Overconfidence, and what it costs in trading

The measured result

Barber and Odean sorted individual accounts by turnover — how much of the portfolio was traded — and compared net returns across the range.

The heaviest traders did not pick much worse stocks. What separated them was what trading costs: after commissions and spreads, higher turnover produced meaningfully lower net returns. The activity itself was the expense.

A companion paper found the same thing along a different cut. Partitioning 35,000 households by gender, men traded about 45% more than women, and trading reduced men's net returns by 2.65 percentage points a year against 1.72 for women — a gap in cost, produced by a gap in activity.

Their explanation for both is overconfidence: the belief that one's own information is better than it is, which produces the belief that acting on it is worthwhile.

Why more information makes it worse

The uncomfortable part is that overconfidence rises with information faster than accuracy does.

Give someone five facts about a company and they will make a forecast. Give them fifty and the forecast barely improves — but their confidence in it climbs substantially. Kahneman calls this the illusion of validity: the feeling of understanding tracks the coherence of the story you can tell, not the evidence behind it.

For an investor with access to a research platform, this is the failure mode to watch. Depth of data is genuinely useful and it is also, reliably, a confidence machine.

The tell

The practical signal is not "am I confident?" — everyone answers yes. It is:

• Can you state what you might be wrong about, specifically? Not "the market could fall" but the particular assumption that carries the thesis.
• Would you accept a bet at your own stated odds? If you say 80% and would not take that price, you do not believe 80%.
• Has your confidence moved recently, and on what? A view that never moves is not being updated by anything.
• How much did you trade last year, and what did it cost? That number is knowable and most investors have never looked at it.

How this connects to Pythia

Two design choices here follow from this evidence.

Scores refuse rather than guessing. A company with no sector bucket gets no PAS; a valuation lens declines when the company sits outside the regime its worked examples came from. A confident number for a case we cannot measure would be the platform being overconfident on your behalf.

And your forecast calibration is scored over time — not whether you were right about one company, but whether things you called 80% happened about 80% of the time. That is the only honest answer to "am I overconfident?", and it requires a record rather than an impression.

The limits of this idea

The Barber–Odean result is about turnover and net returns in retail accounts. It is not a claim that trading is always wrong, that low activity is a strategy, or that confidence is a vice. Rebalancing, tax management, and acting on a thesis that genuinely broke are all trades worth their cost.

Nor does the finding license the opposite error. Doing nothing is a decision too, and a portfolio nobody has examined in three years is not disciplined — it is unattended. What the evidence supports is narrow: activity has a price, that price is usually invisible to the person paying it, and confidence is a poor guide to whether it was worth paying.