Ambiguity and Complexity Aversion: Avoiding the Unpriceable

Risk is not knowing which outcome will arrive. Ambiguity is not knowing the odds at all. People treat the second as a separate bad thing and pay a real amount to avoid it. An expected utility calculation has nowhere to put that amount. Complexity aversion produces the same backing away from something knowable in principle and too effortful in practice, and both end in nobody choosing.

Risk and uncertainty are usually separated by a definition and then left there. The separation carries a number, and the number is large. The preference for known odds is not a philosopher's quibble that disappears once real decisions are examined. The preference is large, it is stable, and it survives being shown to the person that it is costing them something. Surviving the demonstration is what makes the preference worth treating on its own. A preference that vanishes the moment it is pointed out is a slip, and a preference that does not vanish is a mechanism.

What is ambiguity, and how is it different from risk?

Everybody has a coin, so a coin is the natural starting point. A person is asked to bet on heads. The coin has been examined, it is even, and the chance of heads is one half. The bet may be disliked or it may be taken, but there is nothing about the situation that is unknown except which side comes up. Risk is exactly that situation: the outcome is open and the odds are settled.

Now a second coin arrives in an envelope. Nobody will say anything about it. The second coin might be ordinary, it might be weighted, and nobody has said which. The bet on heads is to be placed on that one instead. Most people feel something change. The outcome was open on the first coin too, so the outcome is not what changed. The change is that the oddsThe chance of each outcome, stated as a number a person could check or argue with. are no longer available. Unstated odds are ambiguityNot knowing the odds, as against not knowing which outcome will occur., and ambiguity is a different situation, not a worse version of the same one.

The obvious reply is that a number can simply be invented. If nobody has stated the mix, heads can be treated as one half by symmetry, written down, and the decision carried on as before. The reply is clean and it does not work. People who have written down that number still refuse the second bet at the same terms they accepted on the first. An estimate does not repair what they are responding to. The missing item is anything standing behind the estimate at all. On the first coin the one half is a fact about the coin; on the second, the one half is a fact about the decider's ignorance wearing the same clothes.

Take it out of coins. A household is choosing between two rented flats at the same rent. For the first, the landlord hands over the last two years of maintenance bills, the society's accounts and the electricity readings. For the second, the landlord says the building is fine. The second flat may well be the better one. Nobody in the household believes it is worse for certain. The household takes the first anyway. Pressed for a reason, they say they did not know enough about the second. The coin gives exactly that answer.

What did Ellsberg actually show, and why is it so hard to argue with?

Daniel Ellsberg, writing in the Quarterly Journal of Economics in 1961, built the demonstration that turned a feeling into a contradiction. The setup is two containers. The first holds 100 tokens, stated as 50 pale and 50 dark. The second holds 100 tokens in a mix nobody has stated. A prize is offered for drawing a pale token, and the chooser may pick which container to draw from. Most people take the first.

Now the same two containers, and the prize is for drawing a dark token instead. Most people take the first again. The second choice is the whole demonstration, and it is worth slowing down on. Choosing the stated container for pale says the chooser thinks it holds more pale tokens than the unstated one. The unstated one therefore holds fewer than 50 pale. Choosing the stated container for dark says the unstated one holds fewer than 50 dark. The unstated container holds 100 tokens and they are all either pale or dark, so both beliefs cannot be true. Fewer than 50 of each would leave a container holding fewer than 100 tokens, which nobody claimed.

The contradiction is why the result is hard to argue with. The demonstration does not depend on measuring anybody's attitude to risk, on comparing two people, or on agreeing what a fair bet is. One person, four choices, and a set of beliefs that cannot be written down consistently. And the behaviour is not repaired by explaining it. Show somebody the contradiction and a good number of them will look at it, agree with the arithmetic, and pick the stated container again. Surviving the explanation is the property that separates a taste from an error, and it is why ambiguity aversionPaying to avoid unknown odds even when they might turn out to be favourable. is treated as a preference in its own right rather than as a mistake in mental arithmetic.

What are people actually paying for when they avoid unknown odds?

A preference nobody will pay for is not much of a preference, so put money on it. Set two options in front of Meera Sundaram, an invented investor. The two are stipulated to have the same expected outcome. The first states its composition in full. The second states nothing about what is inside it. She is asked how much she would commit to each.

To the first she commits Rs 1,00,000/-. To the second she commits Rs 82,000/-, or 18.0 per cent less. She has been told, and she believes, that the expected outcome is identical, so the Rs 18,000/- difference is not a judgement that the second option is worse. The Rs 18,000/- buys nothing except knowing, and knowing has been stipulated not to change the outcome. That is the cleanest statement of what an ambiguity premium is: a real amount of exposure surrendered in exchange for information that leaves the expected result exactly where it was.

The boundary around the 18.0 per cent matters as much as the figure itself. No market puts a price of 18.0 per cent on disclosureHow much about a composition or a cost is actually stated to the person deciding., and no such price exists anywhere to be quoted. The 18.0 per cent is what one person demands before she will take one option over another. A demand of that kind is a fact about a decision, not a level.

The shape is the lesson. The discount does not sit at its full height until some threshold is crossed and then switch off; it comes down in a straight line, so the first quarter of disclosure is worth exactly as much as the last quarter. The straight line is why partial disclosure is a real thing to reason about rather than a half measure that fools nobody. Somebody who states four of the eight things that matter has bought back half of the discount, and somebody who states seven has bought back most of it.

What is Complexity Aversion, and how is it a different mechanism?

Now one thing changes, and only one. Everything stays knowable. Every fact about every option is stated, nothing is held back, and the whole is simply made a lot of work to get through. Nothing is ambiguous in the Ellsberg sense at all. Every odd is stated and every composition is written down where it can be read. Effort has risen, and uncertainty has not. And the behaviour that comes back is the same behaviour: people back away.

The Palash decision log carries a measurement of exactly this. Shown a set of 214 options, 11 of 30 people made a choice at all, being 36.7 per cent. Shown a shortlist of 7 drawn from that same set, 21 of 30 chose, being 70.0 per cent. Nothing in the long set was less knowable than in the short one; every option in the shortlist came out of the long set unchanged. The proportion who chose nearly doubled when the work fell, with knowability held perfectly still. The rate rose from 36.7 per cent to 70.0 per cent, a factor of 1.9. Responding to effort rather than to unknown odds is complexity aversionBacking away from something that is understandable in principle and effortful in practice., and its output is the same one: nobody chose.

Everyday version. A government office publishes every rule on its noticeboard, complete and correct, across nine sheets in small type. A person who needs one of those rules walks up, reads two sheets, and goes home to ask a neighbour instead. Nothing was hidden. The whole thing was there. The cost of extracting it was higher than the person had, so the person withdrew, and withdrawal looks identical whether the sheets were incomplete or merely long.

Why do two different mechanisms produce one visible behaviour?

The shared output is the reason both effects are treated together rather than separately. Unknown odds and high effort are genuinely different causes. One is about what is knowable, the other about what is affordable to work out. The two causes arrive by different routes and they respond to different repairs. And they arrive at the same door: the person declines to choose at all.

WithdrawalDeclining to choose at all, rather than choosing something different. is the behaviour, and behaviour is all that can be seen from outside. Devika Rao, the adviser at the invented Palash Advisory Services Private Limited, cannot read the cause off the outcome. A person who did not choose looks exactly the same whether nothing was stated or everything was stated at length, and the two need opposite repairs. If the cause is unknown odds, the repair is stating more. If the cause is effort, stating more makes it worse. More printed sides mean more work. Guessing wrong therefore does not merely fail; it pushes in the wrong direction.

Is avoiding what cannot be assessed actually a mistake?

Here is where most treatments of this subject go wrong, and the honest answer is uncomfortable for anybody who enjoys a tidy list of errors. Backing away from what cannot be assessed is a rule that is right far more often than it is wrong. The reason is not psychological, it is structural. Whoever did not state the composition had the option of stating it. Non-disclosure is nearly always a choice made by somebody, and somebody choosing not to state something generally had a reason to prefer it unstated.

Complexity works the same way. Length is also a choice. A document can be long because the subject genuinely has forty six printed sides of substance in it, and it can be long because length is where cost goes to be unnoticed. The scheme document in the Palash record runs 46 printed sides with its risk statement on the 31st in eight point type; 7 of 30 readers could state the main risk afterwards, being 23.3 per cent, against 24 of 30, being 80.0 per cent, for a 90 word version placed at the top. Nothing was withheld in either version. The information was in both. Only the effort differed, and the effort decided what actually got understood.

So the person who walks away from the unstated and the enormous is not being stupid. The person is applying a rule with a good hit rate to a world that mostly deserves it. Any correction that tells them to stop applying it is a worse rule than the one they have, and that is the trap this subject falls into most often.

What turns this sound rule into a bias, and what is the correction?

Two questions do most of the work, and both of them can be asked by a person deciding alone with nobody to consult. The first is whether anybody chose this level of disclosure. If a party with something to gain decided what to state and what to leave out, the rule fires correctly and no further thought is needed. If nobody chose it, the ambiguity is innocent, and the rule is firing on a situation that has no author.

The second question is whether the effort is genuinely high or only looks high. Forty six printed sides is a real quantity of work. Forty six printed sides with the four things that matter listed on the first one is not, and the difference can be settled by reading that first side. Both questions convert an automatic response into a conditional one, and that conversion is exactly what adding exception handling means. Neither question requires any financial method, any valuation and any market view, which is deliberate: a correction that needs expertise is no correction at all for the person who most needs it.

How does this get exploited, and how is it served honestly?

Take the exploitation first, described as a pattern and attached to nobody. The pattern has two halves and they fit together. The first half uses complexity as a place to keep cost: an arrangement is described in general terms, the parts that carry the charge are stated somewhere they will not be read, and the length of the whole is what makes them unreadable rather than any decision to hide them. Nothing was withheld and nothing needs to be. Effort did the work concealment would otherwise have had to do.

The second half is aimed at the other mechanism. Where an ambiguity premium exists, somebody can be paid for removing it, and the removal can be counterfeited. A reassuring summary is not the same thing as a stated composition, and a confident tone is not the same thing as odds anybody can check. The counterfeit works because the feeling of ambiguity, rather than the ambiguity itself, is what people were paying to be rid of. A person who no longer feels uncertain has bought the whole of the relief and none of the information.

Serving it honestly is the mirror image, and it turns on one fact that the log makes visible: effort and disclosure are separable. One can be lowered without lowering the other. A shortlist of 7 drawn from the same 214, with the basis of the shortlisting stated on the same sheet, lowers the work without hiding a single thing. The full 214 stay available to anybody who wants them, the shortlisting rule is written down where it can be argued with, and the person who chooses from the 7 knows precisely what has and has not been done for them.

What does the avoidance cost the person who does it?

Two costs, and the second is the one that keeps the behaviour alive. The first is arithmetic. Every time the discount is demanded and the other side does not meet it, the option goes untaken. Some of those options were the better ones. The composition was unstated for a reason that had nothing to do with the person deciding. The better options are simply forgone, and on the illustrative scale above the size of what is being forgone at no disclosure is 18.0 per cent of the exposure, or Rs 18,000/- on Rs 1,00,000/-.

The second cost is that the rule never gets corrected by experience, and this is structural rather than psychological. When an option is declined, what it would have done is never learned, so the cases where the rule cost something are invisible afterwards while the cases where it saved something are vivid. A household that walked away from the second flat never lives in it and never finds out that the boiler was newer. The feedback that would tune the rule is exactly the feedback that withdrawal destroys. The asymmetry is why the rule keeps its confidence for decades without ever being tested.

There is a third cost that only appears in aggregate, and it belongs to the effort mechanism rather than the odds one. A person who withdraws from every effortful decision does not end up with no decision; they end up with whatever happens when nobody decides, and what happens is usually money sitting still. Money sitting still is not neutral and it is not free. The mechanism behind staying put is set out under status quo bias and defaults.

How does a practitioner actually use this, and how does somebody deciding alone?

Devika Rao gets one observation and has to work backwards from it: a person who has not chosen. Persuasion aimed at the wrong mechanism makes things worse in a way she can predict, so her first move is not to persuade. Her first move is to establish which mechanism she is looking at, and one question does it. She asks what the person would need in order to decide. If the answer names something the document never states, the cause is unstated odds and the repair is to state it. If the answer is that it is all there and there is too much of it, the cause is effort and the repair is a shortlist with its basis written down.

The second practitioner move is to notice when the discount is being paid to her own side. A practice that answers a client's discomfort with reassurance rather than with composition has collected the ambiguity premium instead of removing it. The check is whether the client can now state the thing they could not state before. If they cannot, and they feel better anyway, then what was sold was the relief.

For a person deciding alone, with no adviser and no committee, the same two moves work without any of the machinery. The first is to write down the one thing that would need to be known in order to decide. The second is to establish whether that one thing is missing from the document or merely buried in it. Missing is a reason to walk away and it is usually a good one. Buried is a reason to spend twenty minutes with a pencil. The information is there, and the only thing standing in the way is work. Missing and buried feel identical from the inside and call for opposite responses, so the single most useful habit is separating them before deciding.

Meera Sundaram, Devika Rao, Palash Advisory Services Private Limited and the Palash decision log are invented.
Educational material. Not advice on any investment, tax, budget or market position.