Trader recency bias — when only the last trade counts

Consider a hypothetical Marek who feels confident after three winning pound trades and wants to abandon the strategy after two losses. A short sequence can dominate his memory, but it cannot establish that the method is profitable or unchanged. Review the broader record, costs and any market changes. This is one issue covered in the trader psychology section.

What recency bias is and where it comes from

Recency bias means giving disproportionate weight to recent observations. It can interact with availability—the ease with which examples come to mind—but the terms are not identical. A last trade is one observation; its relevance should be evaluated alongside the rest of the record and any change in conditions.

Do not dismiss recent information merely because it is recent. New commissions, wider spreads, execution failures or changed volatility can matter to a strategy. Compare the latest result with the wider record and ask what has changed. There is no fixed number of trades at which an edge necessarily becomes known, and no requirement to keep risking money until that number is reached.

How recency bias sabotages sticking to a system

In a hypothetical independent model, a 45% win probability and wins of +2R versus losses of −1R give expectancy 0.45 × 2 − 0.55 × 1 = +0.35R before costs. Losing streaks are possible, not guaranteed within every finite sample. Estimated probabilities can change and costs can remove the apparent edge.

If uncounted costs average 0.4R per trade, the model’s +0.35R becomes −0.05R net. The assumed +2R/−1R amounts describe outcomes, not merely intended target/stop distances. Under additional independence, five losses in one specified five-trade block have probability 0.55⁵ ≈ 5.03%. That is not the probability of such a run anywhere in a whole year; sample length and dependencies matter.

A streak can fit the model while still exceeding an acceptable loss budget. It can also accompany changing costs or execution, so do not automatically label it noise or proof of failure.

The most dangerous effect — position size drift

Raising risk solely after wins and lowering it solely after losses changes the sizing rule. It can amplify losses on larger trades, but it does not predict when the next loss or rebound will occur. Under independent unchanged outcomes, a recent streak makes neither result due.

With a fixed hypothetical starting base of 10,000 currency units and no costs, three gains of 100 followed by a loss of 300 sum to zero. This assumes the fourth loss; it does not make that outcome due. If percentages instead use changing capital, three gains of 1% give 10,303.01, and a 3% loss is 309.0903, leaving about 9,993.92. State the base before claiming exact cancellation. Planned risk also need not equal actual loss.

How to spot recency bias in yourself

• Record the sizing rule. A reduced position can reflect a drawdown limit, changing capital or volatility. Check its justification rather than diagnose a bias from size alone.
• Check the new information. A fee or execution change may require action before a routine review window ends.
• Keep comparable reports. Record dates, rule versions, costs and exposure; do not duplicate overlapping records as new observations.
• Preserve chronology. Shuffling can alter drawdown, available capital and shutdown rules. It is not a decisive psychological test.
• Separate description and prediction. Weighting recent data can be a stated choice; it does not establish that the recent outcomes will repeat.

After an unusually good week, write down any urge to increase risk and compare it with the sizing rule. Keep the decision record so that a later review can distinguish a planned change from a reaction to recent results.

Concrete defences — separating the base rate from the latest sample

Use the full available record to estimate net expectancy, while checking how the sample was selected and whether conditions changed. A larger sample can improve an estimate; it does not certify future edge. Keep sizing tied to a written risk rule, and distinguish a planned adjustment from a reaction to mood.

Use the journal to assess process separately from a single outcome, recording rule versions, exposure, costs and unsuccessful cases. An arbitrary “no change before fifty trades” rule must not override an operational failure or risk limit. Review your confidence against evidence and assess herd pressure as a separate possible influence; these links do not imply that everyone follows the same causal chain.

Different summaries can also reflect different weights. In a constructed sequence, five results of −1R are followed by five of +1R. The full mean is 0R and the last-five mean +1R. Under the chosen recursion mₜ = 0.5xₜ + 0.5mₜ₋₁ with m₀ = 0, m₅ = −31/32R and m₁₀ = 961/1024R ≈ +0.9385R. Those reports are consistent: they use different weights. The recursion is an illustrative description, not a proven estimator of a future edge.

What to do tonight

Estimate net expectancy from the full available journal and record the sample size, costs and market conditions. Keep any sizing change tied to written rules. A minimum-sample review rule must not override mandatory risk limits or prevent a pause when an execution problem is found.