In 1985, Hersh Shefrin and Meir Statman named a pattern that brokers had long observed. Investors are eager to sell assets trading above their purchase price and reluctant to sell those trading below it. They called it the disposition effect. Over a decade later, Terrance Odean's analysis of roughly ten thousand discount brokerage accounts measured it: investors realised gains at about one and a half times the rate at which they realised losses.

The effect is theoretically interesting less for how common it is than for how well it survives the incentives against it. In Odean's data, the winners investors sold went on to outperform the losers they kept. For taxable investors the pattern is doubly costly, because realising gains triggers tax liabilities while realising losses would offset them. An agent with standard preferences over terminal wealth should do roughly the opposite of what most investors actually do.

The disposition effect is therefore best understood as an interaction of several mechanisms rather than a single bias. One is the anchoring of reference points. Another is the hedonic consequence of closing a position. A third is the self-protective accounting that makes certain losses feel intolerable to acknowledge. Each has been isolated experimentally, and some have been traced to measurable neural activity. Separating them matters, because the mechanism determines which interventions should work and explains why so many well-intentioned ones fail.

The canonical explanation starts with prospect theory. Kahneman and Tversky's value function is defined over gains and losses relative to a reference point. It is concave in gains, convex in losses, and roughly twice as steep for losses. Shefrin and Statman's insight was to place the purchase price at that reference point. A stock trading above it sits in the concave region, where risk aversion encourages locking in the gain. A stock trading below it sits in the convex region, where risk seeking encourages holding on in the hope of recovery.

Martin Weber and Colin Camerer tested this directly in a 1998 laboratory study. Participants traded risky assets whose price trends carried information, so holding losers was costly. Participants still sold fewer shares when prices fell below what they had paid. The most telling result came from a treatment that automatically sold all holdings at the end of each period and let participants buy them back freely. The disposition effect fell sharply. Participants faced the same opportunities as before, but the reference point had been reset.

Reference points are not fixed, and they do not update symmetrically. Arkes, Hirshleifer, Jiang and Lim found that reference points adapt more fully after gains than after losses. Investors absorb a price rise into their sense of the baseline more quickly than they accept a decline. Other anchors also compete with the purchase price. Heath, Huddart and Lang showed that employees exercised stock options far more often when the share price passed its previous-year maximum, which suggests that historical peaks can serve as reference points too.

The effect is not limited to securities or small stakes. Genesove and Mayer studied Boston condominium sales during the downturn of the early 1990s. Owners facing nominal losses set higher asking prices, obtained somewhat higher sale prices, and waited considerably longer to sell. The purchase price of a home is economically irrelevant to its current market value, yet it shaped the decisions of sellers dealing with the largest asset most of them would ever own.

The design implication follows directly. Frydman and Rangel showed experimentally that making the purchase price less visible reduces the disposition effect. Most brokerage interfaces do the opposite. They display unrealised profit and loss prominently, often colour-coded, beside every position. These interfaces are not neutral. By making the purchase price the most visible number on the screen, they effectively set investors' reference points for them.

Takeaway

The purchase price tells you nothing about where an asset is going, yet every interface that displays it tells investors where zero is.

Prospect theory alone turns out to be an incomplete explanation. Barberis and Xiong showed in 2009 that when prospect theory is applied to annual trading profits, it often fails to predict the disposition effect, and under some parameters predicts its reverse. Their alternative, developed in 2012, is realization utility. In this model, investors experience a burst of utility at the moment they sell, positive for a gain and negative for a loss, separate from any change in their wealth.

The mechanics are simple. Investors frame each investment as a distinct episode, in line with Thaler's account of mental accounting, and selling closes the episode. If people discount future utility, they want to bring pleasant realisations forward and push painful ones back. Even with linear realisation utility and no loss aversion, the model produces the disposition effect. It also predicts excessive trading and a preference for volatile stocks, two patterns that are well documented in retail portfolios.

Neuroeconomics offers a rare chance to test such a model directly, because realisation utility makes a claim about when reward is experienced. Frydman, Barberis, Camerer, Bossaerts and Rangel ran a trading experiment while scanning participants' brains with fMRI. Activity in the ventral striatum rose at the moment participants sold an asset at a gain. That is the reward signal the theory predicts, and it occurred even though the sale did not change their wealth. Value signals in the ventromedial prefrontal cortex were also consistent with realisation entering the decision.

Alex Imas found converging behavioural evidence. After a paper loss, people tend to take more risk, which fits prospect theory's convex loss region. After a realised loss of the same size, they take less. The money involved is identical, but the psychological state is not. Realising a loss closes the mental account and turns a provisional setback into a final one. This is exactly the moment investors work hardest to avoid.

This creates a real problem for welfare analysis. If the pleasure of selling a winner is genuine utility, it is not obvious that holding losers is a mistake rather than a preference people have chosen to pay for. A defensible position separates experienced utility from decision utility. The burst of satisfaction from selling may be real, yet investors would probably not endorse paying for it once they understood its cost to their long-run returns. Interventions should aim at that gap rather than override preferences wholesale.

Takeaway

If selling is itself a source of utility, the disposition effect is less a miscalculation of wealth than a rational pursuit of the wrong objective.

The financial case for tax-loss harvesting is unambiguous. Realised losses offset realised gains and, in the United States, up to three thousand dollars of ordinary income each year, with any excess carried forward. Constantinides formalised the optimal policy in 1983: realise losses promptly and defer gains. The disposition effect is the exact inverse of that policy. Notably, December was the one month in Odean's data when investors realised losses more readily than gains. The year-end tax deadline achieved what eleven months of identical incentives had not.

Realisation utility explains part of this failure. Harvesting requires the very act that turns a provisional loss into a final one. A second mechanism is responsibility. Chang, Solomon and Westerfield found that the disposition effect holds for individual stocks but reverses for mutual funds, where investors sell losing funds quite readily. Their explanation is cognitive dissonance. Selling a stock you chose yourself means admitting a mistake, while selling a delegated fund lets you blame the manager.

Structural frictions add to these psychological ones. The wash-sale rule disallows a loss if a substantially identical security is bought within thirty days, so harvesting requires finding a suitable substitute or accepting a period out of the market. The benefit is delayed and spread out, arriving at filing time as a lower liability. The cost is immediate and concrete. This pairing of a delayed, diffuse gain with an immediate, concentrated pain is the standard condition under which present bias defeats a clear incentive.

Effective designs work on these mechanisms rather than on information. Automated harvesting, now common through robo-advisors and direct indexing, removes the investor from the moment of realisation. In effect, it applies Chang and colleagues' delegation finding as a design principle. Framing a harvest as a swap matters as well. When a losing position is sold and immediately replaced with a closely correlated alternative, the investor's exposure stays the same and the mental account is transferred rather than closed.

Presentation completes the redesign. Reporting harvested losses as tax savings captured, a positive and realised number, turns a hedonic loss into a hedonic gain without misrepresenting anything. Designers should still be careful. Harvesting lowers the cost basis of the replacement holding, so much of the benefit is a deferral of tax rather than its elimination. Wash sales can also be triggered accidentally across accounts. The goal is better financial outcomes, not simply more trading.

Takeaway

The most effective tax-loss harvesting systems do not persuade investors to accept losses. They make sure no one has to experience the act of accepting them.

The disposition effect persists because several mechanisms reinforce one another. Purchase prices define where gains and losses begin. Realisation creates its own rewards and penalties, separate from wealth. Personal responsibility makes admitting a loss feel like admitting a mistake. Any single mechanism might be manageable. Together, they create a reliable gap between what investors would endorse on reflection and what they actually do.

For system designers, the main lesson is that information alone will not change this behaviour. Investors already know losses are tax-deductible. Interventions work when they alter the reference point, the moment of realisation, or who appears responsible for the decision. Hiding purchase prices, delegating execution, and framing harvests as swaps all succeed because they act on these mechanisms rather than on beliefs.

The broader principle applies well beyond investing. Interfaces, defaults, and reporting choices define what counts as a gain, a loss, and a finished decision. Choice architects are therefore shaping reference points, whether or not they intend to.