Confirmation Bias in Trading: Reading Signals Selectively

Confirmation bias is the reason a trader who is already long counts every green candle as proof and treats every red one as noise. Once a thesis is in place, attention tilts toward the evidence that supports it and away from the evidence that would break it. This primer defines the bias, separates it from the near neighbors it is often confused with, shows where it surfaces in a trading day and a journal review, and gives a way to structure decisions so the disconfirming evidence gets a hearing.

What confirmation bias is

The bias is not lying to yourself on purpose. It is the quiet, automatic slant in what the mind reaches for and how heavily it counts each piece once a position or an opinion exists. A trader who expects the session to trend up will register the higher low that fits and skim past the failed breakout that does not. Both prints happened; only one gets full weight, because one confirms the standing view and the other threatens it.

Nickerson framed confirmation bias as ubiquitous and appearing in many guises, which is the part that matters for a trader.¹ It is not a single mistake made once. It is a standing tilt that colors chart reading, news reading, and the after-the-fact review of a losing trade. The belief does the filtering before the trader is aware there was any filtering to do, so the evidence that survives feels like the whole picture rather than a selected slice of it.

How it differs from its near neighbors

The distinction is worth drawing because the fix differs for each. Recency bias is temporal: the last few trades dominate the read of a strategy no matter what the trader believes about it. Confirmation bias is directional: the trader has a thesis, and the thesis decides which signals get counted. A trader can hold recency bias with no thesis at all, and can run confirmation bias on a belief formed weeks ago.

Anchoring is narrower still, a fixation on a first number such as an entry price or a round level that then frames every later judgment. The sunk cost fallacy is about commitment, honoring money and time already spent by continuing. Confirmation bias can feed all three, because a filtered view of the evidence makes the recent run look cleaner, the anchor look justified, and the losing hold look reasonable. But the mechanism to name here is the filter on incoming signals, and the fix is to force the disconfirming signal back into view.

How it shows up in a trading day

The clearest live case is holding a directional thesis into the open. A trader decides the day is long and then reads the tape through that lens: pullbacks are buying opportunities, failed pushes are shakeouts, and the level that should have said the idea was wrong gets reinterpreted as a retest. The same candles that would warn a flat observer are absorbed as confirmation by the committed one.

A second case is selective use of indicators. A trader who wants the long checks the moving average that is sloping up and looks past the momentum reading that is rolling over, a pattern covered from the tooling side in why stacking more indicators does not strengthen an edge. The third case is news and social feeds: posts that agree with the position are read closely and remembered, posts that disagree are dismissed as noise or as someone talking their own book. In each case the evidence was available. The filter decided what counted.

Why it corrupts a strategy review

A journal is supposed to be the correction for a biased memory, but a biased reading of the journal reintroduces the same error. Suppose a trader believes a particular breakout setup is strong. Reviewing the log, they linger on the eight winning breakouts, feel the belief confirmed, and move on. The twelve breakouts that failed are read quickly, each one explained away by a one-off reason: bad news, a slow day, a mistimed entry. The setup keeps its reputation because the review counted the wins in full and the losses at a discount.

The honest test runs the other way. To evaluate a setup, a trader has to count every instance in the matching condition, wins and losses alike, and let the aggregate speak, which is the discipline behind what makes a behavioral pattern claim trustworthy. If the favored setup shows eight winners at an average that does not cover the twelve losers, the edge is not there, whatever the memory of the good trades says. Confirmation bias is what lets the memory win that argument, and a full count is what settles it.

How to catch it in your trade log

The bias thrives on evidence that is judged after the fact, so the fix is to fix the standard in advance. At entry, alongside the setup and the stop, write the specific condition that would prove the trade wrong: a level that should not trade, a structure that should not break, a reading that should not appear. This is the invalidation, and writing it before the outcome removes the room to reinterpret it later.

With that field in place, the log becomes measurable. Tag each entry as planned or improvised at the moment it is placed, the same binary split covered in how to tag plan and improvised trades, and add a flag for any trade held after its written invalidation broke. Then review the held-through group on its own. If most of the day's red sits in trades that ran past a level the trader had already named as the exit, the position was being read to fit the thesis, and the count now says so in place of the story told after the close.

How to structure decisions so it cannot drive them

Pre-committed invalidation handles the single trade. A level written at entry as the point where the idea is wrong converts the exit from an after-the-fact judgment, which the bias can bend, into a fixed condition set while the trader was still neutral. When that level trades, the position closes, because the standard was agreed before there was a thesis to protect.

The deliberate counter-case handles conviction. Before sizing up or adding, a trader states out loud the strongest reason the position is wrong right now and what price would confirm it. Naming the disconfirming case gives it the weight the filter would otherwise strip away. The full-count rule handles the strategy. Judging a setup on every trade in the matching condition, held to the sample floor described in what sample size a strategy needs, replaces the remembered winners with the complete record. Structured this way, the contrary evidence is written down, spoken, and counted, so the bias never gets to decide by leaving it out.

Frequently asked questions

• q: Is confirmation bias the same as recency bias? a: No. Recency bias overweights the most recent outcomes regardless of any belief, so the last few trades dominate the read of a strategy. Confirmation bias is directional: a thesis is already held, and it filters which incoming signals get counted. A trader can run one without the other.
• q: How does confirmation bias corrupt a journal review? a: A trader reviewing the log for a favored setup lingers on the winners and skims past the losers, explaining each loss away as a one-off. The setup keeps an inflated reputation. The correction is to count every trade in the matching condition, wins and losses alike, and read the aggregate.
• q: What is the single most effective guard against it? a: Write the invalidation level before entry. Recording the specific condition that would prove the trade wrong, while still neutral, removes the room to reinterpret price later. An exit held past that written level is then a measurable flag rather than a matter of judgment.
• q: Can experience remove confirmation bias? a: The tilt does not disappear, because it is how attention works once a belief exists. It can be structured out by fixing the disconfirming test in advance, stating the counter-case before adding, and judging setups on a full count. The skill is procedural, not a matter of trying to be objective.

Sources

Footnotes

1. Raymond S. Nickerson, "Confirmation Bias: A Ubiquitous Phenomenon in Many Guises," Review of General Psychology, Vol. 2, No. 2, 1998, pp. 175-220. ↩ ↩²