Confirmation Bias in Trading: Signs and a Diagnostic

Confirmation bias, in the general psychological literature, is the tendency to seek, interpret, and recall information in ways that favor an existing belief, and it can operate whenever a belief is already in hand — before a decision is made, while it is being acted on, or afterward.¹ In trading, that means it can shape which setup a trader selects before entry, how an open position’s evidence gets handled, and what gets remembered during post-trade review; the section below on where the pattern concentrates covers all three. This page’s diagnostic focuses specifically on thesis protection while a position is open. The relevant test is whether information processing unjustifiably favors the existing thesis under the decision standard that applies at that point. Entry and invalidation standards may legitimately differ, and a standard that stays stable can still be biased if it was never defensible in the first place. The test is not whether the trader remains convinced — conviction can be correct — but whether the evidence was handled in a direction-dependent way without a defensible reason.

That distinction matters because a held position is not, by itself, evidence of bias. A trader can hold through adverse movement because the invalidation condition genuinely has not occurred. The pattern this page diagnoses is unjustified thesis-favoring information processing under the applicable decision standard; it is not established merely because entry and exit decisions use different standards.

What is confirmation bias in trading?

Confirmation bias in trading is the selective seeking, weighting, or interpretation of information in a direction that protects an existing belief — a setup criterion, an open thesis, or a post-trade narrative — from disqualifying evidence, once that belief is already in hand. It is an information-processing pattern, not a market-timing error, and an open position is not a requirement for it. The bias operates in three distinct ways, and a single decision can show one or more of them:¹

Selective search — checking sources likely to agree with the thesis (a bullish forum, a specific analyst) more often or more thoroughly than sources likely to disconfirm it.
Selective weighting — treating confirming information as more decisive than its reliability and relevance warrant, while discounting disconfirming information of comparable reliability and relevance below what it warrants, rather than applying one direction-independent standard to both.
Selective interpretation — reading ambiguous or mixed evidence as consistent with the thesis, when the same evidence would have been read as inconclusive or negative before entry.

None of these require dishonesty. Nickerson’s review describes confirmation bias as showing up “without any intention to defend or support beliefs held” — a default way evidence gets processed once a hypothesis is in hand, not a conscious choice to ignore contrary information.¹

None of this means confirming and disconfirming evidence must always receive identical weight. A more reliable source, a more relevant data point, or a more diagnostic piece of evidence can reasonably count for more than a weaker one, regardless of which direction it points. What the three patterns above describe is a standard that depends on direction — the same source, evidence type, or ambiguity gets a different reading depending on whether it supports or threatens the thesis — not a standard that appropriately weighs evidence of unequal informational value.

Confirmation bias is not the same as conviction, patience, or a wrong prediction

Situation	What is actually happening	What the diagnostic indicates
Price moves against the position; the invalidation condition has not occurred; the trader holds	The plan’s own rule permits holding; nothing here describes how evidence was searched for or weighted	Not indicated — holding under a still-valid rule is not itself an evidence-processing event
Price reaches the invalidation condition; the trader re-reads the same chart, with no new information, until it “still looks fine”	No new evidence entered the picture, but the interpretation of the same chart shifted toward the reading that avoids invalidation	Consistent with confirmation bias — the interpretive standard loosened exactly at the point it was supposed to hold steady. (A stop or invalidation level can also change for defensible reasons — new information, a genuine reassessment of risk — which this row does not describe)
A confirming data point arrives and the trader adds to the position under a predefined scale-in rule	A written rule, set before this occurrence, governs the response to new confirming evidence	Not indicated by this alone — rule compliance is not proof that no bias occurred elsewhere in the decision, but nothing here shows a standard that shifted
A confirming data point arrives and the trader adds size with no predefined rule; a disconfirming data point of similar relevance and reliability from the same period was available, and the trader later says they skipped it because checking it “would’ve just complicated a position that was already working”	The discretionary add is not itself the problem — discretion isn’t inherently biased. The signal is the stated reason for skipping the disconfirming source: it was avoided because it threatened the thesis, not for a reliability, relevance, or availability reason	Consistent with confirmation bias — the reason given for skipping the source is direction-dependent, not evidentiary
The trade loses after the trader held through the invalidation point	The position lost money; nothing here describes how evidence was searched for or weighted at the decision point	Not established by the outcome — a probabilistically sound thesis can produce a losing trade, and a poorly-evidenced one can produce a winning trade. Outcome and process quality are separate questions

The last two rows matter most: confirmation bias is a claim about process, not about outcome, and a rule being followed is not the same claim as evidence being searched for and weighted evenly. A biased decision can still make money, a well-calibrated decision can still lose, and a rule can be followed by the letter while the search behind an earlier, discretionary step was uneven. Grading the pattern by whether the trade worked, or by whether a rule existed, answers a different question than the diagnostic below asks.

A two-question diagnostic for a specific decision

This is an editorial decision-review framework, not a validated clinical or psychometric instrument. It is meant to structure a trader’s own after-the-fact review of one decision point — usually the moment new information arrived or the invalidation condition was approached — not to produce a certified diagnosis.

A difference in how much bullish versus bearish material got checked is not, by itself, evidence of bias. Sources differ in reliability; some evidence is more relevant or more diagnostic than other evidence; information isn’t always equally available; and market conditions genuinely change in ways that justify treating new evidence differently than old evidence. The diagnostic below is only informative when a difference in evidence handling can’t be explained by one of those legitimate reasons.

Was thesis-favoring selective processing identified? Answer YES when relevant evidence was sought, interpreted, or discounted because it agreed with the existing thesis rather than because of source reliability, relevance, diagnostic value, or availability. Answer NO when no unjustified direction-dependent asymmetry is identified. A bearish signal checked less than a bullish one is not on its own the pattern; a comparable bearish signal checked less because it was bearish is.
Is the standard defensible and consistently applied? Answer YES only when both conditions are supported. Answer NO when at least one condition fails. Answer UNKNOWN when the evidence is insufficient to determine either condition. A standard can be biased if it was created before or after entry around sources likely to agree with the thesis; its timing alone does not establish bias. A standard can also be revised legitimately when new information, source reliability, or conditions change; revision alone is not automatically bias. The question is whether the source-selection process and evidence criteria were defensible and applied consistently for this decision, not whether a predefined rule merely existed.

A NO on Question 1 means no unjustified thesis-favoring asymmetry was identified; it does not certify that the entire decision process was unbiased. A YES on Question 2 requires both a defensible standard and consistent application. What the two questions together are checking is whether any direction-dependent asymmetry — new or longstanding — lacks a defensible reason.

Classify the decision using four distinct outcomes rather than collapsing missing evidence into an absence of bias:

Consistent with confirmation bias — evidence processing shows direction-dependent selective search, weighting, or interpretation with no defensible explanation (for example, a comparably reliable disconfirming source was skipped specifically because it threatened the thesis, or a standard — new or longstanding — favors the thesis for no reason the trader can defend).
Possible confirmation bias — investigate further — a direction-dependent difference in evidence handling is present, but there’s an unresolved question of whether it reflects a legitimate reason (reliability, relevance, availability, a defensible standard change) or an unjustified one.
No confirmation bias indicated at this decision point — the available description shows evidence handling that did not differ by direction, or a difference adequately explained by reliability, relevance, availability, or another defensible reason.
Insufficient evidence to determine — the description does not contain enough information to tell whether evidence handling differed by direction or why. Missing evidence is not evidence that the process was unbiased.

Where a scenario explicitly shows evidence being sought or discounted because of its direction, with no legitimate explanation offered, classify it as consistent with confirmation bias rather than hedging — the four-way outcome preserves both a conclusion and an honest unclassified state.

Where confirmation bias concentrates in a trade

Before entry, in setup selection. A trader scanning for a specific setup finds a chart that appears to confirm it and stops looking — skipping a check the trader would normally run for whether the same chart also satisfies the criteria for the opposite setup, specifically because running it might complicate a setup that already looks good. Stopping a search once a confirming setup is found is not automatically confirmation bias — a search has to end somewhere; the pattern is discounting or avoiding a check that would ordinarily run, because of what it might show. Trading decision-making’s checkpoint 2 separates the observation from the method’s interpretation of it for exactly this reason — the interpretation is where a thesis already in mind can shape which observations get recorded as relevant.

At the invalidation point. This is a common point of concentration for the pattern, because the trader has an incentive to find a reason the invalidation does not really apply. A widened stop, a redefined support level, or a reclassified “just noise” reading of the disqualifying move can all be ways the evidence standard loosens exactly when it should hold steady — but none of them is automatically bias. A stop or invalidation level can be revised for reasons that have nothing to do with protecting the current position — new information that changes the risk calculus, for instance — but a revision is not automatically acceptable just because a reason is offered, and it cannot move a hard stop or account-level risk limit past its boundary; that limit holds regardless of how the evidence standard is being reconsidered. The pattern is specifically when the standard actually applied at the invalidation point is looser than the standard the trader’s own plan calls for at that point, without an independently defensible reason for the difference. Entry and invalidation are separate decisions with their own applicable standards — a legitimate difference between them is not, by itself, evidence of bias.

During review. A trader assembling a post-trade rationale can select only the evidence that supported the eventual outcome, producing a record that looks more consistent than the live decision actually was. Trade journal examples covers why separating the contemporaneous observation from the later interpretation is what makes this specific distortion visible in a log instead of invisible in a summary.

Pattern	What changes	Canonical owner
Confirmation bias	Evidence is sought, weighted, or interpreted in a way that unjustifiably favors an existing thesis — whether that favoritism is new or has been there all along	This page
Loss aversion	The exit or risk rule changes because realizing a loss feels worse than an equivalent gain	Loss aversion in trading
Herding	Other traders’ visible actions substitute for the trader’s own qualification criteria	Herding and consensus trading
Outcome bias	A decision is judged by its result rather than by the information available when it was made	Covered within post-trade review

These can co-occur. A trader who moved a stop to avoid a loss (loss aversion) may also seek out confirming commentary to justify that move after the fact (confirmation bias). The two-question diagnostic above isolates the evidence-processing pattern specifically; it does not require ruling out the other mechanisms first.

A worked example: the same new information, two different searches

Consider a hypothetical trader holding a long position on a thesis that a specific demand zone will hold.

Version A. Price approaches the zone. The trader checks one source, finds confirming order-flow data, and holds. A second source — comparably reliable, equally available, one the trader would normally check for this setup, and one that has flagged this zone breaking down before — goes unchecked. Asked afterward, the trader says they knew the second source could show the zone failing, but didn’t check it because the first source had already given the answer they wanted. Diagnostic: Question 1 = YES — a source capable of disconfirming the thesis was skipped specifically because a confirming answer had already arrived, not because of the source’s reliability, relevance, or a predefined stopping point. Question 2 = NO — the normally required source check was not applied consistently and the departure was not defensible. Consistent with confirmation bias, independent of what price does next.

Version B. Price approaches the same zone. The trader checks the same three sources every time this setup occurs, regardless of what they show, using a source-selection process and a two-of-three accumulation threshold that were both set during strategy development — before this position, or any thesis about this specific trade, existed — rather than assembled once the trade was already on. The trader can explain why those sources are relevant and reliable enough for this decision, and the criteria are applied the same way when evidence supports or threatens the thesis. Two of three sources show accumulation; the trader holds. Diagnostic: Question 1 = NO — no unjustified direction-dependent asymmetry is identified. Question 2 = YES — both the source-selection process and the evidence criteria are supported as defensible and consistently applied. No confirmation bias is indicated at this decision point by the available evidence. That does not certify the trader’s entire decision process as unbiased, or prove that the three sources are independent or that the two-of-three rule is well-designed; it evaluates only the decision point described.

Both versions can produce the same closing print. The diagnostic distinguishes them by whether the source-selection process and evidence criteria were defensible and applied on direction-independent grounds — not by whether a rule existed in advance. What separates Version A from Version B is that Version A’s source-skipping was explicitly direction-dependent, while Version B’s sources and criteria were justified and applied the same way regardless of what the evidence showed.

Guardrails for a defensible evidence review

The objective is defensible evidence selection, appropriate weighting, and auditable revisions — not simply preventing every change. A thesis that keeps getting confirmed by evidence reviewed under a defensible standard may be holding up; a legitimate revision should remain reviewable rather than being treated as bias automatically. When the belief being revised is an explicit probability, probability revision quality covers how to review each update’s size, timing, and rationale against the evidence available at the time.

Write the confirming and disconfirming sources down before entry. If the sources that would change the trader’s mind are named in advance, a live decision can be checked against that list instead of assembled from whichever sources are convenient at the time.
Set the evidence threshold before entry, not at the invalidation point. “I will exit if two of three sources turn negative” is a rule that can be audited later. “I’ll know it if I see it” cannot.
At the invalidation point, check the disconfirming source list first. Reversing the usual order — disconfirming evidence before confirming evidence — makes it harder to stop searching as soon as a comfortable answer appears.
Log which sources were actually checked, not just the conclusion. A record that shows “checked A, B; did not check C” makes an uneven search visible on review even when the trader felt the search was thorough at the time.
Ask the reversal question at the decision point. Would this same piece of information change my mind if I currently held no position and were deciding whether to enter fresh? This is a counterfactual for spotting an unjustified double standard, not a rule that entry and holding must use identical thresholds — position management can reasonably weigh transaction costs, existing risk exposure, and information that only arrived after entry in ways a fresh-entry decision would not. The question is whether the trader can name a defensible reason the same information is being read differently now, or whether the more comfortable reading is just protecting the position.

None of these guardrails, individually or together, guarantees that confirmation bias is absent, and none of them substitutes for risk management. A hard stop or account-level risk limit exists independently of how many sources have turned negative and should not be held past that limit because an evidence threshold hasn’t yet been reached. Revising a predefined rule is also not automatically confirmation bias — a source can stop being available, or conditions can genuinely change — the distinguishing question is whether the revision was made for a reason the trader could have defended before knowing it would favor the existing position, or only after.

Confirmation bias among traders specifically

Confirmation bias is not only a laboratory finding. A field study using 502 investor responses gathered through one of the largest stock message-board operators in South Korea found that investors exhibiting confirmation bias — preferentially valuing messages that supported their prior beliefs — were also more overconfident and traded more actively.² The published analysis went beyond self-reported belief: for each investor’s forecast, it computed a direction-adjusted return gap — for a bullish forecast, the stock’s actual return over the following month minus the forecast return; for a bearish forecast, the forecast return minus the stock’s actual return — and reported that stronger confirmation bias was associated with a statistically significant larger gap in the unfavorable direction. That is a finding about how far the stock’s subsequent return fell short of what the investor had forecast, not a measurement of the investor’s own realized trading performance or account P&L, and the result is an association documented among South Korean message-board investors, not a causal proof or a general finding about discretionary traders. It supports treating the selective-evidence pattern as a real tendency among investors who rely on this kind of information, not a purely theoretical concern confined to the lab.

Positive testing and confirmation bias are related but not the same thing. Klayman and Ha’s influential reanalysis showed that many behaviors that look like confirmation bias are better explained by a general “positive test strategy” — a tendency to test the cases most likely to carry the property being checked — and argued that this strategy is often a reasonable, informative way to gather evidence, not proof of a desire to protect a belief.³ But positive testing isn’t universally safe either: under some conditions it produces systematic errors of its own, and it doesn’t account for every pattern that gets labeled confirmation bias. Consistent with Nickerson’s account earlier on this page, neither positive testing nor confirmation bias requires conscious dishonesty or deliberate motivation to protect a belief — both can operate as a default way evidence gets processed once a thesis is in hand.¹ The pattern this page diagnoses is narrower than either concept alone: not “the trader tested the cases most likely to be relevant,” but “the trader’s evidence-handling unjustifiably favored an existing thesis” — whether that shows up in what got tested, how it was weighted, or how it was interpreted.

Where Costante fits

Costante supports session planning, self-defined behavioral guardrails, low-friction trade and behavioral logging, and structured review against the trader’s own plan. That can give a trader a more reviewable record of a decision instead of relying entirely on memory, but it does not by itself record or validate the evidence sources used in this diagnostic.

Costante does not have a dedicated field for confirming or disconfirming evidence sources, does not compare the evidence standard used at entry against the standard used at invalidation, does not run this diagnostic automatically, and does not detect confirmation bias or decide which sources should count as confirming or disconfirming. The guardrails above — naming sources in advance, fixing a threshold, logging what was actually checked — are a manual practice a trader can carry out using general logging fields like these; the trader defines the sources and thresholds and remains responsible for applying them consistently.

Frequently asked questions

Is confirmation bias the same as being wrong about a trade?

No. Being wrong is a directional outcome, and a probabilistically reasonable decision can still lose. Confirmation bias is a separate, process-level claim about how evidence was searched for and weighted, not about whether the prediction turned out right. A correct thesis can still be defended with an uneven evidence standard, and an incorrect thesis can be abandoned using a perfectly even one.

Does holding a position through adverse movement prove confirmation bias?

No. Holding is consistent with the plan if the invalidation condition has not occurred. The diagnostic applies to how evidence was searched and weighted at a specific decision point, not to the fact of holding itself.

How is confirmation bias different from ordinary conviction?

Conviction is a belief state. Confirmation bias is a claim about the process used to test that belief — specifically, whether confirming and disconfirming evidence were searched for and weighted using the same standard. A trader can be highly convinced and still apply an even evidence standard, or hold weak conviction while still searching unevenly.

Can a written trading plan prevent confirmation bias?

A plan can help by fixing evidence sources and thresholds before the decision point, which makes a later change to those sources or thresholds observable and reviewable instead of invisible. It does not eliminate the opportunity for selective search within those sources or selective interpretation of ambiguous evidence, which is why a contemporaneous log of what was actually checked remains part of the diagnostic.

Sources

Costante provides educational workflow tools, not financial advice. Trading involves risk.

Nickerson, R. S. (1998). Confirmation Bias: A Ubiquitous Phenomenon in Many Guises. Review of General Psychology, 2(2), 175–220. ↩ ↩² ↩³ ↩⁴
Park, J., Konana, P., Gu, B., Kumar, A., & Raghunathan, R. (2013). Information Valuation and Confirmation Bias in Virtual Communities: Evidence from Stock Message Boards. Information Systems Research, 24(4), 1050–1067. ↩
Klayman, J., & Ha, Y.-W. (1987). Confirmation, Disconfirmation, and Information in Hypothesis Testing. Psychological Review, 94(2), 211–228. ↩