What is recency bias?

The bias has three components worth separating, because they fail in different ways:

• Weighting. Recent observations are given more influence over a judgment than their share of the evidence justifies, while older observations are treated as background rather than data.
• Extrapolation. The recent run is projected forward as a trend, on the assumption that whatever has been happening lately describes the underlying system rather than a sample of it.
• Reference shift. The recent period quietly becomes the yardstick — "normal" volatility, "normal" returns, "normal" risk all get redefined around the last few months of experience.

In plain terms: recency bias doesn't mean recent information is worthless. Recent data is often the most relevant data. The error is letting it be the only data, particularly when the sample is far too small to say anything.

Recency effects were first documented in memory research. Studies of serial recall consistently find that people remember items from the end of a list far better than items from the middle — the recency effect, one half of the classic serial position curve. What is easiest to retrieve is not what is most important; it is simply what is closest to hand.

Kahneman and Tversky's work on the availability heuristic explains why that matters for judgment: people estimate how likely or how typical something is by how easily examples come to mind. Because recent events are the easiest of all to recall, ease of retrieval gets mistaken for frequency, and the recent past starts standing in for the whole distribution.

A related finding, the law of small numbers, describes the second half of the problem: people expect small samples to resemble the population they came from, so a short run of similar outcomes reads as a pattern long before it is statistically distinguishable from noise. Recency bias is what happens when those two tendencies operate at once — a small, vivid, recent sample treated as a reliable description of the future.

Further reading: the serial position effect and recency in memory research.

The pattern is easiest to see in someone else's decision. These are the shapes it takes most often:

• Trading

  Three losing trades in a row lead to cutting position size in half, even though the strategy's historical record shows streaks of that length are entirely normal and carry no information about the next trade.

• Investing

  Capital moves into whichever fund or sector topped the last twelve months of performance tables, treating a recent stretch as evidence of durable quality rather than as a sample that often mean-reverts.

• Market outlook

  After a long bull run, risk starts to feel abstract and historical drawdowns feel like relics. After a crash, the same investor assumes further declines are the default, because the most recent regime feels like the permanent one.

• Startups and business

  One strong month of sales resets the forecast upward and headcount follows, while two years of steadier, lower numbers are treated as the past rather than as the base rate.

• Hiring and management

  A performance review turns mostly on the last few weeks of work — the recent project, good or bad — rather than on the full period the review is supposed to cover.

• Risk decisions

  Insurance, hedges, or stops get taken seriously immediately after a loss event and quietly dropped once enough uneventful months have passed, tracking recent experience instead of actual exposure.

Three forces keep it in place. Retrieval cost: recent events are cheap to recall and older ones are expensive, so the mind reaches for what is available and treats availability as relevance. Emotional vividness: the last loss still carries feeling, while a loss from two years ago is only a number, and feeling gets more weight than arithmetic when a decision is made quickly. Pattern-seeking: humans are unusually good at detecting trends and unusually bad at rejecting false ones, so three similar outcomes register as a regime change rather than as ordinary variance.

This is why "keep the long view" rarely works as advice on its own. The long view is not psychologically available in the moment — the recent view is. Correcting the bias means making older evidence as retrievable as recent evidence, which is a record-keeping problem before it is a discipline problem.

In a trading account, recency bias shows up as strategy abandonment and inconsistent sizing. A run of losses well within the strategy's historical norms triggers a change of approach, so the process never gets a sample large enough to prove or disprove itself. A run of wins triggers the opposite: size increases at exactly the point where the recent sample, not the underlying edge, is doing the persuading.

For investors it drives performance chasing — buying whatever led the last twelve months and selling whatever lagged, which is close to a systematic way of buying high and selling low. It also resets the perception of risk after long calm stretches, which is why exposure tends to peak shortly before it matters most.

It compounds with other biases too. A recent surge is what makes FOMO and herding feel justified, and a recent winning streak is one of the most reliable inputs to overconfidence.

You cannot make recent events less vivid, but you can build structure that puts older evidence back in front of you before the decision gets made:

1. 1

   Look up the base rate before the recent result

   Before judging a strategy, fund, or decision by its last stretch, write down the long-run number first — win rate, average return, typical drawdown. Deciding the base rate before you look at the recent sample stops the recent sample from setting the reference point.

2. 2

   Fix the sample size in advance

   Decide how many trades, months, or data points constitute a fair evaluation window, and refuse to draw conclusions before you have them. Most streaks that feel meaningful are shorter than any sample capable of proving something.

3. 3

   Keep a written record you actually re-read

   Recency bias operates on memory, and memory is what a log replaces. A decision journal makes older evidence as retrievable as this morning's, which removes the mechanism the bias depends on.

4. 4

   Separate the streak from the process

   Ask whether the recent outcomes were produced by a change in your process or by ordinary variance. Only the first is a reason to adjust; the second is the cost of running any strategy with an edge below 100%.

5. 5

   Pre-commit to rules while calm

   Write position sizing, entry, and exit rules in advance, and require a fixed review cadence — not a losing week — to change them. Rules written before the streak are the only ones the streak cannot rewrite.

6. 6

   Watch your own language

   Phrases like "lately it keeps," "the last few times," "it's been working all month," or "this market always does this now" are markers that a recent sample, not the full record, is carrying the argument.

• Why day traders lose money

  The hub piece: how a recent streak gets mistaken for skill, alongside the six other biases behind the day-trading failure rate.

• Hindsight bias

  The most recent outcome is also the one most likely to be remembered as inevitable, which turns a small sample into a confident lesson.

• FOMO / herding bias

  A recent surge supplies the visible momentum that makes following the crowd feel urgent and justified.

• Overconfidence

  A short winning streak read as evidence of skill is recency bias converting directly into excess certainty.

• Loss aversion

  A recent loss makes the next risk feel larger than it is, which distorts sizing and exits.

• Confirmation bias

  Once a recent trend has set your view, evidence that the trend continues gets favoured over evidence it won't.

The Bias Checker screens for all of them at once, and the articles go deeper on each.