Confidence, Overconfidence, and How to Tell

The short answer

"Am I overconfident?" is not one question. Research distinguishes at least three things that go by that name:

• Overestimation — thinking you performed better than you did.
• Overplacement — thinking you performed better than other people did.
• Overprecision — being too sure that your estimate is close to right.

They are measured differently and they do not move together. In the experimental work, the correlations between the three measures within the same people were close to zero. Someone can be badly overprecise and simultaneously under-place themselves against others. Asking yourself whether you are "overconfident" will return an answer about whichever one happens to be salient, which is why the answer feels unreliable.

The useful consequence is practical. Since you cannot distinguish them by introspection, and since the three behave independently, the only route to an answer is a record of what you predicted and what actually happened.

What the research separates

The three-way distinction comes from work reconciling a literature that had been using one word for different things. Overestimation is assessed by comparing a person's judgement of their own performance to that performance. Overplacement is assessed by comparing their judgement of their standing relative to others with their true standing. Overprecision is assessed differently again: people are typically asked for a numerical answer and a confidence interval around it, and the finding is that those intervals come out too narrow — people are too sure they know the answer.

Three results from the experimental work are worth carrying away.

The three measures were close to uncorrelated. Within the same sample, overestimation and overplacement correlated at around 0.06, overestimation and overprecision at around 0.24, and overplacement and overprecision at around 0.03.

No personality measure predicted them. The researchers compared their three overconfidence measures against seventeen individual-difference measures — extraversion, conscientiousness, neuroticism, narcissism, self-esteem, cognitive reflection and others. Three correlations reached the conventional threshold, but after adjusting for the number of comparisons being made across fifty-one tests, none was significant.

Two of the three are weakly measured. The reliability of the overestimation and overplacement measures in that study was low — around 0.21 and 0.29 — while overprecision was high at around 0.95. That is a caution about the first two rather than a dismissal, and it is the kind of limit this field usually leaves out.

Put together: there is no overconfident personality type to recognise in yourself or in anyone else, and the thing you are most likely to be wrong about — your certainty — is the one you are least likely to notice.

The one clean test in trading

The best evidence connecting overconfidence to trading does not ask anybody how confident they are. It uses a group difference instead.

Theoretical models predict that overconfident investors trade excessively. Barber and Odean needed two groups that differ in overconfidence for reasons unrelated to markets, and used gender, on the basis of psychological research finding men more overconfident than women in domains such as finance. Across more than 35,000 households at a large discount broker from February 1991 to January 1997:

• Men traded 45 per cent more than women.
• Trading reduced men's net returns by 2.65 percentage points a year, against 1.72 points for women.
• Among single account holders, single men traded 67 per cent more than single women, and gave up a further 1.44 percentage points a year.

The design is what makes this persuasive. The prediction was made in advance, it could have failed, and the direction and magnitude both came out as the theory said. It does not show that any particular man was overconfident or that any particular woman was not; it shows that a group difference in a measured trait tracked a difference in trading behaviour and in cost.

The mechanism by which that cost arrives is frequency, which is the subject of why people trade too often.

Why you cannot answer this by asking yourself

Three reasons, each from the evidence above.

1. The question is ambiguous. Three constructs, one word, near-zero correlation between them. "Yes" and "no" are both incomplete answers.
2. Overprecision is the one that matters most for trading and the hardest to feel. A trade is an interval judgement: you are implicitly saying where price will and will not go. Intervals being too narrow is precisely the documented failure.
3. The trait measures do not help. Being a cautious person by temperament predicted nothing in the data. The null result cuts both ways — it is also why "I know I'm not arrogant" is not evidence.

What this leaves is uncomfortable but clear: the only instrument available is a record of predictions and outcomes made before you knew the result.

What to record, to get an answer

These are the three measures translated into things a trading log can hold. The practical setup is in what to actually record in a journal.

For overprecision — the interval test. Before entry, write the range you expect price to stay within over the trade's horizon, and the confidence you would put on it. Afterwards, count how often it was breached. If you state ninety-per-cent ranges and they are wrong far more than one time in ten, you have measured overprecision in yourself. This is the single most informative thing in this article, and it takes one extra line per trade.

For overestimation — the pre-commitment test. Before a period, write down what you expect your results and your rule-following rate to be. Compare afterwards. Not against the market: against your own prior statement.

For overplacement — mostly, don't. It is the least measurable of the three outside a lab and the least actionable. Knowing whether you are better than other traders does not tell you what to change.

And the behavioural proxy. If the Barber and Odean mechanism applies, the observable consequence of overconfidence is trading volume. Your turnover is a number you already have. A rise in it that you cannot attribute to a change in conditions is the most accessible signal available.

What this does not mean

It does not mean confidence is bad. Nothing in this research says a person should be less certain across the board; the overprecision finding is about calibration, and a systematically under-confident trader has a different problem with the same cause.

It does not mean you can diagnose other people. If seventeen personality measures failed to predict overconfidence in a controlled study, your read on a stranger's temperament from their posts is not going to.

And it does not mean overconfidence is why most retail accounts lose money. The regulators' figures — 74 to 89 per cent of retail investor accounts typically losing on CFDs across EU jurisdictions — have several mechanisms underneath them, of which costs are demonstrably large. We traced the loss statistics themselves in the numbers this industry repeats, and the wider map of what has actually been established is in our guide to the behavioural evidence.

Limits

The three-way distinction and the correlation figures come from experimental work with student and online participant samples, not from traders. The Barber and Odean study is of American equity investors in the 1990s and uses gender as a proxy for a trait rather than measuring the trait directly, which is a well-known limitation of that design. We have not found a study measuring overprecision in traders and relating it to their results.

What is solid is the distinction itself, the near-independence of the three measures, and the finding that trading volume tracked a group difference in overconfidence at a substantial cost.

Nothing here is advice. Trading leveraged products carries a high risk of losing money quickly.