Recency bias
Also called extrapolation bias, over-extrapolation
Giving too much weight to the latest events or returns and assuming they will continue, for example expecting a market to keep rising because it has risen recently.

On this page· 5 sections
The last few weeks of a chart are the easiest part to remember and the hardest part to ignore. They are also a small sample.
Recency bias is what happens when that small sample sets the forecast. A good month starts to look like the new normal; a bad one, like proof that something has broken.
How it works
Part of the cause is memory. Tversky and Kahneman noted in 1974 that recent occurrences are likely to be more available to the mind than earlier ones[3]. Their everyday example: the felt probability of a traffic accident rises for a while after seeing an overturned car by the road[3].
The other part is inference from too little data. The FCA calls it over-extrapolation: making predictions from a few observations while assuming they are representative and show a real pattern, and so underestimating uncertainty[1]. Its investing examples include predicting future stock returns from a few past returns and overestimating an investment product’s future return because of its past performance[1].
Data on investors’ expectations shows the pattern at the scale of a whole market. Robin Greenwood and Andrei Shleifer studied investor expectations of future stock market returns from six data sources between 1963 and 2011[2]. The measures moved together, and they were positively correlated with past returns and with the level of the market[2]. They were also strongly negatively correlated with model-based expected returns[2]. In plain terms, investors expected the most after prices had risen, when the models expected less[2].
Example
Say a trader’s method wins half its trades, and each result is independent, like a coin toss.
- In 100 trades, a streak of four or more wins in a row is expected about 3 times.
- The chance of seeing at least one such streak is about 97%.
- Streaks of four or more losses are equally likely.
After a winning streak, recency bias says the method has improved. After a losing streak, it says the method has stopped working. In this example neither streak says anything about the next trade, because nothing about the method changed.
Watch out
Recent is not the same as representative. The FCA notes that people judge advice as good on the strength of a few successful investments that may reflect pure luck[1].
It runs in both directions. Expectations that track past returns rise after gains and fall after losses[2].
A shorter look-back is not always wrong. Conditions do change; the bias is in treating a few observations as proof that they have[1].
Sources
- RegulatorFinancial Conduct AuthorityApplying behavioural economics at the Financial Conduct Authority (Occasional Paper No. 1, April 2013)Checked November 9, 2026
- ResearchOxford University Press (Review of Financial Studies)Expectations of Returns and Expected Returns (Greenwood & Shleifer, Review of Financial Studies 27(3), 2014)Checked November 9, 2026
- ResearchUniversity of Washington (course copy of the journal article)Judgment under Uncertainty: Heuristics and Biases (Tversky & Kahneman, Science 185(4157), 1974, pp. 1124-1131)Checked November 9, 2026
