Behavioural biases in money decisions

Everyone makes mistakes with money. A bias is a particular kind of mistake: one that leans the same way every time. Random errors average to nothing, which is not the same as costing nothing: each one still moved a decision off the best available, and the damage does not reverse because the next error leans the other way. What random errors do is add up slowly, since one partly offsets the next. Systematic errors add up fast, because the second one points where the first one did, so the gap widens with every decision instead of hovering.

Most biases are shortcuts that usually work. Treating a loss as more serious than an equivalent gain is sensible when a loss can end you. Weighting recent events heavily is sensible when the world changes. Sticking to a commitment is how anything long ever gets finished. Each shortcut misfires in a particular setting, and money is that setting: stakes are large, feedback is slow and noisy, outcomes are partly luck, and the horizon is decades rather than minutes.

Six of them show up repeatedly.

They overlap, and they stack. An investor who will not sell a falling position is running loss aversion, then reads only the bullish case for it, which is confirmation, then justifies staying in by how much has already gone in, which is sunk cost. One holding, three errors, all pointing the same way.

Loss aversion is the finding that losing something registers more strongly than gaining the same thing. In early experiments the ratio sat near two to one: people refused an even chance of winning or losing unless the win was around twice the loss. That central figure has held up: a meta-analysis published in 2024, pooling 607 estimates from 150 papers, put the average at about 1.96 with a narrow interval around it. What has not held up is treating it as a constant. Individual estimates scatter with the person, the size of the stake and the way the choice is put, some settings show no asymmetry at all, and a minority of researchers argue the effect is not general enough to carry the weight put on it. Read the direction as the finding and the multiplier as an average, not as a number any particular decision has to obey.

$$\lambda = \frac{\text{weight put on a loss}}{\text{weight put on an equal gain}}$$

The money versions are easy to spot once you know the shape.

• Holding losers, selling winners. Selling at a loss turns a paper number into a settled one. Investors realise gains at a noticeably higher rate than losses, a pattern named the disposition effect. Wherever realised gains are taxed, a realised loss usually carries a tax value an unrealised one does not, so the reluctance can cost twice over in a taxable account. The rules differ by country, and they matter: in the United States the wash sale rule is the one to read before acting.
• Over-insuring the small, under-insuring the large. A very low excess or deductible buys expensive cover against a loss you could absorb, while the loss that would actually change your life goes uninsured. Loss aversion prices the frequent small loss, not the rare large one.
• A charge against a discount missed. The same money moves people further when it is labelled a fee than when it is labelled a saving forgone.
• Cash held for years after a fall. Two different things look identical from the inside. Discovering during a fall that you cannot sit through one is real information about yourself, and holding less risk afterwards is a considered answer to it. Staying in cash only because you will not repeat a loss already taken is a decision aimed at the past.

The reference point does the work. Loss aversion is defined against a starting line, so whoever sets that line decides what counts as a loss.

A dollar or a pound is fungible: it does not remember where it came from. Mental accounting is the habit of treating it as though it did, by sorting money into separate pots that get different rules.

The label is the whole of it, and experiments isolate it cleanly. Describe an identical windfall as a bonus and people spend more of it; describe the same money as a rebate, a return of what was already theirs, and they save more. Nothing but the wording moved. A tax refund usually wears the rebate label honestly, because where tax is withheld through the year most of a refund is your own income handed back rather than a prize. Two things stop that from being universal: cash paid out through a refundable credit arrives by the same route and is not returned withholding at all, and where tax settles through payroll most employees never file for a refund in the first place. Bonuses, gifts and winnings sit at the other end of the same scale, which is the house money effect: money that feels won is risked in ways money that feels earned is not.

The expensive version is holding two accounts that point in opposite directions. A savings pot earning a low rate sitting beside a card balance charged at a much higher one gives away the difference on the overlapping amount every month. Some of that gap is bought deliberately and is worth the price. Cash you can reach quickly has a value of its own, which is what an emergency fund is for, and a limit paid down is not always a limit you can draw on again. Past reasons like those, the labels are doing the arguing, and a label is not a reason.

Then the honest complication. Mental accounting is also one of the few devices that changes behaviour reliably, and it works precisely because a label sticks. A named sinking fund for the next car, a separate account holding tax owed, and zero-based budgeting all run on the same mechanism the refund does. The difference is who chose the labels. Money sorted by a system you designed while calm is a rule. Money sorted by how it happened to arrive is an accident.

Both of these hand a single number more authority than it has earned. Anchoring gives it to the first number. Recency gives it to the latest.

Anchoring. An opening figure pulls the estimates that follow it, and it does so even when everyone can see the figure is arbitrary. In the original experiment a rigged wheel of fortune was spun in front of people before they estimated a quantity that had nothing to do with it, and where the wheel stopped moved their answers. In money terms that is the crossed-out original price, the asking price on a house, and the first salary figure spoken aloud in a negotiation. One of the sharpest cases is the minimum payment on a card statement. In a published experiment, people who were shown a required minimum and chose to pay only part of the balance paid less than people shown a statement with no minimum on it. Read the finding for what it is: it is about partial payers rather than about everyone, and the minimum is a contractual floor rather than a suggestion, which is exactly why it should not have moved anything.

One neighbouring pull is worth separating out, because it gets filed under anchoring and works differently. On a three-tier menu the expensive top tier is usually there to make the middle one look reasonable, not the other way round: adding a more extreme option raises the share who pick the middle, which is a preference for avoiding extremes rather than the drag of a first number.

Recency. The last stretch of results gets read as the trend. Money tends to arrive in an investment after a strong run and leave after a weak one, so the return an investor actually lives through can trail the return the fund reports. That gap is the distance between a money-weighted return and a time-weighted one, and it repays careful reading: some of it is mistimed buying and selling, and some of it is arithmetic, because a fund holds more money in its later years than its earlier ones whatever anybody decided. Day-to-day volatility reads as news when much of it is noise, and a rare risk feels impossible until one happens, after which it feels likely to happen again soon. Risk and return is the same argument made with the arithmetic.

The counter to both has one shape: bring your own number. A price you settled on before seeing theirs, and a plan written before the last twelve months happened, cannot be pulled by either.

These two guard a position you already hold. Confirmation bias guards the belief. Sunk cost guards the spending.

Confirmation. Once a choice is made, evidence supporting it becomes easier to find, easier to believe and harder to forget. Someone who has bought a share reads the bullish case for it. Someone who has decided to buy a property reads the pieces arguing prices will rise. Someone who has picked a fund checks its best three-year window rather than its worst. Keep the two claims apart: past performance is a weak guide because the scorecards keep finding that a good run rarely repeats, and confirmation is why a weak guide still feels convincing. Both are reasons index funds against active management turns on costs instead. The tell is asymmetric effort: you can say what would confirm you, and not what would change your mind.

Sunk cost. Money already spent and not recoverable is identical under every option in front of you, so it cannot tell them apart. Only what happens next can.

$$\text{Relevant} = \text{future costs and benefits that differ between the options}$$

The renovation that has already swallowed a large sum, the qualification half finished, the car whose next repair exceeds its resale value, the holding that has already fallen a long way: in each case the question is what the remaining spend buys, asked as though you had arrived today.

It grips for two reasons. Walking away makes the earlier loss final, which is loss aversion again, and conceding that the first decision was wrong costs something in front of other people, which is one reason groups and organisations tend to escalate further than individuals do. Sunk cost is not always a fallacy, though, and the last FAQ below sets out the test.

Reading this page will help you name a bias afterwards and will do much less for what you do under pressure, which is not a failure of attention. A bias acts on judgement rather than on your stock of facts. The standard demonstration is an optical illusion: measure the two lines, know they are identical, and they still look different. Knowing does not repair the picture. What recognising a bias after the event does buy is a better rule for next time, which is where the value of reading about them sits.

Money is also poor ground for learning your way out.

• Feedback is slow and noisy. A good decision can lose for years and a bad one can win, so results teach the wrong lesson.
• The moments that matter are rare. Nobody gets many market falls or house purchases to practise on.
• The bias blind spot. People see bias in others far more readily than in themselves, which makes self-inspection the weakest instrument in the box.

So change the situation instead of the person. Decide once, while calm, and make the decision awkward to revisit later.

Automation carries most of the weight because it removes the moment of decision rather than trying to win it: a standing transfer on payday, sized by the savings goal calculator, or contributions on a schedule, which is dollar-cost averaging. Be clear what that buys. It removes the timing question and the temptation to wait, and it does not raise the expected return. Investing from income as it arrives is not a choice between staging and going in at once, because there is no lump sum sitting there; where there is one, spreading it out lowers the expected return rather than raising it, in exchange for a smaller worst case.

A rule you can abandon in ten seconds is an intention. A rule needing a form, a call or a week is a rule. This is educational material rather than personalised advice, and the opportunity cost of money is what most of these rules exist to keep in view.