The job defines what ought to be done, but people make the decisions. Human attention is limited, and most decisions rely on quick intuition. People measure gains and losses against a reference point, and the pain of a loss is about twice the pleasure of an equal gain. The resulting biases are not random errors; they have a direction and can be predicted: people sell the stocks that made money and keep the ones that lost; the more often they check their accounts, the less risk they are willing to bear; and they are overconfident in their own judgment. Because the biases have a direction, design can change them: the same people make quite different decisions under different defaults and forms of presentation.
Consider this question: a bat and a ball cost $1.10 in total, and the bat costs $1 more than the ball. How much does the ball cost? The answer that springs to most people's minds is $0.10; the correct answer is $0.05. Frederick (2005) used questions of this kind to measure cognitive reflection and found that even among students at leading universities, a large share answered wrongly. Kahneman (2011) divides human thinking into two modes: one fast, automatic and effortless, the other slow, deliberate and demanding attention. Most everyday decisions are made by the first, which in most settings is efficient; but on problems that require calculation or resisting a first impression, fast intuition errs systematically. Simon (1955) had pointed out earlier that human rationality is bounded: people do not search for the optimal solution but for one that is good enough, given limited information and computing power.
Fast intuition and careful checking
A bat and a ball cost $1.10 in total, and the bat costs $1 more than the ball. How much does the ball cost? Pick an answer first.
Simon (1971) made an often-quoted observation: information consumes the attention of its recipients, so a wealth of information creates a poverty of attention. Investors face far more information than they can process, so decisions tend to depend on what is seen first rather than on what matters most. DellaVigna and Pollet (2009) found that the market's same-day reaction to earnings announcements made on Fridays is markedly weaker than on other weekdays, and the subsequent price drift is larger: when attention is scattered, information enters decisions more slowly. For individuals, the most prominent information is usually price changes, rankings and news, while the information most relevant to long-term results, such as fees, how concentrated the portfolio is and how far it remains from the goal, is often not prominent at all.
Kahneman and Tversky's (1979) prospect theory holds that people judge outcomes not by total wealth but by gains and losses relative to some reference point, and that the pain of a loss is clearly greater than the pleasure of an equal gain. Tversky and Kahneman (1992) estimated that losses weigh about 2.25 times as much as equal gains. In addition, the feeling of both gains and losses diminishes as the amounts grow: going from a $100 gain to a $200 gain adds less pleasure than going from $0 to $100. The reference point is usually the purchase price, but it can also be the account's peak or other people's returns, and it largely determines whether an outcome is felt as a gain or a loss.
The value function of prospect theory
The horizontal axis is the gain or loss relative to the reference point; the vertical axis is the value felt. Drag the outcome to compare how a gain and a loss of the same amount feel.
Reference points combined with loss aversion produce one of the most robust regularities in individual investor behaviour. Shefrin and Statman (1985) called it the disposition effect: investors tend to sell stocks that have gained too early and to hold stocks that have lost for too long, because selling a losing stock means admitting the loss. Odean (1998) analysed the trading records of about 10,000 accounts at a brokerage: counting only days on which a sale took place, about 14.8% of winning positions were sold, against only about 9.8% of losing positions; and over the following year, the winners that were sold outperformed the losers that were kept by about 3.4 percentage points on average. Not only does the disposition effect bring no gain; under most tax systems it also means paying more tax.
A portfolio with positive long-term returns has a fairly high probability of showing a loss over short periods. Benartzi and Thaler (1995) proposed myopic loss aversion: if a person evaluates a portfolio frequently and each loss seen counts about twice as heavily in how they feel, they will find stocks too risky and allocate too conservatively. They estimated that an evaluation period of about one year is enough to explain why stocks command such a high risk premium. Experiments by Gneezy and Potters (1997) likewise showed that, faced with the same gamble, participants required to look at results more frequently invested less.
The more often you look, the less worth holding stocks seem
The same stock portfolio, with positive long-term returns. Drag how often the account is checked to see the probability of a loss at each check and the felt value under loss aversion. Above zero, holding stocks feels worthwhile; below zero, it does not.
People generally overestimate how accurate their judgments are. In calibration studies, participants are asked to give ranges they are 98% sure contain the true answer, and the share of true answers that fall inside the ranges is far below the confidence claimed (Lichtenstein, Fischhoff and Phillips, 1982). Overconfidence leads people to underestimate risk, concentrate too heavily and trade too much. Barber and Odean (2001), using gender as a proxy for overconfidence, found that men traded about 45% more than women, and that trading reduced men's annual net returns by about 2.65 percentage points and women's by about 1.72 percentage points.
Because biases are systematic, decisions change with presentation and defaults. Madrian and Shea (2001) studied what happened when a company changed its retirement savings plan from "opt in voluntarily" to "enrolled automatically, free to opt out": participation among new employees rose from about 37% to about 86%, with no reduction at all in employees' freedom of choice. The "Save More Tomorrow" plan designed by Thaler and Benartzi (2004) had employees commit in advance to raising their savings rate at each future pay rise; participants' savings rates rose from 3.5% to 13.6% over about 40 months. Thaler and Sunstein (2008) called this approach choice architecture. The same tools can also be used to lead users into doing things against their own interest: designing trading to feel like a game, hiding losses, removing the confirmation step before an order. Choice architecture is neither good nor bad in itself; what matters is whether it serves the user's job or the platform's revenue.
That design can change biases does not mean a product should decide for the user. A user's practice may not amount to a complete method: finding opportunities, buying and reviewing may each rest on a different set of reasons. This piecemeal practice is itself a stable state, and the tension between it and the practice of mature investors comes from the user's own wish for better results. Simply going along with the current practice lets the cost of frequent trading pile up year after year; correcting the user at every turn is likely to drive them away. A workable approach is to respect the user's habits first, use the user's own trading records to show the results, then invite rather than require them to try another method, and promise only to improve the decision process, not the outcome. Thaler and Sunstein (2003) called the approach of preserving freedom of choice while steering people toward choices they themselves consider better libertarian paternalism.
Because biases have a direction, a product's presentation and defaults can never be neutral. They should be placed on the side of the user's job, while respecting the habits the user already has and replacing lecturing with invitation.
Fees, concentration and the distance from the goal should be as prominent as price changes; do not compete for attention with rankings and push notifications.
Besides the purchase price, show where the portfolio stands relative to the plan and the goal; when selling, explain how different ways of selling affect the portfolio and the tax bill.
For long-term jobs, show by default a period consistent with the horizon, and do not put daily price changes in the most prominent place.
Defaults, confirmation steps and presentation should all serve the user's job; do not use game-like design to induce trading.
Carry out the user's habits as usual, use their own trading records to show the results, then ask whether they would like to try another method; promise only to improve the process, not the outcome.