脑会对抗你的计划

Your trading plan has an opponent, and it is not the market. It is the standard-issue decision machinery every human brain ships with — a set of shortcuts that served our species well around scarce food and visible predators, and that misfire reliably around leveraged positions and flickering quotes. This sheet covers the five misfires with the largest documented cost to traders. The premise is honest and slightly uncomfortable: reading about them will not fix them. Knowing the name of a bias has roughly the effect on the bias that knowing the name of gravity has on falling.

Firmware, not character flaws

Decades of decision research — the same experiments, replicated across cultures, professions, and levels of expertise — converge on an awkward finding: these patterns are not what stupid people do, they are what people do. Experienced traders exhibit them; researchers who literally wrote the studies report exhibiting them. They fire before conscious reasoning gets involved, which is why they feel, from inside, like judgment rather than error.

That origin dictates the repair strategy, and it is the actual thesis of this sheet. You cannot patch firmware with awareness, but you can route around it with procedure — rules made in advance, executed regardless of how the moment feels. Keep that in mind as each bias is introduced below, because otherwise the list reads as a self-diagnosis quiz, and self-diagnosis is itself performed by the instrument being diagnosed.

Loss aversion: why stops move and winners get cut

In measured choices, losses weigh roughly twice as much as equal gains — losing $200 hurts about as much as winning $400 pleases. Now put that asymmetry in front of a price feed. A position showing $200 of ▲unrealized profit feels like a win that could still be taken away, so the urge is to close it and make it safe. A position showing $200 of unrealized loss feels unfinished — closing it would convert a maybe into a fact, and the asymmetry makes that fact feel like $400. Same number, opposite urges.

The resulting behaviour is the most documented pattern in retail trading records: winners closed early, losers held long — sometimes inverted into moving the ▲stop-loss further away as price approaches it, which converts a planned $200 loss into an unplanned $400 one. Note the perfect symmetry of the instrument and the perfect asymmetry of the response: the market offers identical odds on both sides of your entry; the brain prices the two directions differently and trades the mispricing — against you.

Part of the trap is linguistic. The platform labels an open loss “unrealized,” and the word does exactly what loss aversion wants: it suggests the loss is not yet real. It is real — your ▲equity, the number a margin calculation actually uses, already includes it tick by tick. Closing changes nothing about the amount; it only ends the story the brain was telling about a comeback. Traders who internalise that one accounting fact lose a surprising amount of the urge to hold and hope.

Confirmation and recency: curated evidence

Confirmation bias is the analyst's failure mode: once you want the trade, research quietly becomes advocacy. Five charts get opened and the one that agrees gets kept; the indicator contradicting the entry is re-parameterised until it stops; the headline supporting the position is read twice and the one against it is skimmed. The output looks like analysis — it has charts and reasons — but the conclusion was chosen first and the evidence hired afterwards.

Recency bias compounds it by shrinking your dataset to whatever just happened. Three breakout trades win, so breakouts “work now”; a strategy has two losing weeks, so it is “broken” — verdicts rendered on samples of three and ten, by a brain that experiences the recent past as more real than the statistics. Together the pair manufactures conviction: recency supplies a fresh, vivid hypothesis and confirmation guards it from contrary evidence. Conviction built this way feels identical, from inside, to conviction built from a hundred journalled trades. The feeling does not distinguish its sources — only a record does.

There is one cheap counter-procedure worth installing immediately: before entering, write a single line stating what would prove the trade wrong — the price level, the failed condition. Confirmation bias can defend a position against evidence it gets to evaluate live; it has a much harder time against a sentence you wrote before you had a position to defend. That line later becomes the journal's most informative field.

Sunk cost: averaging into losers

The sunk-cost fallacy is the refusal to re-decide. The rational question in front of any open position is always the same: knowing what I know now, would I open this trade at this price? If no, close it. The fallacy answers a different question — “I have already lost $80 on this idea, and closing makes the $80 permanent” — and so the losing idea is kept, and often fed.

Feeding it has a precise arithmetic. Say you are long 0.2 lots of EUR/USD from 1.0900 and price sits at 1.0860: minus 40 pips, minus $80. Adding another 0.2 lots at 1.0860 drops your average entry to 1.0880 — break-even is suddenly half as far away, which is the whole psychological appeal — but you now hold 0.4 lots moving $4 a pip in both directions. If price recovers 20 pips, you are flat instead of down $40: the averaging looks brilliant. If it falls another 40, you are down $240 instead of $160. The technique does not change the trade's odds at all; it doubles the stakes on the idea you already have the most evidence against, at the moment you are least able to weigh that evidence.

Overconfidence: the post-streak size creep

Overconfidence has a favourite entrance: the winning streak. Five good trades in a row and the brain quietly reclassifies luck as skill — the analysis felt sharper, so the next position can surely be larger than the plan allows. But five consecutive wins arrives by pure chance about once in every 32 five-trade runs for a coin; a modest real edge makes it common. The streak is weak evidence of skill, and strong temptation precisely when discipline has the most to protect: size creep after wins means your largest positions are systematically placed at your most confident moments, which are not your most accurate ones.

The same worked numbers as ever: a plan that risks $50 a trade survives any ordinary losing streak. The post-streak version risking $150 “because it's working” gives back three wins with one loss — and the symmetric upside (three extra wins of $150) is exactly the lure that makes the oversizing feel justified. The mechanism pays identically in both directions; the difference is that the plan priced one of them and the mood priced the other.

The procedural fixes

If awareness fixed biases, reading this far would have done it. The studies on debiasing are blunt about this: warnings and education change measured behaviour barely at all, while changed procedures change it durably. So the fixes below are not insights to remember — they are mechanical changes to when and how decisions get made, which is the only door the firmware does not guard. Three procedures cover the five biases:

• A pre-entry checklist — setup matches a named plan entry; size comes from the sizing rule, not from conviction; stop and target are placed with the order. Loss aversion cannot move a stop that was committed at entry, and confirmation has to argue with a written setup definition instead of a mood.
• Pre-commitment — daily loss limits and trade caps decided in advance (the circuit breakers from the ▲overtrading sheet), plus orders that execute without you: the stop in the market is a decision your calmer self already made.
• Journal review — the only debiasing tool that uses your own data. Recency says the strategy is broken; the journal says it is two losing weeks inside a profitable quarter. Overconfidence says size up; the journal shows what happened the last three times size drifted above plan.

None of these procedures requires you to feel unbiased — that is the design. They work like the checklist in an aircraft cockpit: not because pilots are forgetful, but because procedure executes under exactly the conditions in which perception fails. The plan sheet and the journal sheet in this region of the survey are the two halves of that cockpit, and this sheet is the reason they exist.