Confirmation Bias Turns Research Into a Search for Agreement
Once a view is formed, evidence gets filtered rather than weighed. The more research you do, the more confident you become, and confidence is not the same as accuracy.
The Mechanism
Confirmation bias operates in three stages, and each is a separate failure.
Search is selective: you look in places likely to support the view. Interpretation is asymmetric: supporting evidence is accepted readily while contradicting evidence receives scrutiny until a reason to dismiss it appears. Recall is biased: the confirming instances are the ones remembered later.
The result is that someone can conduct a genuine investigation and emerge more confident regardless of what the evidence actually showed.
Why Research Makes It Worse
This is the counterintuitive part. Spending more time on a position typically increases conviction more than it increases accuracy.
An investor who has read every filing, listened to every earnings call, and built a detailed model has invested substantial effort. That effort creates commitment, and each additional confirming data point feels like accumulating proof, while each disconfirming one gets absorbed as a temporary issue or a misunderstanding by the market.
Depth of research produces confidence reliably and accuracy only sometimes. Confusing the two is how well informed people hold wrong positions for a long time.
The Modern Amplifier
Information environments now sort themselves. Following accounts that discuss a position favourably, reading commentary that shares the thesis, and participating in communities organised around a specific view produces an evidence stream that is filtered before it arrives.
The filtering feels like consensus. What it actually reflects is who is talking, and the people talking are disproportionately those who already agree.
How Professionals Attack It
| Practice | What it forces |
|---|---|
| Write the bear case yourself | Genuine engagement with the other side |
| Pre mortem | Assume failure and explain it |
| Assign a devil's advocate | Someone paid to disagree |
| Track record of falsifiers | Stated conditions that would change the view |
| Read the best opposing argument | Not the weakest one available |
The pre mortem is the most useful and the most uncomfortable. Rather than asking what could go wrong, it assumes the position has already failed badly and asks what happened. That framing bypasses the defensive response, because the failure is stipulated rather than argued about.
The Falsification Standard
The strongest single discipline is stating in advance what would prove you wrong.
A thesis that cannot be falsified is not a thesis, it is a preference. If a stock rising confirms the view and a stock falling means the market has not understood it yet, no possible observation can change your mind, and the analysis has become unfalsifiable.
Writing specific conditions in advance, such as a named metric failing to reach a level by a date, converts the position into something that can be tested. It also makes the moment of being wrong identifiable, which is what allows exiting before it becomes expensive.
Where It Compounds With Other Effects
Confirmation bias is rarely alone. Combined with anchoring it defends the purchase price. Combined with loss aversion it supplies the reasons for not selling a loser. Combined with the sunk cost effect it justifies additional investment in a deteriorating position.
The pattern of averaging down repeatedly into a declining position, with a fresh explanation each time, is usually all three operating together rather than any one of them.
The Bottom Line
Confirmation bias filters evidence at the point of search, interpretation, and recall, which means more research reliably produces more confidence and only sometimes more accuracy. The defences are structural rather than attitudinal: write the opposing case yourself, run a pre mortem, and state in advance the specific observations that would prove the thesis wrong.