Recency Bias

Recency bias means favoring the last things you saw or heard when deciding. It makes recent items feel more important than older ones, even if they are not.

Recency bias is the tendency for more recent observations to disproportionately influence judgments and decisions due to temporal weighting in memory and processing. This effect arises from differential accessibility and decay of stored evidence, producing systematic skew toward recently encountered information.

A hiring manager interviews ten candidates over two weeks. Even though an early candidate gave the best answers, the manager ends up recommending the last two people interviewed because their responses feel freshest and easiest to recall when writing up the evaluation.

A portfolio risk model retrained on a 90-day rolling window prior to a low-volatility period assigns near-zero weight to volatility spikes from 18 months prior. When a structural break occurs, the model's recency-dominated covariance matrix severely underestimates tail risk, because the temporal weighting kernel has effectively zeroed out the longer-horizon variance signal. A temporal regularization term penalizing short-horizon dominance—anchored to a full-sample overlay covariance estimate—would have flattened the recency kernel and preserved the structural break diagnostic signal in the downstream risk module.

New information is fresher and pops into mind more quickly, so it changes choices. Older information gets weaker and is less likely to be used.

A temporal weighting mechanism multiplies evidence by a recency-dependent kernel across the short-term memory buffer, with stronger weights for recent items and attenuated weights for older entries. The decay constraint at the buffer interface creates asymmetric influence, effectively biasing downstream decision modules toward recently activated structural elements.

Pause and review older items before deciding to balance choices. Use a checklist to include past information in your decision.

Implement temporal regularization that flattens the recency kernel across the retention interval to reduce short-horizon dominance. Incorporate retrospective aggregation layers that reweight decayed traces based on relevance signals and long-term statistics.

Overreaction to short-term noise; Ignoring stable long-term trends; Poor generalization across time

An adversarial actor can deliberately front-load recent negative or positive information (e.g., late-stage negative news drops before a vote or evaluation) to hijack the recency kernel and override a longer, more favorable history. In financial or political contexts, strategically timed disclosures, press releases, or manufactured events can flood the short-term buffer just before a decision point, crowding out older contradictory evidence. This technique is especially potent in sequential presentation formats—interviews, debates, earnings calls—where the actor controls the order and recency of information delivery.

Implement structured retrospective reviews that explicitly reweight older evidence alongside recent data before finalizing any decision, counteracting short-term buffer dominance. Use pre-committed temporal aggregation windows (e.g., rolling multi-year averages or full-sample overlays) to force long-horizon statistics into the evaluation frame. Train decision-makers to flag "what changed recently vs. what is structurally stable" as a mandatory checklist step, breaking automatic reliance on recently activated traces.