Recency Bias Trading: A Teardown of How Five Trades Resize Your Risk

By Sofia Harchich | Trading Psychologist & Behavioural Finance Writer | thewealthmirror.com

Recency bias trading is easiest to detect not in what a trader thinks, but in the number they type into the size field.

Ask most traders whether four winners in a row makes the fifth trade more likely to work, and the answer is immediate and correct: no. Then look at the size of the fifth trade. The knowledge and the behaviour are stored in different places, and only one of them reaches the order ticket.

This is worth taking apart properly, because the usual treatment — a description of the recency effect, followed by an instruction to zoom out and look at more data — addresses a stage of the mechanism where nothing is going wrong.

The mechanism in four moves

Move one: unequal encoding. The most recent items in any sequence are recalled more readily than those in the middle. This is the serial position effect, it is one of the older findings in experimental psychology, and it is not a defect. A system that weighted last week’s information identically to information from three years ago would be a considerably worse system for staying alive.

Move two: fluency read as frequency. The availability heuristic does the damage. Because the last four trades come to mind quickly, the ease of recall gets read as evidence about how often this outcome occurs. Nothing in the experience distinguishes I can remember this easily from this happens often, and the second is what the trader ends up believing.

Recency bias trading is often described as though it stopped at move two. It does not.

Move three: the belief updates. A conclusion forms, usually unspoken: the setup is working at the moment. The market is in a phase that suits me. Or, after losses: something has changed and I am reading it wrong.

Move four: the size changes. Not the strategy, which the trader would notice and question. The size, which is rarely written down as a decision at all.

The first two moves are memory functioning as designed. The instruction to zoom out is aimed at those two, which is why it fails so consistently: you cannot correct an encoding property by resolving to encode differently.

Where recency bias trading enters the loop

The underlying circuit runs in five stages. Trigger, emotional activation, reactive behaviour, outcome, reinforced belief — and back to trigger.

Recency bias trading does not enter at stage one. It enters at stage five, and this is the whole diagnostic point. The trigger is not distorted; four winners genuinely happened. The activation is proportionate. The behaviour, at the moment it occurs, is coherent given the belief. What has gone wrong is that the outcome updated the belief far more than the sample size justified, and the updated belief now sets the threshold for the next cycle.

That makes it a belief-updating problem wearing the costume of a memory problem, and it changes what will work. Beliefs update on evidence; the fix is to change what counts as evidence for a sizing decision, before the sizing decision is being made.

The size of the next trade usually says more about the last five trades than about the next five.

The loop in question is the instability loop, and recency bias enters it at step five: the last five outcomes harden into a belief about the next five, and that belief resizes the position before the setup has been read.

The step that actually breaks

Installing a change between trigger and behaviour has an order: detect, regulate, decide, execute, enforce.

Here the break is at decide — specifically, the absence of a decision protocol for size. Most traders have written entry criteria, written invalidation, and no written rule at all governing how large the position is. Size is improvised each time, under load, by a person whose most recent results are the loudest thing in the room.

That absence is not carelessness. Size feels like the flexible, discretionary, skill-expressing part of trading, and treating it as fixed feels like giving something up. What it actually gives up is the only variable recency bias has access to.

Detection is not the failing step here, which is why so many traders can name this bias precisely and still exhibit it. Naming a bias is a step-one competence. The gap between seeing it and acting on it — the awareness–action gap — stays wide until something changes at step three.

The test that separates bias from adaptation

There is a legitimate objection to all of this, and it deserves a real answer rather than a dismissal: sometimes conditions genuinely change, and adjusting exposure to changed conditions is competence, not bias.

True. So the two need separating, and they separate cleanly.

Bias: size varies with recent results. Adaptation: size varies with measured conditions — realised volatility, spread, session, the distance to invalidation.

The test is falsifiable and takes one spreadsheet. Log position size, the previous three trades’ outcomes, and one volatility measure for every trade over a month. Then look at which column size actually tracks.

If size correlates with recent results and not with volatility, the mechanism above is running. If it correlates with volatility and not with results, it is adaptation and there is nothing here to fix. If it correlates with both, the separable part is still worth fixing.

Most traders are confident they know which one they are doing. The distribution frequently disagrees, and it disagrees in a specific direction: size expands after wins more readily than it contracts after volatility increases.

Why the losing-streak version is more expensive

The winning-streak version of recency bias trading gets all the attention because the failure is spectacular — an oversized position into the end of a favourable run.

The losing-streak version does more damage and attracts none, because it looks like discipline from every angle. Three losses, size cut to a quarter, and the trader receives internal credit for prudence. What follows is a period in which the strategy’s actual edge is being expressed at a fraction of its intended exposure, which amounts to taking the strategy’s losses at full size and its wins at reduced size, decided by recent memory rather than by anything written down.

It also feeds a second circuit. Reduced size produces reduced returns, which confirms something has changed and I am reading it wrong, which justifies further reduction. The belief that lowered the size is strengthened by the consequences of having lowered it — the same stage-five loop, running in the direction nobody flags because it feels responsible.

The rule that survives contact

The replacement is a decision protocol at step three, written when nothing is at stake.

1. Size is a function of account risk and distance to invalidation, and of nothing else
2. It is written down before the session rather than chosen at the ticket, which is where recency bias trading gets its access
3. Any deviation requires a stated reason logged at the time — not afterwards, when the reason will be constructed
4. Recent results are explicitly excluded as an input, by name, in the written rule
5. Volatility-based adjustment is permitted and must reference a measure, not an impression

Point three is the one that does most of the work. A deviation that has to be written down while it is happening drops sharply in frequency, without any rule against deviating — because the writing is a step-one intervention placed in front of a step-three failure, and it is the only moment in the sequence when detection is still fully available.

The second target: the strategy itself

Sizing is the fast expression of recency bias trading. Strategy abandonment is the slow one, and it costs considerably more.

The sequence is familiar. A run of losses, an accumulating sense that the edge has stopped working, a search for something better, and a replacement adopted at roughly the point where the original system’s recovery was statistically due. Nobody experiences this as bias. It is experienced as adapting to a market that changed.

What settles the question is arithmetic almost nobody runs.

A strategy with a genuine 55% win rate does not produce alternating wins and losses. Simulated across 100 trades, it produces a longest losing run of about five, with a roughly one-in-five chance of a run of seven or worse. At 50% the typical longest run is six. At 40% — entirely viable with a large enough average win — the typical longest run is close to eight, and a run of ten or more turns up about one time in five.

Win rate	Typical longest losing run in 100 trades	Chance of a run of 5+	Chance of a run of 7+
40%	8	97%	69%
50%	6	81%	32%
55%	5	65%	18%
60%	5	46%	9%
65%	4	29%	4%

Find the row matching your system. The losing run that persuaded you to abandon it is very likely sitting inside the range that system produces while working exactly as designed.

This is where the availability heuristic does its most expensive work. Six consecutive losses are vivid, recent, and effortless to recall. The other ninety-four trades are none of those things. Ease of recall gets read as evidence about the edge, and the evidence appears to say the edge has gone — when six-loss runs turn up in simulations of profitable systems roughly one time in three.

The correction is a number written down before the drawdown begins: the losing run length that would constitute genuine evidence against this system, derived from its win rate rather than from how the run feels at trade five. For most systems that number is markedly larger than the trader expects, and having it in advance is what makes it possible to sit through a run the system was always going to produce.

One caveat, because the arithmetic can be overread. All of it assumes the win rate is known. A trader with fifty trades of history does not know their win rate to within ten percentage points, so the table describes what randomness can produce rather than granting permission to hold a system indefinitely. For most traders both the edge and the drawdown are uncertain at once, which is an argument for smaller size rather than for greater patience.

Three markers in the record

Recency bias trading leaves the same three traces in almost every journal, and none of them mention memory.

• Size variance without a stated reason. Two entries a fortnight apart on the same setup, at different sizes, with no note explaining the difference. The explanation existed at the time and was not written down because it did not feel like a decision
• Timeframe language after a run. Phrases like conditions have shifted or this setup has stopped working, appearing within five trades of a drawdown and absent from any pre-drawdown entry
• Asymmetric recovery. Size returning to normal within one or two trades after a win, and taking eight or more after a loss of the same magnitude

The third is the most diagnostic, because it cannot be explained by adaptation. A genuine read on changed conditions would move size symmetrically in both directions. A read on recent results does not.

What the teardown does not explain

Two honest limits.

It does not explain why the effect is stronger on some instruments than others for the same trader. Instrument-specific activation is real — a trader can hold sizing discipline across a book and lose it on one symbol, usually the one carrying the most personal loading — and this mechanism does not account for that. It sits on a different layer, closer to what money means to that person than to how memory weights a sequence.

And it does not explain the trader who runs the test, finds the correlation, and changes nothing. That is not a recency problem. That is the awareness–action gap, and it is addressed in what is trading psychology rather than here.

Why the knowledge sits so far from the behaviour

Nobody reading this believes four winners make a fifth more likely, which is what makes recency bias trading so persistent. That was never the failure.

The failure is that the belief lives in a part of the system that does not have write access to the size field, and the part that does have write access is updating on whatever happened most recently, at a speed the evidence does not justify. Recognising this is a reason to move one decision — how large — out of the moment when it is made worst and into the one when it is made best, and not a reason to think less of your discipline.

For the wider set of biases this belongs to, start with behavioural finance for traders, and with confirmation bias in technical analysis for the bias that most often defends the position this one sized. The pattern quiz identifies which emotional pattern your behaviour under pressure fits.

---

About the Author

Sofia Harchich is a Trading Psychologist and Behavioural Finance Writer with a Master’s in Psychology. She works at the intersection of Jungian shadow work, neuroscience, and market behaviour — helping traders understand the psychology driving their decisions, not just the strategy.

Read more at thewealthmirror.com/about