Noise Traders: Trading on Something Other Than Information

A noise trader is somebody trading on something other than information: a story, a display, a recent move, a feeling, or the visible action of others. Noise is not stupidity and it is not rare. Noise is the ordinary condition of most trading. Two informed parties rarely want opposite sides, so without noise there would be almost no trading at all.

Here is a word that sounds like an insult and is not one. The word is not a verdict on a person, and calling a decision noisy says nothing at all about the intelligence of whoever made it. Black set the term out in the paper Noise, in the Journal of Finance in 1986, and what he was pointing at was not carelessness. He was pointing at the ordinary condition of a room full of people, every one of whom has a perfectly good reason to act and almost none of whom is holding anything nobody else holds. The useful question is never whether somebody is a noise trader; it is what a given decision was made of.

What is noise, and what exactly is it noise against?

The word only means something once what it is being measured against has been stated. Every use of it carries a hidden benchmarkThe standard a thing is judged against. Change the standard and the same act can pass or fail., and that benchmark is information about the thing being decided. Set the benchmark down in the open and the word becomes usable. NoiseAnything moving a decision that is not information about the thing decided. is anything that moves a decision without being information about what was decided.

Take it out of investing first. A household is choosing a school. Information about the school is the class size, the distance from home, the fees, and what the last three years of leavers went on to do. Noise is the fact that a cousin mentioned the name at a wedding on Saturday. The mention is real, it genuinely moved the decision, and it contains nothing whatever about the school. The structure does not change when the decision is about money instead of a school.

So the benchmark is not a test of intelligence. The benchmark is a question about content. Adding Rs 1,00,000/- to a holding on the evening a television segment named it may be a perfectly sensible act, and the segment may even have carried something worth knowing. But the mention itself is not information about the holding; it is a prompt, and a prompt and information are two different things that happen to arrive at the same moment. The 19 February entry in the invented Palash decision log has exactly that shape: Rs 1,00,000/- went into Suvarna Chemicals Limited the same evening a segment named it, and the record holds the date and the amount and nothing at all about what was examined.

What does somebody trading on noise believe they are doing?

Nobody thinks of themselves this way, and that is a structural fact rather than a moral one. From the inside, a decision moved by a promptSomething that triggers a decision without necessarily carrying any information about it. feels exactly like a decision moved by information. The prompt arrives wearing the same clothes. The television segment felt like learning something. The recent move felt like evidence. The visible action of other people felt like a fact about the world, and in a narrow sense it was one.

A person trading on noise believes they are trading on information, and that belief is the ordinary state rather than the exception. So the check has to be built out of something other than a feeling of confidence. Confidence is produced by fluency, and a story told well is fluent whether or not there is anything behind it.

There are only two questions available, and both are asked about one decision, after the act rather than during it. The first is what did I examine before I acted. The second is whether the same act would have followed had the prompt never arrived. Neither can be answered about a person in general, and neither can be answered from a record of trades. Only the person who acted can answer them, honestly, about one act at a time. If the answer to the first is nothing in particular, and the answer to the second is no, then the prompt did the work, and the label belongs to that decision and to nothing else.

What is an Information Cascade, and how does it manufacture information from nothing?

Noise is not only the absence of information. Very often it is the product of a process that felt informative to everybody standing inside it, and that is the uncomfortable part.

The mechanism is simple to state. People decide in sequence, each one can see what the people before them did, and after roughly three of them the private evidence stops arriving. Banerjee set the structure out in the Quarterly Journal of Economics in 1992, and Bikhchandani, Hirshleifer and Welch set out the cascade that locks, in the Journal of Political Economy in the same year. The output of the mechanism is what matters. Each new person adds one visible choice to the tally and no fresh examination to the evidence. The tally grows. The evidence stands still.

Picture ten people choosing between two food stalls on a street none of them knows. The first two look at the queues, read the menus, glance at the state of the counters, and pick the left stall. The third arrives, sees two people at the left stall, and joins them. So does the fourth. By the tenth, ten people are eating at the left stall and exactly two of them examined anything at all. An observer arriving at that moment sees what looks like ten independent judgements and is in fact looking at two judgements and eight copies. That is how an information cascadeA sequence in which copying manufactures apparent information out of order alone. manufactures apparent information out of sequence, and it is why noise gets produced in quantity by people every one of whom acted sensibly.

Noise Trading vs Informed Trading: what actually separates them?

Informed tradingTrading on information that other people do not yet hold. means acting on information others do not yet hold. Noise tradingActing on something that is not information about the thing being decided. means acting on something that is not information about the thing decided. Written down side by side the difference looks obvious. In a record of what people actually did, it is invisible.

Take two entries. On 19 February somebody bought Rs 1,00,000/- of a holding after reading three years of its accounts, working out what it earns against what it costs, and concluding that the terms on offer were worth taking. The same evening somebody else bought Rs 1,00,000/- of the same holding because a television segment named it. The two entries carry the same date, the same action, the same amount and the same instrument. The examination that separates the two entries happened before either purchase, and it left no trace in the record of either.

The blank is not a fault in one particular log. Records of actions carry the same property in general. A record captures what was done. The examination lives in the head of the person who did it, and unless somebody writes it down at the time, the examination is gone. Reconstructing it a year later is worse than useless. The reconstruction gets built out of what happened next rather than out of what was known before. So the test cannot be a figure at all. The test has to be a process that captures the reason at the moment of acting. A written reason does exactly that job.

What does the decision log actually show?

The Palash decision log covers sixty investors across eight quarters, and holds 240 logged decisions, with Meera Sundaram among them. The log has the shape every record of this kind has.

Of the 240 decisions, a written reason was recorded on 84. Work that out: 84 divided by 240 is 35.0 per cent. The remaining 156 decisions record what was done and never record why. Now work the second measurement the same way. Of the 96 buys, 41 followed a media mention within three days. Divide 41 by 96 and the share is 42.7 per cent. The base rateHow often the thing happens in general, before any particular case is considered., meaning the share of the eligible list mentioned anywhere at all in a given week, was 11.0 per cent. So the buys clustered on mentions at 42.7 against 11.0. The clustering runs 3.9 times as often as chance alone would put them there.

The clustering leaves one question. How many of those 41 buys were informed?

The log cannot say, and no record built the same way ever can. A mention is not information and it is not the absence of information. A mention is a prompt, and whether anything informative arrived alongside it is precisely what a record of actions never captures. The clustering is real and it is measurable. The contents of the clustering are unknown, and they will stay unknown. The 156 blank reason fields are where the answer would have been.

Why is noise necessary rather than merely tolerated?

One consequence surprises people. Without noise there would be almost no trading at all.

Think about what a trade requires. Somebody has to want to sell exactly what somebody else wants to buy, at the same moment, in the same amount. Now suppose both parties are informed, both have read the same accounts, and both have reached the same conclusion about the value of the holding. Neither of them wants the opposite side of the other. Agreement produces no trade whatsoever.

A trade needs somebody on the other side who is there for a different reason, and most of the time that reason is not information. Black made exactly this the centre of the 1986 paper: noise is what makes trading possible, and it is also what makes trading hard to interpret, and those are the same fact seen twice. De Long, Shleifer, Summers and Waldmann carried the argument further in the Journal of Political Economy in 1990, showing that noise becomes a risk in its own right, one that anybody betting against it has to carry.

Correcting a mispricing is costly and risky, so a mispricing can persist. Shleifer and Vishny set that idea out in the Journal of Finance in 1997, and it is taken in full under limits to arbitrage. The lesson for one person is smaller and more human. The person on the other side of a decision is usually there for a reason that has nothing at all to do with what was examined.

What does noise cost the person producing it?

Noise is not free, and the cost does not appear where people look for it. The log's sixty investors sort into five turnoverHow much of a holding is bought and sold over a year, measured against the size of the holding. groups of twelve. Annual turnover runs 9, 34, 71, 128 and 210 per cent across the five. Gross returns, before costs, run 11.2, 11.0, 11.1, 10.9 and 11.0 per cent. Net returns, after costs, run 10.9, 10.4, 9.6, 8.4 and 6.9 per cent.

Read the two spreads rather than the five levels. The gross returns sit inside 0.3 percentage points of each other, from 10.9 to 11.2. The net returns run 4.0 points apart, from 6.9 to 10.9. The distance between the busiest group and the quietest is almost entirely what the trading cost and almost nothing to do with what was picked. Costs here mean dealing charges, the spread and tax, counted together.

Odean measured a shape of this kind in the Journal of Finance in 1998, and Barber and Odean in the Journal of Finance in 2000. The cost of acting on prompts lands not in worse selections but in a larger number of them.

Where does trading on something other than information make sense?

Not all of it is damaging.

Somebody sells a holding to pay a hospital bill. The sale is not informed by any definition, and it is not the damaging kind of noise. The reason is real, it sits entirely outside the holding, and it would have been the same reason whatever the holding happened to be doing that week. Rs 25,000/- a month going in by standing instruction is the same shape approached from the other direction: the reason is a rule adopted in advance, not a signal read on the day. Pausing that standing instruction because income stopped is a change in the facts of a life rather than a change in the facts of a holding.

The harmless kind rests on a real need, and the damaging kind rests on a signal mistaken for a need. A bill is a need. A recent move is a signal, and treating a signal as a reason is where the cost sits. The log does not record whether income stopped or a segment ran the night before, so the 24 pauses of a standing instruction across its eight quarters prove nothing by themselves. The gap is the same one that runs through every record of actions, and it is why a written line matters more than any figure.

How does an adviser use any of this without accusing anybody?

Devika Rao, the adviser at the invented Palash Advisory Services Private Limited, cannot look at a record and separate informed from noisy. Nobody can. She can change what the record holds in the first place.

On 4 November, 20 of the sixty investors adopted a written checklist: before acting, write one line saying what was examined. Across quarters 5 to 8 that group recorded a written reason on 34 of 41 decisions, a share of 82.9 per cent. The other 40 recorded one on 19 of 63, a share of 30.2 per cent. Their realisation ratio, the measure covered under selling winners and holding losers, fell from 3.2 to 1.6.

Eight quarters and sixty people cannot support a claim about returns, so the checklist says nothing about whether returns improved. The checklist produced a record that can answer the question the log could not. Changing what the record holds is the whole of the professional use, and it is available to a person deciding alone just as much as to a practice deciding for others. A household can run the same thing on one line in a notebook: before acting, a single line recording what was examined, read back a year later. The value is not the discipline of writing it; the value is that the person arguing against that line next year is the one who wrote it, and by then the feel of the prompt will have been completely forgotten.

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