Self-Attribution Bias in Trading: Skill vs. Luck

Self-attribution bias in trading is the tendency to explain winning trades with your own skill and losing trades with something outside your control — the news, the spread, a bad fill, “just variance.” The pattern is not the wrong part by itself: some losses really are caused by events a trader could not have foreseen, and some wins really are caused by a sound read. The problem appears when the explanation changes with the result while the recorded decision inputs do not. If two rule-aligned trades share a setup and a pre-trade expectation, but one is filed under “my read” and the other under “bad luck” only because of the P&L, the review note may be tracking the outcome rather than the decision.

That makes it a different problem from the ones the neighboring articles cover. Process vs. outcome feedback is a review framework, not a bias: it decides which kind of evidence answers which grading question. Outcome bias is a known result relabeling the classification of the same decision as aligned or deviated. Self-attribution is what happens one step later, in the causal story the trader writes about why a result occurred and, over many trades, what the trader concludes about their own skill.

What is self-attribution bias?

In psychology, the self-serving attributional bias is the tendency to credit oneself for good outcomes and to attribute bad outcomes to external circumstances. In finance research, “self-attribution bias” usually refers to the same asymmetry applied to investment results, and to what it does to a person’s belief in their own ability.¹

Two features of the research matter for a trader reading it, because they limit how strongly it can be applied.

The asymmetry is widely observed, but its cause is disputed. A meta-analysis of 266 studies found the self-serving pattern in nearly all samples it examined, with an average effect size of d = 0.96, but the size varied by age, culture, and clinical status — for example, it was smaller in Asian samples than in U.S. samples and smaller in samples with depression.² Those studies were general psychology tasks, not trading, so the finding says the pattern is common in people, not that it is present in any particular trader’s journal.

Not every asymmetry is motivated. In a classic review, Miller and Ross found only limited support for a general self-serving bias. They found some support for self-enhancing attributions after success and minimal support for self-protective attributions after failure, and they proposed a non-motivational alternative: part of the asymmetry can arise because people expect their behavior to produce success, among other information-processing tendencies, rather than from a wish to protect self-image.³ Applied to trading, that means a gap between how wins and losses are explained can come from real differences in expectation or information, not only from motive. A review process therefore should not label the gap “ego”; it should check whether the gap persists among reasonably comparable trades.

Why it matters for a trader specifically

Trading gives the bias an unusually clean channel: a result arrives quickly, is expressed as a number, and can be explained in a sentence. The sentence is what accumulates into a belief about your own ability.

Gervais and Odean built a model of exactly that accumulation. In it, a trader who takes too much credit for successes and too little blame for failures becomes overconfident, and the expected overconfidence rises early in a career before experience lets the trader estimate ability more accurately.⁴ That is a theoretical model of learning, not a measurement of any individual trader, and it should be read as a mechanism to check for rather than a prediction that it is happening.

There is also survey evidence with real brokerage records. Hoffmann and Post matched monthly survey answers to the trading records of clients of a large Dutch discount broker and found that investors with higher recent returns agreed more strongly that those returns reflected their investment skill — and that this held for the investor’s own returns but not for overall market returns.¹ The result is an association in one sample of self-selected survey respondents. It does not show that any specific investor was mistaken about their skill, and it does not show the belief caused any particular decision.

What follows for a discretionary trader is narrow and practical: the risk is not that you feel good after a win. It is that the running tally of “skill” in your notes is filled by outcome rather than by evidence, and that tally can then shape how much size, confidence, or deviation from the plan feels earned next session. Escalating position size after a run of wins is one place where that confidence can surface.

How is self-attribution different from outcome bias and confirmation bias?

These patterns share a review file and are often confused. They act on different objects.

Bias	What it distorts	The question it can answer wrongly	Owning article
Outcome bias	The classification of a single decision (aligned, deviated)	“Was this decision rule-aligned?” — answered by looking at the P&L	Outcome bias in a post-trade review
Confirmation bias	Which evidence is sought, believed, or remembered, during a trade or in later review	”Does the evidence support my view?” — answered by selective evidence	Confirmation bias in trading
Anchoring	Weight given to a fixed reference number	”What is this worth now?” — answered from an old price	Anchoring bias in trading
Self-attribution	*The stated cause* of a result, and the trader’s running belief about their skill**	”Why did this happen, and what does it say about me?” — answered differently for wins than for losses	This article

A single trade can involve more than one of these. Each has a different repair: reclassify from the plan (outcome bias), run a disconfirmation check (confirmation bias), or — for self-attribution — hold the explanation to the same evidence standard whichever way the trade ended. Process-versus-outcome feedback is not a row in the table because it is a review framework rather than a bias; the outcome-bias repair leans on it.

Where it enters a trading record

The bias needs a place to live, and in most journals it lives in free-text fields: “why it worked,” “why it failed,” “lesson.” It rarely lives in the entry, size, or exit fields, which are facts. Three habits make the free-text channel easy to skew:

Different effort by outcome. A losing trade prompts a search for a cause, so more explanation gets written; a winning trade often gets one line. Effort itself can push the record toward “external cause” on losses, though this is an ordinary possibility rather than a finding from the research above.
Unfalsifiable causes. “Market was choppy,” “smart money,” “unlucky” cannot be checked against the record. “Skill,” “good read,” “patience” often cannot either.
Causes chosen after the result is known. The available explanation set is not fixed in advance, so whichever fits the result gets picked.

None of these proves bias. They describe conditions under which a bias, if present, would go unnoticed.

The attribution symmetry check

The check has four steps. Its purpose is to compare how wins and losses are explained among reasonably comparable trades, and to test whether each explanation is held to the same evidence standard. It is a descriptive review indicator, not a validated psychological instrument, and it does not reconstruct why any single trade produced its P&L.

1. Fix the coding vocabulary before the result exists. Use two separate fields, the same for wins and losses. Factor type is the kind of factor a note names: decision, execution (fill, timing), market condition (regime or session type), or event. Attribution is whom the note assigns that factor to: own action, external factor, mixed, or unresolved. The two fields are not equivalent. An execution factor such as a late fill can be attributed to the trader’s order, to liquidity, or to neither, and a note that names a factor without assigning it is coded “unresolved.” A note can name several factors; its primary attribution is the main explanation the trader explicitly recorded, and secondary factors can be recorded separately. A trade with no note is coded “no explanation recorded,” not invented and not dropped.

2. Classify process from the contemporaneous plan and record. Use the same first step as process-versus-outcome grading: aligned, deviated, or unclassified, decided from the plan as written before the trade and the execution record, not from a plan rewritten around the result. Hiding the P&L during this first grading pass is an optional practice that reduces contamination by the outcome; it is not required for the check to be useful.

3. Record the pre-trade expectation before the result is known. Write down, before entry, how strongly the setup was expected to work — even a coarse label such as standard or lower-quality setup. This is a partial control for the expectancy explanation of Miller and Ross: if the trader expected the trades to work about equally well, differing explanations are less likely to reflect differing expectations. An expectation written after the result cannot serve this purpose, so trades without a contemporaneous record are marked “not recorded” rather than reconstructed.

4. Keep three things separate for every explanation, win or loss.

The recorded explanation: what the trader’s note claims.
The observable evidence: what behavior or circumstances the record establishes.
The causal conclusion: whether the claimed contribution to P&L can be causally established, which it often cannot.

An explanation is “supported by an identifiable record” or it is not, and the same question is asked of every note. A record makes a claim checkable, not correct, and it supports a claim narrowly:

An entry that followed a written rule supports process compliance. It shows the behavior a note refers to occurred. It does not show that the trader attributed the result to that behavior, or that the behavior produced the profit.
A timestamped news release shows that an event occurred. It does not show the event was unforeseeable, that it caused the loss, or that the trade would otherwise have won.
A fill measurably away from an appropriate reference quote shows an execution discrepancy, not who or what caused it: order timing, liquidity, order type, routing, or other circumstances. It does not measure the discrepancy’s contribution to realized P&L, and it does not by itself validate either an own-action or an external-cause attribution. If the record does not identify the source, responsibility stays unresolved.

An explanation with no identifiable record is a placeholder, not a finding. That does not make the opposite explanation correct, and an explanation with a record can still have an unresolved cause. The step tests whether the evidence standard is consistent, not what caused each trade’s result.

Matching is partial. Setup, process classification, and pre-trade expectation are partial controls: they reduce some sources of variation without eliminating confounding, and a coarse expectation label cannot capture every difference in expected return or outcome probability. Where the record allows, also note differences in session or market regime, execution conditions, relevant events, position size, and trade timing. Matching need not be perfect — demanding that would leave comparison groups too small to read — but the comparison stays observational and descriptive.

Then compare across a sample. Define one coding rule and apply it identically to wins and losses:

Own-action attribution: a review explanation in which the trader attributes some contribution to the outcome to their own decision or execution.

A note that only describes following a rule, without attributing any part of the outcome to that behavior, is not own-action attribution. Nor does own-action attribution mean the action broke a rule, that the trader is skilled, or that the trader caused the whole outcome. A rule-aligned loss can still attribute part of the result to the trader’s own decision, such as choosing one of several entry options the plan explicitly permits. Attributing an outcome partly to one’s own decision does not imply a rule violation, and a win can attribute part of its result to market conditions.

The comparison uses each trade’s primary attribution, the main explanation the trader explicitly recorded, under the same rule for wins and losses:

primary own-action attribution share, by outcome =
  trades whose primary recorded attribution is own action
  / all trades in that outcome group

Compute separately for wins and for losses, within the comparable
group. Report the counts of mixed, unresolved, and
"no explanation recorded" trades for each outcome next to the share.

Secondary contributing factors can be recorded separately, but they are not part of this comparison. A review that also tallies any-mention shares should keep them under a separate label, since they answer a different question.

The denominator is every trade in the group. A trade that does not count toward the share has either no own-action primary attribution recorded (its primary attribution is external, mixed, or unresolved) or no explanation recorded at all. These are reported separately, and neither is silently counted as an external attribution. Unequal documentation rates between wins and losses, such as long loss notes and one-line win notes, can affect how the shares read. No threshold declares a bias. If the share differs noticeably by outcome among comparable trades, that is a flag to inspect the notes, not a diagnosis.

A worked example (hypothetical)

A trader reviews sixteen trades from one setup over a month. Using the contemporaneous plan and record, all sixteen are classified aligned, and all were labeled standard setups before entry. Eight ended as wins and eight as losses. These shared labels are partial controls: the example makes no claim that size, timing, volatility, liquidity, or execution were the same across the sixteen trades.

This example reports each note’s primary attribution only, as defined above. It does not record whether a note also mentioned a secondary factor, and secondary factors are not part of the comparison. Coding to primary attribution is a simplification of the hypothetical trader’s notes, not an exhaustive causal model. Every note carried a primary attribution of own action or external factor; none was mixed, unresolved, or missing.

Outcome group	Trades	Primary attribution: own action	Primary attribution: external factor	Mixed, unresolved, or no explanation recorded
Wins	8	7	1	0
Losses	8	1	7	0

Primary own-action attribution is 7 of 8 on wins and 1 of 8 on losses. Among trades that share a setup, a process classification, and a pre-trade expectation, the trader’s primary explanations name own action far more often on wins than on losses. That is an attribution asymmetry worth investigating. It does not show that the explanations are false, and it does not show that self-attribution bias, or any particular motive, produced the difference: trades with matching labels can still differ in ways the labels do not capture, and the losing trades may genuinely have met worse conditions.

Applying step 4, the trader checks whether each of those fourteen majority-pattern attributions, which are claims made in the notes, has an identifiable supporting record:

Outcome group	Claim examined	Identifiable supporting record	No identifiable supporting record
Wins	Primary attribution: own action (7 of 8 wins)	5 — the record shows the exit or entry followed the written rule	2 — “good read”
Losses	Primary attribution: external factor (7 of 8 losses)	3 — two timestamped news releases and one fill measurably worse than the quote	4 — “chop” or “unlucky”

Two winner-side own-action claims and four loser-side external claims have no identifiable supporting record. The records that do exist support narrow things. On five wins, they verify the behavior the claim refers to (following the rule), not that the behavior contributed to the profit. On the losses, two records show that an event occurred, and one shows an execution discrepancy without identifying its source. The three external explanations remain the trader’s claims: none of these records independently validates an external-cause attribution or establishes what caused any trade’s result.

This screen finds unsupported explanations. It does not produce a revised attribution gap. Two notes were not screened here (the one external primary attribution on a win and the one own-action primary attribution on a loss), and a revised comparison would require re-coding all sixteen trades under one rule and recalculating each denominator. Until then, the descriptive comparison stays 7 of 8 versus 1 of 8, and an unsupported explanation is not evidence that the opposite explanation is right. Because this is a hypothetical teaching example, no inferential test is performed. The observed difference is descriptive and does not establish a persistent attribution pattern. The practical conclusion is to track the next window with the fixed vocabulary from the start; it does not settle whether the trader is skilled, unskilled, or biased.

What a passing or failing check does not mean

A gap is not proof of bias or motivated reasoning. A real difference in expectations can produce it, and so can real differences in market conditions, execution, size, or timing that the matching did not capture.
No gap is not proof of skill. Symmetric explanations can still be equally wrong; the check tests consistency of the standard, not the correctness of the causal claims.
A small sample is a prompt to keep recording. See why trading data is easy to over-read before treating a share from one month as a trait.
Blaming yourself more is not the goal. Attributing part of a loss to your own action is not the same as calling it a mistake, and filing a rule-aligned loss as a violation is the outcome-bias error in the other direction. The goal is one evidence standard for both.

What to do with a flagged pattern

Choose one response and test it over a defined window, rather than adopting several at once:

Record the pre-trade hypothesis before the result. At entry, write why the trade should work, the conditions you expect, and what would invalidate it. The post-trade explanation is then compared against those contemporaneous records instead of being written from scratch.
Split “own decision” from “own outcome.” Record separately whether the decision followed the plan and what the result was. A rule-aligned loss and a deviating win then never share a label.
Review losses and wins in the same session, with the same questions. Use the same review questions for winning and losing trades, in the same order.
Tie any skill claim to a data standard. Before writing “I’ve improved,” check that the record could show it: a fixed classification, a comparison window, and a sample large enough to survive ordinary variance.

None of these guarantees a better outcome. They make the explanation reviewable.

Where Costante fits

Costante supports session planning, low-friction trade logging, and structured review of recorded trades against the trader’s written plan, including setup type, execution discipline, mistake tags, and outcome quality. That gives a trader plan-adherence and outcome records to hold explanations against.

The attribution symmetry check itself is a manual review practice. A fixed explanation vocabulary, a pre-trade expectation label, coding each note under one rule, and comparing the shares are steps the trader defines and carries out; this article does not describe them as Costante features. Costante does not detect self-attribution bias, code review explanations, match comparable trades, judge whether a stated cause is correct, decide whether a result reflects skill or luck, or determine whether a trade should have been taken. The trader defines the categories, applies the same evidence standard, and draws the conclusion.

Frequently asked questions

Is self-attribution bias the same as overconfidence?

No. Overconfidence is a broader family of miscalibration about one’s own knowledge or ability. Self-attribution is one proposed way overconfidence can build up: asymmetric explanations of results feeding a rising belief in skill. Gervais and Odean’s model describes that pathway; it does not say every overconfident trader arrives there this way.⁴

Is it always a bias when I blame the market for a loss?

No. Markets do produce events a trader cannot anticipate, and a loss can be external. The question is whether the explanation is supported by an identifiable record, and whether the same standard would have been applied had the trade won. Even a supported explanation, such as a timestamped news release, shows that an event occurred, not that it caused the loss.

How is this different from outcome bias?

Outcome bias changes how a decision is classified once the result is known. Self-attribution changes the reason given for the result and, cumulatively, the trader’s belief about their own skill. Both can be present in one review. The outcome-bias article covers the classification error; this article covers the explanation asymmetry.

How many trades do I need before an attribution gap means anything?

There is no universal threshold. The comparison is easiest to read within a group of reasonably comparable trades, for example ones that share a process classification and a pre-trade expectation, and small groups swing widely. Treat the share as a descriptive review indicator, keep recording under the same vocabulary, and see trading data statistical reliability for why small samples mislead.

Should I attribute more of my losses to myself?

Not as a rule. A rule-aligned loss should be classified as aligned, even if the note also attributes part of the result to your own decision, such as choosing one of several entry options the plan explicitly permits. Attributing a result partly to your own decision does not imply a rule violation. The aim is one evidence standard for wins and losses, not a default of self-blame.

Can a trading journal remove the bias?

No. A journal gives the bias a record to be examined against, which is what makes a pattern visible. It does not stop the explanation from being written selectively, and it cannot verify that a stated cause is true.

Sources

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

Hoffmann, A. O. I., & Post, T. (2014). Self-attribution bias in consumer financial decision-making: How investment returns affect individuals’ belief in skill. Journal of Behavioral and Experimental Economics, 52, 23–28. ↩ ↩²
Mezulis, A. H., Abramson, L. Y., Hyde, J. S., & Hankin, B. L. (2004). Is there a universal positivity bias in attributions? A meta-analytic review of individual, developmental, and cultural differences in the self-serving attributional bias. Psychological Bulletin, 130(5), 711–747. ↩
Miller, D. T., & Ross, M. (1975). Self-serving biases in the attribution of causality: Fact or fiction? Psychological Bulletin, 82(2), 213–225. ↩
Gervais, S., & Odean, T. (2001). Learning to be overconfident. The Review of Financial Studies, 14(1), 1–27. ↩ ↩²