What Is Behavioural Finance? The Biases and the Evidence

Terrance Odean's 1998 study of 10,000 US brokerage accounts, covering 1987 to 1993, compared two kinds of stock on every day an account made a sale. A stock is "on a gain" when its price is above what the account paid. In the study, 14.8% of stocks on a gain were sold, against 9.8% of stocks on a loss.

That gap is an example of the disposition effect, defined as selling stocks on a gain more readily than stocks on a loss. It is one finding in behavioural finance, the study of how people's money decisions differ from what standard economic models assume. This post reports each finding as a named study with its sample and period.

Nicholas Barberis and Richard Thaler's 2002 survey states the field's core idea this way: "some financial phenomena can plausibly be understood using models in which some agents are not fully rational".

The model it is set against is expected utility theory, in which a person picks the option with the best probability-weighted result.

These studies cover households and experiment participants. They say nothing about rule-based methods, which What Is Systematic Investing? and Systematic vs Discretionary Investing define.

Prospect theory is Kahneman and Tversky's account of how people choose between risky options, with outcomes judged as gains or losses. Their 1979 paper in Econometrica gave hypothetical choices to students and university faculty in Israel, Sweden and the United States.

In Problem 3, 95 respondents chose between a sure 3,000 Israeli pounds and an 80% chance of 4,000, which averages 3,200. Of them, 80% took the sure amount.

In the mirror-image Problem 3', 95 respondents chose between a sure loss of 3,000 and an 80% chance of losing 4,000. Now 92% took the gamble. The authors called the flip the reflection effect.

The authors also argued that a loss weighs more than an equal gain, later called loss aversion. In a 1992 paper, Tversky and Kahneman reported a median loss aversion coefficient of 2.25 from 25 graduate students at Berkeley and Stanford. That means a loss weighed 2.25 times an equal gain.

A 2024 meta-analysis by Alexander Brown, Taisuke Imai, Ferdinand Vieider and Colin Camerer pooled 607 estimates from 150 articles, reported from 1992 to 2017. Its mean was 1.955, with a 95% interval of 1.820 to 2.102.

The disposition effect is the pattern of selling stocks on a gain more readily than stocks on a loss. Hersh Shefrin and Meir Statman named it in 1985 in "The Disposition to Sell Winners Too Early and Ride Losers Too Long: Theory and Evidence".

Odean's records came from a large US discount brokerage, January 1987 to December 1993. On each day an account sold a stock, he counted the stocks sold and the stocks kept, each on a gain or a loss.

Over the whole period, 14.8% of gains were realised (sold), against 9.8% of losses. In December the gap reversed, with 10.8% of gains sold against 12.8% of losses.

Chart 1 shows the share of gains and losses sold in each period. Excluding December, 15.2% of gains were sold against 9.4% of losses.

Odean's Table VI (returns after the sales) followed the stocks afterwards. Over the next year, stocks sold at a gain outperformed stocks kept at a loss by 3.4 percentage points, measured against a market index.

The pattern is better established than its cause. Itzhak Ben-David and David Hirshleifer's 2012 paper studied 77,037 US brokerage accounts from 1990 to 1996. It found little sign of a jump in selling at zero profit, where a preference for realising gains would predict one. They concluded that the preference does not mainly drive the effect.

Brad Barber and Terrance Odean followed 66,465 households at a large US discount brokerage from 1991 to 1996. They split households into fifths by turnover, the share of a portfolio sold and replaced.

Before trading costs, the stocks held by each fifth earned between 18.5% and 18.7% a year. After the bid-ask spread (the gap between buying and selling prices) and commissions, the most active fifth earned 11.4% and the least active fifth 18.5%. The average household earned 16.4%, against 17.9% for a market index.

Chart 2 shows the net returns, with the index for comparison. The authors called the pattern consistent with overconfidence, a belief in one's own judgement that exceeds its accuracy. Trading records show what households did but cannot show why.

Anchoring is the pull of an irrelevant number on a later estimate. In 1974, Tversky and Kahneman spun a wheel that set a starting number between 0 and 100. Subjects then estimated the share of United Nations members that are African countries.

Groups given 10 as a starting point had a median estimate of 25%. Groups given 65 had 45%. The paper gives no sample size.

Richard Klein and colleagues' 2014 Many Labs project ran four anchoring questions across 36 samples with 6,344 participants. All four effects replicated, with anchoring among the largest of the 13 effects tested.

Market evidence is indirect, because prices record trades but not the numbers people had in mind. George and Hwang reported in 2004 that nearness to a stock's 52-week high helps forecast its later return.

A 52-week high is a candidate anchor, so a check on 27 exchange-traded funds (ETFs) in our own price data, with dividends counted, asks the same question.

At each month-end from August 2002 to August 2026, the funds were ranked by price as a share of their 52-week high. That share is 1 minus the drawdown from the high.

Each month ranked 12 to 27 funds, as not all existed in 2002. The top third beat the bottom third over the next 21 trading days by 0.20 percentage points a month on average (0.85% against 0.65%).

The t-statistic (the gap divided by its standard error, which comes from the standard deviation) was 0.72, below the conventional threshold of about 2. The gap is indistinguishable from chance.

How to Read a Backtest covers the cautions on testing past prices.

Don Moore and Paul Healy's 2008 review cites studies in which people gave a range they were 90% sure contained the right answer. The right answer fell inside it less than half the time.

Markus Glaser and Martin Weber's 2007 study matched 215 online brokerage customers' questionnaire answers to their trades. Those who rated their investment skill or past performance above average traded more. A measure of miscalibration, meaning confidence beyond accuracy, was unrelated to trading volume.

A replication repeats a study with new participants. The Open Science Collaboration's 2015 project repeated 100 psychology studies from three journals' 2008 issues. Of 97 originals with significant results, 35 replications were significant in the same direction (36%).

Colin Camerer and colleagues' 2016 paper repeated 18 laboratory economics experiments from 2011 to 2014. Eleven replicated (61%).

Many Labs counted 11 of its 13 effects as replicated when all samples were pooled. Flag priming and currency priming did not replicate. Imagined contact had weak support.

Kai Ruggeri and colleagues' 2020 study put 17 of the 1979 items to 4,098 participants in 19 countries. Twelve of the 13 original theoretical contrasts replicated when all countries were pooled.

On the stricter test of replicating consistently across samples or countries, both Many Labs and the 2020 study found 10 of 13.

The one 2020 contrast that did not replicate was between items 4 and 8 in the study's own numbering. It tests the reflection effect, but the authors found respondents largely indifferent between the options in both items, so there was "simply no preference to reflect".

Because the other reflection contrasts replicated, they read the result as a peculiarity of the items.

In the studies reported here, the gap between gains and losses sold and the cost of heavy trading appear in trading records. Anchoring appears in laboratory replications. The reasons people trade as measured remain unsettled.

Behavioural finance is the study of how people's actual money decisions differ from what standard economic models assume. Researchers measure those differences in experiments and in trading records. In Terrance Odean's 1998 study of 10,000 accounts, 14.8% of gains were sold against 9.8% of losses.

Standard models such as expected utility theory predict choices from probabilities and outcomes. Behavioural finance tests those predictions against measured choices. In Kahneman and Tversky's 1979 questionnaire, 80% of 95 respondents took a sure 3,000 over an 80% chance of 4,000, although the gamble averages 3,200.

Loss aversion is the finding that a loss weighs more than an equal gain. Tversky and Kahneman's 1992 median estimate was 2.25. In Brown, Imai, Vieider and Camerer's 2024 meta-analysis of 607 estimates, the mean was 1.955, with a 95% interval of 1.820 to 2.102. A loss weighed about twice an equal gain on average.

Parts of its laboratory base have been replicated. Others have not. In Kai Ruggeri and colleagues' 2020 study of 4,098 participants, 12 of 13 prospect theory contrasts replicated. In the Open Science Collaboration's project, 36% of 97 significant psychology results replicated.

This explainer is part of the YX Insights Academy. Each one takes a single idea and checks it against real data.

The same approach runs through everything else we publish:

Good places to start on the website:

DISCLAIMER: This article is strictly educational. Any information or analysis in this note is not an offer to sell or the solicitation of an offer to buy any securities. Nothing in this note is intended to be investment advice and nor should it be relied upon to make investment decisions. Any opinions, analyses, or probabilities expressed in this note are those of the author as of the note's date of publication and are subject to change without notice.