Preferences: What People Actually Want, Not What They Should

A preference is an ordering over outcomes, not a feeling about them. Economists read the ordering from what a person chose rather than from what they said. The measurement is called revealed for that reason. The ordering is assumed to be complete, transitive and stable, and the interesting part of behavioural finance is what happens when it is not.

Almost everything in behavioural finance is a claim about one of two things: what somebody believed, or what somebody wanted. The first is judgment, set out under judgment under uncertainty. Wanting is the subject here, and the first move is to strip the everyday warmth out of the word. In ordinary speech a preference is a liking. In this subject a preference is a structure, and the whole point of turning wanting into structure is that a structure can be checked for contradictions while a liking cannot. Once wanting is written as an ordering, an inconsistency in it stops being a personality note and becomes an amount of money.

What is a preference, precisely, in the sense economists use?

A preference is an ordering over a set of outcomes. Given any two of them, it says which one is ranked above the other, or that the two are ranked equally. Ranking above, ranking below, ranking level: those three answers are the whole of it. The ordering does not say by how much, it does not attach a number to any outcome, and it does not report how anybody feels while holding one. If a person ranks tea above coffee, the ordering has recorded everything it is entitled to record. Whether that person loves tea or merely tolerates it is outside the object entirely.

Dropping the strength of feeling sounds like a loss of information, and it is, deliberately. An ordering is weaker than a feeling and that weakness is what makes it usable. Two people cannot compare how much they each enjoy something, and neither can an adviser, and neither can a regulator. But anybody can look at a ranking and ask whether it holds together. A ranking that puts the first above the second, the second above the third, and the third above the first is broken in a way that needs no access to anybody's inner life to detect.

Take it out of money first. A household is choosing between three flats to rent. One is nearer the school, one is cheaper, one is larger. The household does not need to say how much it values a shorter school run, and it could not say it in any unit anybody else would recognise. Of any two flats put in front of it, the household only needs to be able to say which one it would take. The answer, repeated over every pair, is the preference. The ordering is the object; the strength of feeling behind it is not part of the object and never enters the arithmetic.

How is a preference measured, if asking will not do?

The awkward problem is this. The ordering sits inside a person and cannot be observed. Asking is possible, and people answer politely, and the answers are worth very little. So economists took a different route: assume the ordering exists, then read it off the choices the person actually made. If somebody could have had the deposit and took the mid-cap scheme instead, the mid-cap scheme is ranked above the deposit in their ordering. Repeat that over enough pairs and the ordering assembles itself out of behaviour. An ordering assembled that way is revealed preferenceAn ordering read from what somebody actually chose rather than from what they said they wanted., and Paul Samuelson set it out in Economica in 1938.

The move is more radical than it first looks. Samuelson was not saying that choices are a good proxy for wants. He was saying that for the purposes of the theory, the choices are what the want means. The behaviour is the definition, so no separate inner ranking waits to be checked against it. A disagreement between a questionnaire and a decision log is therefore not a puzzle about honesty. Only one of the two documents is a measurement at all.

The Palash decision log, an invented record, shows the difference plainly. The log holds 240 decisions taken by 60 investors over eight quarters. Of those 240 decisions, 96 were buys, 84 were sells, 36 were switches and 24 were pauses of a standing instruction. The four counts sum back to 240. Set that record beside what the same 60 people said about themselves on a form, and the two tell different stories. The form records what somebody would like to be true about their own ordering; the log records the ordering that was actually exercised.

A record that is mostly one kind of decision reads one corner of an ordering rather than the whole of it. So the shape of the log is worth seeing before the log is trusted as a measurement.

Why is what somebody says about what they want weak evidence?

A stated preferenceWhat a person says they would choose, which is regularly not what they do choose when the choice is real. is weak evidence for four separate reasons. The four reasons need different fixes, so they are worth keeping apart. The first is that saying costs nothing. Describing oneself as patient carries no penalty, so the answer drifts towards the version of the person the respondent would like to be. The second is that the question is usually hypothetical, and a hypothetical loss is a sentence while a real one is money leaving. The third is that the form itself shapes the answer. In the log, 11 of 30 readers shown 214 options chose anything at all, against 21 of 30 shown 7 options. The fourth is that people often genuinely do not know, and a form gives them nowhere to say so.

The cost point deserves its own sentence because it is the one doing the theoretical work. A choice is trusted precisely because it was expensive. In the log, the top turnover group ran 210 per cent a year and gave up 4.1 points to dealing charges, spread and tax together, against 0.3 points in the lowest group. Gross returns across the five groups sat within 0.3 points of each other, between 10.9 and 11.2 per cent. Net returns ran from 10.9 down to 6.9 per cent, a spread of 4.0 points. Whatever those investors said about wanting to keep costs down, the ordering they exercised put activity above 4.0 points a year, and that ordering was paid for in full.

The rows on their own hide how far apart the second spread is. Put the two spreads on one scale before reading them.

What are the three properties an ordering is assumed to have?

John von Neumann and Oskar Morgenstern wanted to know what an ordering has to obey before a single number can be attached to each outcome and the whole thing handled with arithmetic. They wrote the properties down carefully in Theory of Games and Economic Behavior in 1944. Three of those properties carry the weight: completeness, transitivity and independence.

CompletenessAny two options can be ranked against each other, with ties allowed. Completeness rules out the shrug, not the tie. says that for any two outcomes a person can say which is ranked higher, or say the two are level. There is no third answer and no permitted shrug. Completeness also does not demand a strict winner, and the omission matters just as much. Ranking two things level is a real position called indifferenceRanking two options equally. Indifference is a definite position in the ordering, not a failure to decide., and it satisfies completeness perfectly. Being unable to compare at all is what breaks completeness, and that inability is more common than the theory would like. Shown 214 options, 19 of 30 readers in the log chose nothing.

TransitivityIf the first is ranked above the second and the second above the third, then the first is ranked above the third. says the rankings fit together. A person who ranks the index scheme above the mid-cap scheme, and the mid-cap scheme above the deposit, is committed to ranking the index scheme above the deposit. The third ranking is not decided separately; the first two answers already settled it. IndependenceAdding the same third possibility to both options should not flip which of the two is preferred. says that if the same extra possibility is bolted on to both of two options, the ranking between them should not flip. Completeness makes an ordering usable, transitivity makes it consistent, and independence lets probabilities be handled by multiplication. Independence is also the one that breaks first.

What happens when the ranking goes round in a circle?

Transitivity is the one people find obvious until they watch it fail. Suppose somebody ranks the index scheme above the mid-cap scheme because it costs less to hold. The same person ranks the mid-cap scheme above a deposit because it has more room to grow. And they rank the deposit above the index scheme because the deposit cannot fall. Each of the three comparisons was made on a sensible ground. The three grounds were different grounds, and that is exactly how a circle gets built: by comparing each pair on whichever feature is most obvious while that pair is the one in front.

A circle is not an exotic failure. Circles form in any committee that votes on pairs, and they form for a person deciding alone whenever the pairs arrive one at a time with a gap in between. A household comparing three schools on distance, then two of them on fees, then two of them on results, can produce a circle without a single careless answer. The circle is not visible from inside any one comparison. The circle appears only when all three comparisons are written down together, and writing them down is the entire practical value of drawing an ordering out.

Why is independence the fragile one?

Independence is the axiom that lets probability be handled cleanly, and it is also the one that people break most readily once uncertainty enters. The idea is simple. If two options both come with the same one in ten chance of the same irrelevant side outcome, that shared component should cancel and the ranking between the two should stay where it was. Nothing about the shared part distinguishes them, so nothing about it should move the answer.

Maurice Allais, writing in Econometrica in 1953, built a pair of choices designed to test exactly that and watched careful people flip. The construction worked on a certainty in the first pair that vanished in the second: shave a sure thing down to a very high probability and the ranking reverses, even though the arithmetic of the shared component is identical in both pairs. The point was not that people are careless. The people who flipped were mostly economists who could see the structure and flipped anyway. Independence fails because certainty is treated as a different kind of thing rather than as a probability of one, and no ordering built on multiplication has anywhere to record that.

The failure deserves proportion. Independence breaking is what eventually forced a rebuild of the machinery underneath, and that rebuild is set out under prospect theory. The three properties are separately testable, so an ordering can satisfy two of them and break the third, and naming which one broke is the difference between a diagnosis and a complaint.

What does an ordering that circles actually cost, in rupees?

Now the money. Suppose a person holds a deposit of Rs 3,00,000/-, and suppose their ordering circles in the way described: the mid-cap scheme is ranked above the deposit, the index scheme above the mid-cap scheme, and the deposit above the index scheme. Somebody who knows the ordering can now make an offer. The holder ranks the mid-cap scheme above the deposit, so the offer is a swap of the deposit for the mid-cap scheme at a fee of Rs 500/-. By the person's own ordering that is an improvement, so they accept. Then the same offer takes them from the mid-cap scheme to the index scheme, another Rs 500/-. Then from the index scheme back to the deposit, another Rs 500/-.

Count what happened. Three swaps, Rs 1,500/- paid, and the person is holding exactly what they held at the start. Nothing about the world changed in between. No price moved, no news arrived, nobody was misled about any fact, and every single step was one the person wanted by their own stated ranking. The sequence is the money pumpA sequence of swaps somebody accepts one at a time, each an improvement by their own ranking, that returns them to where they began but poorer., and the money pump is the reason transitivity is treated as a requirement rather than a preference about preferences.

Withdrawing a single one of the three answers is enough, so the repair is smaller than it sounds.

What does one number say about a whole ordering?

The log measured the ordering rather than asking for it, and one measurement compresses a great deal. Each of the 60 investors was asked what gain would make a fifty-fifty gamble against a Rs 10,000/- loss worth taking. The median answer was Rs 22,000/-. Divide Rs 22,000/- by Rs 10,000/- and the measured coefficient is 2.2. Read as an ordering, that says a loss of Rs 10,000/- and a gain of Rs 22,000/- sit at the same place: the point at which the person is indifferent between taking the gamble and walking away.

Notice how much structure one number carries. If the ordering treated the two directions alike, the balancing gain would be Rs 10,000/- and the coefficient would be 1.0. It is not. The gain has to be more than twice the loss before the two sit level. The asymmetry is a fact about the measured ordering rather than an explanation of it, and the machinery that produces the asymmetry is set out under prospect theory. A measured coefficient states what an ordering does; explaining why the ordering does it takes a whole model.

Are preferences stable, or does the starting point get into them?

Stability is the assumption that the ordering holds still while nothing relevant has changed. Stability is the quiet assumption, and behavioural finance exists largely to interrogate it. Two more measurements from the same afternoon, put to the same 60 people, are enough to make the question sharp.

First, a certain Rs 5,000/- was offered against a half chance of Rs 11,000/-. Half of Rs 11,000/- is Rs 5,500/-, so the expected valueThe average of the outcomes weighted by how likely each one is. of the gamble is Rs 5,500/-. The gamble is worth Rs 500/- more than the certain Rs 5,000/- in average money. Forty two of the sixty took the certain Rs 5,000/- anyway, being 70.0 per cent, handing over Rs 500/- of average value to be rid of the spread. Second, the same shape was turned upside down: a certain loss of Rs 5,000/- against a half chance of losing Rs 11,000/-. Thirty nine of the sixty took the half chance, being 65.0 per cent, taking on Rs 500/- of average cost in order to keep the spread.

Read those two together. Above the starting point the spread was a thing to be paid to avoid. Below it the spread was a thing worth paying to keep. Same people, same room, same Rs 500/- at stake either way. An ordering written over final amounts has no way to know which side of the starting point an outcome is on, so no single such ordering can produce both of those answers. Something other than the final amount is getting into the ranking, and that something is where behavioural finance finds its subject.

Where does a preference stop and a judgment begin?

Two people look at the same holding and rank it differently. Before anybody starts arguing, one question sorts out what kind of disagreement this is. Do they disagree about what is likely to happen, or about how much it would matter if it did? The first is judgment and it is a claim about the world. The second is preference and it is not a claim about the world at all.

The distinction earns its keep because the two have completely different repairs. A disagreement about likelihood can be narrowed with evidence: go and look, count something, read the statement, wait for the next set of results. A disagreement about how much an outcome matters cannot be narrowed that way at all. Send both people away with better information. Nothing they learn touches the ordering, so they come back agreeing on the odds and still ranking the two holdings differently. Trying to fix a preference disagreement with more information is the most common wasted argument in this whole subject, and one question in advance prevents it.

The sorting question matters just as much for one person deciding alone. The decision is stuck on one of the two, and naming which one is the whole job. If the likelihood of the bad outcome cannot be stated, that is judgment, and there is work to do. If the likelihood is known exactly and the decision still cannot be made, that is preference, and no further reading will help. The person still has to decide what they are willing to live with.

How does an adviser or a person deciding alone use any of this?

Devika Rao, the adviser at the invented Palash Advisory Services Private Limited, does not go looking for a client's ordering in a conversation. She reads it off the record. A practice has duties around capturing what a client says, so the form is still collected. The form is treated as testimony and the log is treated as evidence. Where the two disagree she does not decide which one is honest. She writes down both and asks the client about the gap. Asking about the gap is far more useful than asking somebody to rate their own caution out of five.

Three specific uses follow from reading an ordering off a record. The first is the pair test: before recommending anything, the adviser checks whether the client's answers across the last several decisions can be arranged in a chain, or whether they circle. A circle is not a character flaw and it is not permanent; it is three comparisons made on three different grounds, and it closes the moment they are written side by side. The second is the cost question: whenever a client's stated ordering and their exercised ordering diverge, the divergence is priced. In the log the top turnover group gave up 4.1 points a year, and a person who says they want to keep costs down has an ordering that says otherwise. The third is the sorting question from the section above, asked before any argument begins.

For a person deciding alone, with no adviser and no committee, the same three uses work unchanged and cost nothing. The last five decisions go on one sheet of paper. A circle is what to look for. Where one thing was said and another done, the cost of the doing is worked out. And where the decision is stuck, the question is whether it is stuck on what is likely or on what could be lived with. All three of these are checks on the ordering itself, and not one of them requires knowing anything about any market.

Set the three side by side and what they have in common becomes visible.

Meera Sundaram, Devika Rao, Palash Advisory Services Private Limited, the Palash decision log, the Palash 100 index, the Vindhya index scheme, the Nilgiri mid-cap scheme, Suvarna Chemicals Limited and Kesari Logistics Limited are invented.
Educational material. Not advice on any investment, tax, budget or market position.