Men in one large US brokerage sample traded 45% more than women. That cost them 2.65 percentage points a year in net returns, against 1.72 points for the women. Their stock picks weren't measurably worse. They simply acted on them more often.
That gap is the cleanest published test of an old idea: that investors trade far more than their circumstances warrant because they overrate what they know. Rebalancing, tax and cash needs can't explain the volume we observe. Something psychological is filling the gap, and overconfidence is the leading candidate.
How much turnover are we actually talking about
Barber and Odean's 1999 working paper on a large US discount broker put the average household's annual common stock turnover above 75%. The quintile that traded most turned over more than 250% of its portfolio a year. As a reference point, the New York Stock Exchange itself reported turnover of 76% in 1998.
Turnover of 75% means the average household replaced three quarters of its share portfolio every twelve months. Liquidity needs don't move that fast. Odean's blunt version of the objection is that liquidity shocks as an explanation for 250% annual turnover belie common sense.
The costs of that activity in the 1990s were heavy. In the same sample, the trade-weighted round trip cost roughly 1% in bid-ask spread and about 3% in commissions. In Odean's earlier study of 10,000 accounts at a nationwide discount broker, the average commission was 2.23% on a purchase and 2.76% on a sale, plus an estimated 0.94% effective spread. Round trip: about 5.9%.
Where the money went
Here's the part that shapes everything after it. Across the broker sample, gross returns were fine. The market's annualised geometric mean return was 17.9%; the average household earned 18.7% before costs and 16.7% in aggregate after them. Barber and Odean also report very little difference in gross performance between households with monthly turnover above 8.8% and those who barely traded.
Read narrowly, that says the damage is friction, not judgement. Trade less and you keep more, because you pay less. It's the same accounting that drives the gap between what funds return and what bad timing costs the investors who own them.
But Odean's 1999 paper tested something sharper: whether the stocks these investors bought beat the stocks they sold at all, before a penny of cost. Over the 252 trading days after a trade, market-adjusted returns to purchases came in 3.22 percentage points below returns to sales, with a bootstrapped p-value under 0.001. The buys weren't good enough to pay for the trading. They weren't good enough full stop.
He then stripped out the innocent reasons to trade. He kept only purchases made within three weeks of a sale, only sales made at a profit, only sales of a complete position, and only cases where the stock bought was the same size decile or smaller than the one sold. That removes most liquidity, tax-loss, rebalancing and risk-reduction motives at once. The gap got worse, not better: 5.82 points over a year, on 7,503 purchases and 5,331 sales.
Odean's own reading is stricter than the popular summary. Overconfidence about the precision of your information, on its own, only predicts that you'll pay too much in costs. To lose money before costs, you have to be misreading the information itself.
The gender test, and what it does and doesn't show
The 2001 Quarterly Journal of Economics paper split more than 35,000 discount brokerage households by the gender of whoever opened the first account, then tracked them from February 1991 to January 1997. Psychologists had already found that confidence gaps between men and women are largest for tasks seen as male-coded, and largest again where feedback is slow and noisy. Stock picking is both.
Mean monthly turnover came out at 6.41% for men and 4.40% for women. Among single households the gap widened: 7.05% against 4.22%, or 67% more trading. Net of costs, men gave up 2.65 percentage points a year to trading and women 1.72.
The detail that matters most is the one usually left out. Stocks men bought underperformed the stocks they sold by 20 basis points a month. For women the figure was 17 basis points. The difference between those two isn't statistically significant. Men didn't choose worse. They chose more often, and each choice carried a cost.
The authors also had survey evidence. Gallup ran a poll for PaineWebber fifteen times between June 1998 and January 2000, roughly 1,000 respondents each time. Both men and women expected their own portfolios to beat the market over the coming year. Men expected to beat it by 2.8 points, women by 2.1, a difference significant at t = 3.3. Almost nobody expected to be average.
Two caveats belong next to those numbers. The sample is discount brokerage customers, who are not the general investing population, and the return window closes in January 1997. Odean also notes that closed accounts were not replaced, so the later years carry some survivorship bias toward investors who did well.
The natural experiment nobody planned
The strongest causal evidence comes from a change in plumbing. Barber and Odean tracked 1,607 investors at the same broker who switched from phone-based to online trading during the 1990s.
Before the switch, these people were good. They beat the market by more than 2% a year. After it, their average annual turnover rose from 73.7% to 95.5%, their speculative turnover nearly doubled from 16.4% to 30.2%, and they lagged the market by more than 3% a year. Size-matched investors who never went online saw turnover drift down over the same window, from 53.2% to 48.2%.
The obvious explanation is that trading got cheaper, so they did more of it. The paper checks. Round-trip commissions for the switchers fell from 3.265% to 2.507% and round-trip spreads from 1.133% to 0.859%. Cheaper, yes. Not cheap enough to explain a swing of five percentage points in relative performance. The authors reach instead for self-attribution: people who had just been right about the market concluded they were skilful, and a faster tool let them act on it.
The case against reading turnover as overconfidence
Gender is a proxy, not a measurement. Grinblatt and Keloharju went after the thing directly, using Finnish trading records linked to tax filings, driving records and the psychological profile every Finnish male sits at around age 19 for the armed forces. One scale on that test measures self-confidence from 1 to 9. They treat the part of it left unexplained by measured intellectual ability, income and life outcomes as overconfidence.
They also had a second variable: speeding convictions, as a proxy for sensation seeking. Each extra ticket raised the probability of trading by 4.7%, the number of trades by 9.8%, and turnover by roughly 3.6 percentage points a year from a base near 40%.
When both variables ran together, the result cut against the standard story. Sensation seeking stayed significant for the number of trades and for turnover. Overconfidence was significant for whether someone traded at all, but only marginally insignificant for turnover itself. The authors argue that it's equally plausible the Barber and Odean gender result reflects a gender difference in sensation seeking rather than in overconfidence.
Two other studies point the same way, and I've read only their published abstracts, so take them at that weight. Glaser and Weber surveyed around 3,000 online broker clients and matched 215 of them to trading records. Investors who thought they were above average traded more. Miscalibration, which is the version of overconfidence the theory actually models, was unrelated to trading volume. And Cueva, Iturbe-Ormaetxe, Ponti and Tomás ran an experiment with incentivised confidence measures taken before a simulated market. Men were more confident and traded more, but the abstract reports that confidence differences did not account for the trading gap, and that risk aversion, financial literacy and competitiveness were unlikely to either.
So the pattern is solid and the label is contested. Trading more predicts earning less. Whether the driver is overconfidence, thrill-seeking, boredom or something not yet named is genuinely open.
The other objection: trading is nearly free now
The second challenge is arithmetic. If most of the penalty was friction, and friction has collapsed, the penalty should have collapsed with it.
The Taiwan evidence makes that case better than any critic could. Using every trade on the Taiwan Stock Exchange from 1995 to 1999, Barber, Lee, Liu and Odean found individual investors lost 3.8 percentage points a year in aggregate, worth 2.2% of Taiwan's GDP and 2.8% of total personal income. The breakdown is the interesting bit: gross trading losses 27%, commissions 32%, transaction taxes 34%, market-timing losses 7%. Two thirds of the damage was cost and tax. Institutions, meanwhile, gained 1.5 points a year after their own commissions and taxes.
Strip commissions to zero and that 3.8-point penalty shrinks a long way. US brokers moved to commission-free equity trading, and Taiwan's exchange turnover in that window averaged 294% a year, which no modern retail market approaches. Frictions are lower and turnover comparisons across eras aren't like-for-like.
Two things blunt the objection. First, taxes didn't vanish. UK investors still pay 0.5% Stamp Duty Reserve Tax on electronic share purchases, and 0.5% Stamp Duty on stock transfer forms above £1,000, per HMRC guidance current in August 2026. That's a real cost on every buy, and it scales directly with how often you trade.
Second, and more damaging, the pre-cost evidence never depended on commissions. Odean's 3.22-point gap is measured on market-adjusted returns before trading costs. The zero-commission era has produced its own version. Barber, Huang, Odean and Schwarz tracked Robinhood user counts from May 2018 to August 2020 and defined a herding episode as the top 0.5% of daily percentage increases in users holding a stock, about ten a day and roughly 5,000 in all. Those stocks lost 3.55% in abnormal returns over five days and 4.74% over twenty. Nearly two thirds were negative at day twenty. Robinhood users concentrated 35% of net buying in ten stocks against 24% for retail investors generally.
Commissions there were zero. The losses weren't. That paper measures user counts rather than dollar trades, it covers a wild 26-month window, and its subject is attention rather than confidence, so it's supporting evidence and not a replication.
High turnover isn't uniformly stupid
One more finding keeps the picture honest. Barber, Lee, Liu and Odean also studied Taiwanese day traders from 1992 to 2006. In an average year 360,000 individuals day traded, accounting for 17% of exchange volume, and about 13% of them made money net of fees in a typical year.
Sort them by last year's returns, though, and the top 500 go on to earn 49.5 basis points a day before fees and 28.1 after. The worst-ranked go on to lose 17.5 before fees and 34.2 after. Fewer than 1% of day traders, roughly 1,000 out of 360,000, repeat reliably. Skill exists. It's just rare enough that assuming you have it is exactly the error being described.
What would change the conclusion
Three findings would move this materially.
A modern account-level dataset with near-zero commissions showing high-turnover retail accounts matching low-turnover ones after costs would kill the friction channel outright, leaving the selection channel to carry the whole result. That test hasn't been published at the scale of the 1990s brokerage studies, and the Robinhood work is a poor substitute because it tracks stocks rather than accounts.
A run of direct overconfidence measures that keeps failing to predict turnover would mean the mechanism is mislabelled even where the pattern holds. Grinblatt and Keloharju, Glaser and Weber, and Cueva and colleagues each land partway there already. If sensation seeking wins that race, the practical implication changes: you'd be managing a taste for stimulation, not a belief about skill.
And a failure to reproduce Odean's pre-cost gap outside the 1987 to 1993 US discount-broker window would undercut the strongest claim in the literature. The Taiwan and Finland work is consistent with it, but those are different markets with different microstructure.
What wouldn't change it is cheaper dealing. Free trades fix the commission line and nothing else. If the stocks you buy trail the stocks you sell by three points a year before costs, paying nothing to make the swap doesn't help. That's also why the pre-commitment literature matters more than the fee literature here, and why the evidence on writing an investment policy statement and the comparison of annual rebalancing against 5% threshold bands both approach the problem by removing discretion rather than by making it cheaper to exercise.
The finding that has survived every recut since 1999 is narrow and durable. Across large samples, in several countries, investors who trade more end up with less, and the gap doesn't come from picking worse. It comes from picking more often on evidence that was never strong enough to act on.