Confirmation Bias

Confirmation bias operates at three distinct points in the research process. At the search stage, the analyst chooses sources already aligned with the thesis: the bullish sell-side note, the management interview, the forum populated by fellow holders. At the interpretation stage, ambiguous data is read charitably: a weak quarter becomes "transitory," a competitor's price cut becomes "desperation." At the recall stage, the mind retrieves the datapoints that worked and lets the misses fade, which inflates the perceived hit rate of the original analysis.

In Bayesian terms the failure is asymmetric weighting of the likelihood. Supporting evidence is treated as highly diagnostic and moves the posterior sharply; contradicting evidence is treated as noise and barely moves it at all. The prior therefore hardens instead of converging on the truth as the evidence set grows.

Confirmation bias sits among the cognitive errors, in the belief perseverance family, alongside conservatism, representativeness, illusion of control and hindsight bias. The classification matters because it drives the recommended remedy. Cognitive errors stem from faulty reasoning and incomplete information processing, so they respond to education, structured process and better data: they can be moderated. Emotional biases stem from feeling and impulse and generally must be adapted to rather than corrected.

The near neighbour is conservatism bias, where an investor under-reacts to all new information and clings to a prior view. Confirmation bias is not indiscriminate stickiness: the investor updates readily, but only in one direction. A candidate should be able to separate the two from the description in a vignette.

The portfolio signature is concentration. An investor who only registers supportive evidence sees less risk in a position than exists, so position sizes creep up and the portfolio drifts toward a single theme, sector or, in the classic case, employer stock. Diversification is judged unnecessary because each holding feels individually well understood. Correlations across holdings are typically higher than the investor believes, since the same underlying thesis often drives them all.

It also shows up in quantitative work. Screens are built with criteria chosen after the conclusion is formed; backtests are re-specified until they support the strategy; disappointing factor results are attributed to a regime that will normalise. In manager selection, due diligence questions are framed to elicit the answers the allocator already expects.

Three tasks recur. First, identify the bias from behavioural evidence in a case: an investor who reads only bullish coverage, dismisses short-seller reports, or celebrates confirming news while explaining away misses. Second, classify it correctly as a cognitive error in the belief perseverance group, and justify the classification. Third, recommend a remedy consistent with that classification: moderate rather than accommodate.

Practical moderation techniques worth being able to name include disciplined collection of contradictory evidence, formal devil's advocate or red-team review, written investment theses with explicit falsification conditions, pre-mortem exercises, and maintaining a decision log so that recall of past reasoning is checked against the record rather than reconstructed from memory.