Preference is a relation on options
A ≻ B says the chooser ranks A strictly above B. The options may be cakes, careers or lotteries; the relation does not care, and that is why one condition on it can apply to all of them.
Preference is a relation on options, and rational choice needs it to be complete and transitive. Build a cyclic preference on the relation grid, read the pair that breaks transitivity, and compute what a trader earns by walking an agent round the cycle.
Core ideas
A theory of choice has to start from what the chooser prefers. Before any number appears, there is a structure the preferences must have.
A ≻ B says the chooser ranks A strictly above B. The options may be cakes, careers or lotteries; the relation does not care, and that is why one condition on it can apply to all of them.
Complete: for any two options one is preferred or they are tied. Transitive: if A ≻ B and B ≻ C, then A ≻ C. Together they let the options be lined up, with ties sharing a rung.
If A ≻ B, B ≻ C and C ≻ A, every option is beaten by another, so choosing “the best” has no answer. The money pump shows the same defect from the outside, as a loss.
The material
Two conditions, then the argument that the second one is not optional.
Choice theory begins by asking what the chooser would pick from a pair. Write A ≻ B when A is strictly preferred to B, and A ≽ B when A is at least as good: preferred or tied. Everything the course computes later, expected values and rules for ignorance included, assumes that this relation can be turned into a ranking.
A preference is complete when, for any two options A and B, at least one of A ≽ B and B ≽ A holds. The chooser is never left without a view on a pair.
A preference is transitive when A ≽ B and B ≽ C together give A ≽ C. A comparison that passes through a third option does not change the answer.
The relation grid below makes both conditions things you can see. Read a 1 in row a, column b as “option a is preferred to option b”. A complete strict preference has a 1 on one side of the diagonal or the other for every pair, with no gap. A transitive one has no two-step path a to b to c without the direct step a to c; the property table names the first offending pair when there is one.
The preset is the relation a < b on five options, and as a strict preference it is both complete and transitive: option 1 beats all, option 5 beats none. Now click the cell in row 5, column 1. That adds the pair 5 ≻ 1 and closes a loop, because the preset already ranks 1 ≻ 5 through the cell in row 1, column 5. The table reports transitivity failing at the pair (1, 5) and (5, 1) without (1, 1): a cycle forces each option to be preferred to itself, which no one means by preference.
Nothing in those three lines is false as a report of taste. The trouble is what a cyclic chooser will do. Suppose you hold A, and you prefer C to A, B to C and A to B. A trader offers C for A plus a small fee ε, and you accept because you prefer C. The trader then offers B for C plus ε, and you accept. Then A for B plus ε, and you accept again.
You began with A and you end with A. You are 3ε poorer, and each trade was one you wanted at the moment you made it. Another lap costs another 3ε, and the trader can run as many laps as you will go round.
That is the money pump. It is an argument that transitivity is a condition on what you prefer, not a feature of what you like. The options can be anything and the fee as small as you please; the loss is 3ε a lap in every case, so a cycle among three kinds of tea is exposed in the same way as a cycle among three jobs.
The argument has a strongest objection, and it deserves a hearing. A person may rank three flats by different standards, price against size, size against light, light against price, and each pairwise verdict can be reasonable. The reply is that this is a cycle in the pairwise verdicts and not in the ranking of the flats as wholes, so one of the standards has to give way or a weighting has to be chosen before the person can act without being walked in circles. Accepting the cycle costs the chooser the loss above; rejecting a pairwise verdict costs a reason that looked sound.
The limit of the argument is worth stating. It assumes the chooser really will pay for each swap and has no memory of the lap. An agent who sees the trader coming and refuses the third offer has changed her preferences, which is exactly the repair that transitivity asks for. The pump does not prove cycles unthinkable; it prices them.
Interactive
Decided by checking, not by name
Method
Four moves, in this order. The first three decide whether the ranking exists and the fourth puts a price on its failing.
Write each X ≻ Y the chooser holds, one per pair of options. If some pair has no entry and no tie, the preference is not complete, and a different question is being asked.
For each A ≻ B and B ≻ C, look for A ≻ C. A missing pair is the witness that transitivity fails, and it is the only thing the check has to find.
If a chain of strict preferences returns to where it began, the preference has a cycle, and no option in the cycle is best.
Start at any option in the cycle and take each preferred swap in turn, paying the fee each time. A lap of k swaps costs k·ε, and the agent holds exactly what it started with.
The chooser prefers C to A, B to C and A to B, and pays 1 cent for each swap to something preferred. She starts holding A.
Each line, taken alone, is a choice the chooser endorses. It is the sequence that has no defence: she ends with what she began with and three cents fewer, and the pump may run for as long as she keeps saying yes.
Given a set of options and the pairs a chooser prefers, you should be able to say whether the preference is complete, name the pair that breaks transitivity if there is one, and compute what a trader collects over a given number of laps at a given fee.
Something to change about this lesson? Write it down, tick it off when it is handled, and export the list when you want to act on it. Kept in this browser only: nothing is sent anywhere, and nobody is notified.
No recommendations yet for this lesson.