Luck, Skill and What a Result Is Attributed To

The attribution, and why it decides everything downstream

Attribution is the assignment of a cause to an outcome, and it is where learning either happens or does not. A trade that loses $2,000 because the stop was placed inside the ordinary noise carries one lesson; the same loss because the position was 20% of the account carries another; the same loss because the whole sector fell 6% carries a third. The result is identical in all three cases and the correct response is different, so a trader who attributes by outcome cannot get better even with perfect records. The systematic error is self-serving. People attribute their successes to their own ability and their failures to circumstance — a bias documented in the attribution literature and reproduced wherever a person is explaining their own performance. In an account it produces a record that reads like a highlight reel: the winners confirm the method, the losers were unpredictable, and the two conclusions are drawn from one dataset with opposite rules. The mirror image is a trader who attributes everything to luck and changes nothing, which is the same failure with the sign flipped. The correction is mechanical rather than motivational, and it has three parts. Attribute by mechanism: name the thing that caused the outcome — the size, the stop distance, the sector, the beta, the schedule — and check whether it was inside the trader’s control. Separate the market’s part from the process’s part: a portfolio up 20% in a market up 22% did not demonstrate anything, and one up 4% in a market down 12% did, because the second result required the holdings to behave differently from the average. Then insist on a sample, since a single trade attributes to nothing, and a written record of forecasts is the only way to know whether the ability being credited actually exists. Three results, three attributions — Portfolio +20%, market +22%: The environment did the work; the process added nothing — and lagged ← · Portfolio +4%, market −12%: The process earned the difference: the holdings did not do what the average did · Twelve trades at +4R against a +0.2R expectancy: Not yet attributable: the draw is small enough that noise dominates The first row is the one that flatters and misleads at the same time — a positive number that contains no evidence about the decision-making, and the most common way a rising market manufactures confidence.

Where experience does and does not produce skill

There is a widely held assumption that time in markets produces skill. The evidence says the assumption is conditional. Environments teach when they provide valid cues, a rapid and unambiguous outcome, and enough repetitions to distinguish signal from noise — the conditions under which professionals in other fields (chess, firefighting, anaesthesia) genuinely develop fast, accurate judgement. Where the cues are not predictive, where feedback is delayed or confounded, and where the outcome is mostly luck, experience produces confidence without accuracy, which is the most dangerous combination available. Markets sit at the far end of that spectrum. The feedback is delayed by position and by horizon, noisy enough that a losing decision and a winning decision look identical for months, and adversarial in the sense that any durable pattern invites competition that removes it. A trader can therefore accumulate two decades of experience in conditions that do not teach, and be no better calibrated than when they started, while feeling considerably more certain. That finding is not a reason to give up on learning, because it also names what does work. Feedback can be manufactured where the environment does not supply it: trade in a written, checkable way so the outcome is attributable to something specific; keep the record of forecasts so calibration can be measured rather than assumed; study the base rate for the class of situation rather than accumulating case memory; and treat the survivorship in every sample you are shown as a variable, because the funds, traders and strategies still visible are the ones that survived a process with a large random component. The distinction the whole rung rests on is between experience, which is time spent, and practice, which is time spent with valid feedback on a defined skill. • Attribute by mechanism, before attributing to ability: size, stop, sector, beta, schedule. • Separate the market’s contribution from the process’s contribution. • Insist on a sample; one trade attributes to nothing. • Manufacture feedback where the environment does not supply it, using written forecasts. • Treat survivorship as a variable in every sample you are shown. A distinguished record is evidence about a period and only weakly about the next one, and the temptation is greatest at exactly the moment it is least informative — after a run. Mean reversion is not a forecast that a good manager will fail; it is the statement that part of the result was not caused by the manager, and that part does not carry forward.

How long a record would have to be

The intuition that a good record over a few years demonstrates skill does not survive an arithmetic check. The right question is not whether outperformance happened but whether it was **large relative to how much of it chance alone would produce**. That comparison has two inputs: the size of the average outperformance, and how much the individual periods vary around it. When the period-to-period variation is large, as it is for any equity strategy, a run of good years is well within what a no-skill process generates. The statistical machinery is the standard error of the mean, and its message is blunt: with the dispersion typical of stock returns, distinguishing a genuinely skilled manager from a lucky one takes a **very** long record — for many strategies, decades. That result has a corollary that matters more than the conclusion. If the record of an ordinary investor or a discretionary trader is too short to separate skill from luck, then **the record cannot be the primary basis for confidence in the process**. What can be evidence in a short record is the part that does not depend on the market: whether the rules were followed, whether the sizes were what the plan said, whether the exits were taken. Those are countable from day one because they do not require a large sample to observe — they are facts about behaviour, not statistical estimates. This is the practical reason this subject keeps returning to process: outcomes need years to speak, and decisions speak immediately. The third thing a short record cannot do is survive survivorship. Any group of traders you can observe — the people you follow, the funds that exist today, the strategies with published track records — is the subset that did not quit or blow up, which biases every aggregate you compute from it upward. If the record you are comparing yourself with comes from that group, the comparison embeds a selection effect that no amount of careful reading of the numbers removes. What a short record can and cannot show — Three years of outperformance, dispersed returns: Not yet evidence of skill — well inside what chance produces · Rules followed, sizes as planned, exits taken: Evidence, available immediately, and independent of the market · Comparing against the funds or traders still standing: Biased upward by survivorship before you start ←

Self-attribution: the credit you keep, the blame you export

Once a result has been attributed to skill or to luck, the attribution decides what happens next — and the attribution people make about their own performance has a direction. Successes are claimed as the product of judgement; failures are assigned to circumstance. The finding is one of the most reliable in the psychology of performance, and markets are the domain where it survives longest, because the feedback is noisy enough to accommodate it indefinitely. The trading version is easy to recognise in other people: a winner becomes “my thesis”, and a loser becomes “the market was irrational”, “the algos”, “the news”, “an unfair halt”. Over a year of decisions that produces a record in which nothing was ever wrong and the world simply misbehaved — and no learning occurs, because the only thing that generates learning is a mistake that has been located. What makes markets unusual is the noise. In an activity with a clean signal — a chess rating, a golf score — the scoreboard corrects the attribution error whether or not the person wants it to. In trading a well-reasoned loss and a badly reasoned loss arrive as the same number, so the ambiguity is resolved by preference rather than by evidence. The bias does not need to survive the data; it only needs the data to be unable to speak. The symptom to look for in yourself is not a sentence but a gap. The narrative version of your record and the actual log diverge, and the size of the divergence is a rough measure of the bias. Recall is reconstructive: the positions that worked are remembered as decisions, and the ones that were lucky are remembered as skill on both sides of the ledger. The countermeasure is to attribute the decision before the outcome exists. Write the thesis, the mechanism and the falsifier at entry; then, at review, ask whether the mechanism operated rather than whether the trade made money. For the second half of the bias — the export of blame — the test is a single sentence that must not mention anything outside your control: what specifically did I get wrong, in this decision, that I could have got right? If the sentence cannot be written without the market in it, the attribution has not been made. The mirror error deserves equal space, because it is common among careful people: attributing a good result to luck and refusing to credit a process that worked. That produces the opposite paralysis — an inability to learn what is working — and the goal is accuracy in both directions rather than humility in one. • The bias has a direction: credit inwards for wins, blame outwards for losses. • Noisy feedback lets the bias survive, because outcomes cannot arbitrate the question. • Compare the remembered record with the written one; the gap is the measurement. • Attribute the decision at entry, and require the explanation of a loss to name something you controlled. A useful monthly exercise: take the three worst and the three best trades of the quarter and write the mechanism that produced each, without referring to the P&L. If the best three read as skill and the worst three read as weather, the attribution is doing the remembering.

Sources

• Self-serving biases in the attribution of causalityMiller & Ross (1975), Psychological Bulletin
• Persistence in mutual fund performanceCarhart (1997), Journal of Finance
• Conditions for intuitive expertiseKahneman & Klein (2009), American Psychologist