Overconfidence Bias

Overconfidence bias is the tendency to hold more confidence in one's own judgment, ability, or forecasts than accuracy warrants. The standard treatment in psychology is Don Moore and Paul Healy's "The trouble with overconfidence," in Psychological Review 115(2) in 2008, and its central contribution is a distinction rather than a measurement: what had been treated as a single bias is three separable things.

Overestimation is thinking your own performance, ability, or chance of success is higher than it is, judged in absolute terms. Overplacement is thinking you rank above others, which is the familiar better-than-average effect. Overprecision is excessive certainty that your own estimate is close to correct, and it shows up as a confidence range that is too narrow. Ordinary usage collapses all three into "overconfidence," which matters because they behave differently and because only one of them reliably damages a portfolio.

The three constructs are dissociable, and that is the finding worth carrying. Moore and Healy's argument is that the apparent contradictions in decades of overconfidence research resolve once the three are separated, because they can move in opposite directions on the same task. On a difficult task people tend to overestimate their own absolute performance while underplacing themselves relative to others; on an easy task the pattern reverses. So a person can simultaneously overrate what they scored and underrate where they finished, which is impossible if overconfidence is one thing. Any explanation that treats overconfidence as a single dial gets this wrong, and most consumer explanations do, usually by describing overplacement and calling it the whole phenomenon.

Overprecision is the financially expensive one, and it is expensive through position size rather than through direction. A forecast has a central estimate and a range around it. Overprecision leaves the central estimate alone and shrinks the range, and almost every consequential portfolio decision runs off the range rather than the midpoint. How much of one holding is too much, how large a cash reserve has to be, whether a plan survives a bad decade: each of those is a question about the width of the distribution. Two investors with identical forecasts and different ranges will build substantially different portfolios, and the one with the narrower range will hold more of whatever they are most certain about.

The trading evidence is about behavior rather than about beliefs, and it points the same way. Brad Barber and Terrance Odean, in "Trading Is Hazardous to Your Wealth" in the Journal of Finance 55(2) in 2000, examined the accounts of tens of thousands of households at a discount brokerage and found that the households that traded most actively earned the lowest returns net of costs, with gross returns that did not justify the turnover. Their 2001 paper "Boys Will Be Boys," in the Quarterly Journal of Economics 116(1), found that men traded more than women and that returns followed the same ranking. The papers do not read minds, and overconfidence is offered as the explanation for the trading rather than measured directly. What they establish is a specific and useful thing: activity itself was costly, and the confidence required to act frequently was not accompanied by the accuracy that would have paid for it.

Where the three constructs surface in ordinary financial life. Overestimation appears in projections of one's own future behavior, including the saving rate you expect to reach next year and the spending discipline you expect to bring to retirement. Overplacement appears in the belief that you will do better than average at picking funds or timing an exit, which is a claim about other participants rather than about a market. Overprecision appears in single-point plans: one assumed return, one assumed inflation rate, one assumed retirement date, each stated as a number rather than a range, with a plan built to survive exactly that combination.

The countermeasures are structural, and the reasoning for that follows from the construct rather than from any tested intervention. Because overprecision is a claim about range width, the only version of it you can audit is one you wrote down. A range recorded before an outcome is known can be checked afterward; a range recalled afterward tends to have widened to accommodate what happened, which is a separate documented effect. So the workable habits are recording the range and the decision rule in advance, in an investment policy statement or anywhere durable, and then counting how often reality landed inside it. Counting is the part that does the work, since it converts a belief about your own judgment into a number.

Calibration is the mechanical test, and it is arithmetic rather than introspection. Set ten ranges you are 90% confident about, on questions whose answers can be checked. If your ranges are honest, about nine of the ten should contain the true answer. If far fewer do, the ranges are too narrow, which is what overprecision means operationally. The exercise says nothing about whether your central estimates are good; it says whether your certainty is.

A hypothetical example of what range width does to a decision. Amara has $250,000 in investments, of which $150,000, or 60%, is her employer's stock. She is confident the position could not fall by more than 25%, and under that assumption her worst case is a $37,500 loss ($150,000 × 0.25), leaving $212,500. Nothing about that arithmetic is wrong. The range is.

A single company's shares can fall much further, and rank behind every creditor if the company fails. At a 70% decline the loss is $105,000 ($150,000 × 0.70), leaving $145,000, which is a 42% fall in her total investments ($105,000 ÷ $250,000) caused by one holding. Her forecast of the most likely outcome never entered the calculation. What changed the answer was how wide a range she was willing to entertain, and the position size that survives the wider range is smaller than the one she chose.

Pros

• Separating the three constructs turns a vague warning into three specific questions, each of which points at a different decision.
• Overprecision is measurable against your own record, so it is one of the few biases you can actually audit rather than merely acknowledge.
• Some confidence is required to act at all, and the research on trading concerns frequency and cost rather than the act of investing.
• The countermeasure, writing a range and a rule down in advance, costs nothing and leaves a record.

Cons

• It is invisible from the inside, because the feeling of certainty is identical whether or not it is justified.
• The popular version of the bias is the better-than-average effect, so an ordinary explanation of overconfidence often never mentions overprecision, which is the one that sets position size.
• Ranges recalled after the fact are unreliable, so the audit only works if the range was recorded before the outcome.
• Overprecision compounds with concentration: the more certain you are, the larger the position, and the larger the position the more a wrong range costs.
• Being told about it does not visibly reduce it, which is why the fixes are procedural rather than educational.