Overconfidence and the cost of trading

If you take one chapter from this subject into practice, take this one. The evidence is the strongest in the field and the cost is the largest.

What overconfidence means precisely

Not optimism about markets. Overconfidence is a belief that your own information or judgement is better than it is — too narrow a sense of how wrong you might be.

It has a direct behavioural consequence: if you believe your estimate is sharper than it is, trading on it looks worthwhile. So models of overconfident investors predict excessive trading, and that prediction is testable because turnover is observable.

The American evidence

Barber and Odean sorted roughly 65,000 households into quintiles by monthly turnover, 1991–1996.

Turnover quintile	Annual return, net of costs
Most active 20%	11.4%
Least active 20% (buy-and-hold)	18.5%

A spread of 7 percentage points a year, which over a working life is not a detail but the difference between outcomes. The average individual investor's stock portfolio earned a three-factor alpha of −31.1 basis points per month after costs, about −3.7 points a year.

The mechanism is costs, not stock-picking. The survey's conclusion on a larger 78,000-investor sample is that individual portfolios underperform "largely because of trading costs". The trades were not notably bad; there were simply too many of them, each paying a spread and a commission.

Taiwan gives the aggregate version: individual traders' losses equalled 2.8% of total personal income and 2.2% of GDP, with net returns about 3.8 points below the market — attributable roughly equally to poor selection, commissions and transaction tax, with a smaller role for market timing.

The turnover finding, isolated

The cleanest demonstration that activity rather than ability is the problem comes from a comparison the authors drew between two groups of account holders who traded at different rates. Barber and Odean measured annual turnover of about 80% among men and about 50% among women, with the higher-turnover group earning worse net returns — while gross returns on their trades were similar.

That is the whole argument in one observation, and the mechanism is the finding rather than the grouping. The authors state that neither group appeared to have stock selection ability, and that virtually all of the difference in performance traces to one group trading more aggressively. Same ability to pick, different amounts of trading, different net outcomes — so what the comparison isolates is the cost of turnover, with any question of who trades more left to the overconfidence literature that motivated the test.

The Indian evidence

SEBI studies the equity derivatives segment directly, using the top 13 brokers covering about 96 lakh of roughly 107 lakh unique traders.

Year	Loss-makers	Net losses (₹ crore)	Average per person (₹)
FY22	90.2%	−40,824	−95,517
FY23	91.7%	−65,747	−1,12,677
FY24	91.1%	−74,812	−86,728
FY25	91.0%	−1,05,603	−1,10,069

Net losses widened 41% year on year to ₹1,05,603 crore in FY25, after transaction costs.

The stability of the first column is the finding. Four years, a near-tripling of participants, and the loss-making share sits between 90% and 92% every year. Whatever is happening, it is not a cohort of beginners who are learning. The structure produces the same distribution as it scales.

Working the problem

The "join the 9%" inference fails on several counts.

A pure game of chance with costs produces exactly this. If outcomes were random and every participant paid transaction costs, a large majority would still lose and a minority would still win. So the 91/9 split is not evidence that the 9% have a skill; it is what you would see even if nobody did.

The winners cannot be identified in advance. You could only join them by knowing beforehand who they are, which is the thing the statistic does not tell you. Afterwards, the set is defined by having won.

It is close to zero-sum before costs and negative after. Derivatives gains and losses between participants largely offset; costs do not. The average participant's expected outcome is therefore negative before any question of skill arises, which is why the aggregate is a loss of ₹1,05,603 crore rather than roughly nothing.

Survivorship hides the base rate. The visible traders are the ones still trading, who are disproportionately those who have done well so far. The 91% are mostly not posting.

And the stability of the share answers the "I will learn" version. If experience solved it, the loss-making percentage would fall as the cohort aged. Across FY22 to FY25 it does not move.

What the evidence does not say

It does not say markets are unbeatable — some investors do beat them, and the survey notes measurable cross-sectional differences in skill.

It does not say never trade. Rebalancing, raising cash and correcting a genuine mistake all require trades.

It says the default should be inactivity, because the measured cost of activity is large and the measured benefit is not there. Every trade should have to argue for itself.

The point

Across roughly 65,000 households the most active fifth earned 11.4% a year net against 18.5% for the least active — a 7-point gap driven by trading costs rather than by bad selection, since gross returns on their trades were similar. SEBI finds 91% of individual derivatives traders lost money in FY25, a share stable between 90% and 92% for four years, which rules out the explanation that participants are simply learning. A majority losing in a near-zero-sum game with costs is what chance alone would produce, so the 9% are not a group you can join by intending to.