Projection Bias

People expect their current feelings or wants to stay the same in the future. They make choices as if their future self will feel exactly like they do right now.

Projection bias is the cognitive tendency to overestimate how much current preferences, affective states, or needs will persist when making predictions or decisions about the future. It leads decision processes to overweight present-centric signals during forecasting and choice, producing suboptimal selections once the future state actually changes.

A person goes grocery shopping while hungry and loads their cart with far more snacks and ready-to-eat food than they need, assuming they will always feel this ravenous — only to find the food uneaten a week later because their appetite returned to normal.

A quantitative portfolio manager, evaluating a new momentum strategy during a high-volatility regime, overweights recent high-volatility signals in the model selection phase, projecting that the current turbulent market state will persist. When volatility reverts to normal, the model's decision pipeline — shaped by projection bias — suffers significant performance degradation, because present-regime salience dominated the temporal encoding layer at the moment of selection. A rigorous resistance protocol would require rolling-origin evaluation across multiple historical regimes, plus explicit downweighting of present-regime inputs relative to the long-run distributional priors.

Current emotions or wants make people assume the future will feel the same. That assumption is exactly what leads them to pick options matching how they feel right now.

Projection bias operates by overweighting present-state feature activations in the preference module, constrained by limited temporal encoding and an asymmetrical gating of future predictors; current affective signals dominate simply because they carry stronger connection weights. That structural asymmetry between present and prospective representations biases the expected utility estimation toward the immediate state.

Pause and imagine how you might feel later, before choosing. Listing out ways your feelings could change helps too.

Structured future-self perspective taking, along with scenario sampling, discounts the present-state salience and reweights the temporal predictors when estimating future utility. Calibrated temporal decay factors reduce the present-signal dominance during selection further.

Overcommitment to short-term preferences; Underpreparation for changed future needs; Regret when future differs

An adversarial actor can exploit projection bias by engineering high-affect or high-need moments immediately before a target makes a consequential decision — for example, inducing hunger, fear, or desire at point-of-sale or negotiation — ensuring the target's decision pipeline is saturated with present-state signals that overwhelm future-state predictors. Subscription and commitment products can be deliberately priced and framed to lock in consumers during peak enthusiasm states, betting that projection bias will cause them to overestimate future usage and underestimate cancellation effort. In influence operations, adversaries can manufacture emotional crises or scarcity events to spike present-state salience, causing targets to commit to alliances, purchases, or policy positions that serve the adversary's interests but misalign with the target's stable long-run preferences.

Structured future-self simulation exercises — explicitly imagining oneself in emotionally neutral or contrasting future states before committing — can attenuate present-state salience dominance in the decision pipeline. Incorporating mandatory cooling-off periods or temporal commitment delays in institutional choice architectures forces the decision process to sample across multiple affective states rather than a single high-salience present moment. Applying calibrated temporal decay factors to present-state feature weights during expected utility estimation can reduce projection errors in forecasting and planning contexts.