When other people become the evidence

Copying is usually the right move

Following a crowd is often correct. Other people's actions carry information you do not have, and reading it off their behaviour costs far less than gathering it yourself.

Two restaurants on the same street, one full and one empty, and you have eaten at neither. Choosing the full one is not weakness. It is an inference from the only evidence available, which is that a hundred people who did know something about the food acted on what they knew.

Markets run on that same inference, and mostly it works. A price is a compressed summary of what a large number of people concluded after doing work you did not do. Treating other people's decisions as evidence is the ordinary, sane behaviour of somebody who has a job and cannot research everything.

Which is exactly what makes the failure hard to catch. The failure is not some different, worse behaviour. It is the same behaviour, still being applied after the condition that made it work has quietly stopped holding — and that condition never announces its own departure.

What makes a crowd informative is independence, not size

Francis Galton reported in Nature in 1907 that when a crowd at an English country fair guessed the weight of an ox, the middle guess landed close to the true weight — in a crowd that mixed butchers and farmers used to judging cattle with people who were not. The version of the story that travels is stronger than the paper, and the strengthening always happens in the same direction: it makes the crowd sound magical rather than arithmetical.

The arithmetic is worth doing, because it is what names the condition. Average a set of estimates that are each wrong in their own direction and the errors partly cancel: the spread of the average shrinks with the square root of the number of estimates, so a hundred independent guesses produce an average whose typical error is roughly a tenth of a single guess's. That, and nothing more mysterious, is where the accuracy comes from. Which also says exactly when it fails — the errors have to point in different directions, and a mistake everybody shares does not cancel, however many people share it.

Now break the independence. Suppose ninety-nine of those hundred people read the first person's card before writing their own, and wrote down something close to it. The average is now the first person's guess wearing a crowd's clothes. The square root did no work at all, because there was only ever one estimate in the pile.

So the quantity that decides whether a crowd is evidence is not how many people are in it. It is how many of them decided before seeing the others — the number of independent readings. A crowd of a hundred thousand containing four independent readings is a crowd of four with a very loud voice, and its size tells you nothing about which of the two you are looking at.

How a crowd stops carrying information

The mechanism has a name and a formal treatment. Two papers published in 1992 — Abhijit Banerjee in the Quarterly Journal of Economics, and Sushil Bikhchandani, David Hirshleifer and Ivo Welch in the Journal of Political Economy — worked out what happens when people decide one after another, each able to see what the earlier ones did but not what they knew.

Set it up concretely. Ten people decide in turn whether to buy something. Each holds a private hint that is right more often than it is wrong, and each can see the choices made before theirs. The first two act on their own hints, so their choices reveal something about what they held. By the time the third person sees two buyers, two visible purchases outweigh one private hint, and buying is the reasonable move even for someone whose hint pointed the other way.

Follow what that does to the information. The third person's purchase now says nothing about their hint, because they would have bought either way — and the fourth, watching three purchases, sits in the same position. Every decision after the second transmits no new evidence while still adding to the pile that everyone afterwards is reading. The authors called it an informational cascade.

Two properties follow, and both are mechanical rather than psychological. A cascade can settle on the wrong option, because it locks in on the first two or three hints and those may have been the unlucky ones. And it is fragile — one small piece of public information can outweigh the entire visible crowd, precisely because the crowd was only ever carrying two hints' worth of evidence.

Lisa Anderson and Charles Holt reproduced this in a laboratory in 1997, in the American Economic Review, using draws from an urn so that the correct answer and each participant's private signal were both known to the experimenter. Cascades formed, including on the wrong urn. How far that carries into a market is a separate question, and an open one — the laboratory controls the single thing a market never does, which is who saw what.

There is a second channel, distinguished by Morton Deutsch and Harold Gerard in 1955. Informational influence is copying because you think the others know something; normative influence is going along because being the odd one out is costly. Solomon Asch's line-judgement studies of the early 1950s showed people agreeing with a majority that was plainly wrong about something they could see with their own eyes. The size of that effect is genuinely contested, and popular retellings overstate it — but the existence of the second channel matters here, because it means a crowd can hold together among people who privately disagree with it.

The confidence gradient runs backwards

Here is the part that makes this so hard to see from the inside, and it falls straight out of the cascade above.

In a crowd that genuinely carries information, the first movers are the informed ones and they are the least confident — acting on a private hint that might be wrong, with nobody yet agreeing with them. The last movers are the most confident, because they can see the largest amount of agreement. They are also holding the least information, because agreement is the thing they substituted for their own hint.

Confidence and evidence therefore run in opposite directions along the queue. A crowd is loudest and most certain at the point where its information content is lowest, because certainty here is manufactured out of observed agreement, and observed agreement is manufactured out of copying.

Two consequences worth carrying. “Everyone I know is doing it” is a weaker argument the more people it covers, which is the exact reverse of how the sentence feels when you say it. And a reversal, if one comes, needs very little new information to start — there was not much information holding the position up.

FOMO is a comparison, not greed

The pull people describe as fear of missing out is not a hunger for money. It is a response to a comparison, and the thing being compared with is a version of yourself who acted.

Graham Loomes and Robert Sugden set this out formally in 1982 as regret theory: a choice carries not only its own outcome but the outcome of the alternative you turned down, and the gap between the two is felt as part of the result. On that account, watching a stock you considered and skipped rise sharply afterwards is not neutral information. It is experienced as something that happened to you, which is why it produces an urge to act rather than a fact to file.

That explains a specific asymmetry in what people find intolerable. A portfolio that did nothing while nobody else made money is comfortable. The identical portfolio, while a visible group made money, is not — the same holding produces a different urge depending on what is visible beside it, and nothing about the asset changed between the two cases.

It also explains why the pull strengthens as the evidence weakens. The visible group is largest, and its gains most conspicuous, late in a run — which by the section above is exactly when the crowd is mostly reflection. Regret and information peak at opposite ends of the same episode, and acting on the first while believing you are acting on the second is one of the routes into the gap between what a fund returned and what its investors got.

IPO subscription: a number that partly measures itself

A company selling shares to the public for the first time — an initial public offering — publishes its own demand while that demand is still forming. Through the bidding window the Indian exchanges show category-wise subscription, meaning how many times over the retail, institutional and non-institutional portions have been applied for, and by the last day that multiple is the headline in every report.

Read it as evidence and it turns out to be a strange kind. Part of what it measures is the allotment rule rather than conviction: the retail portion is allotted in minimum lots and, where demand exceeds supply, by draw, so one large application does not buy a proportionally larger allotment. The rule itself produces a great many small applications. The multiple is partly an artefact of how allotment works, and reading all of it as intensity of belief counts the same thing twice.

Then the grey market premium, quoted informally by dealers ahead of listing. It is not an exchange-cleared price and carries no settlement obligation, and yet it functions as the crowd's estimate of what the crowd will pay on listing day. Keynes described this shape in 1936, in chapter 12 of The General Theory, with his newspaper beauty contest: entrants were not picking the faces they found prettiest but the faces they expected other entrants to pick, and the more skilful players were working a level further up again.

Note carefully what is and is not being claimed. That public issues are systematically bad investments is not a claim this article makes — long-run performance after listing is an empirical question, studied since Jay Ritter's 1991 paper on US issues, and the answer moves with the sample, the period and the benchmark chosen. What is mechanical is much narrower: a subscription multiple is a count of applications, not an independent valuation, and a grey market premium is an opinion about opinions.

The sectoral fund that arrives after the run

A fund house launches what it can sell. What it can sell is what has recently done well. Those two sentences together are the whole mechanism.

So a launch calendar is a lagging record of returns rather than a leading one. Sectoral and thematic schemes cluster after a sector has already run, because that is when the trailing numbers make the pitch and the pitch is what fills the window in which a scheme sells units before it owns anything — a new fund offer. Nobody has to behave badly for this to happen — a sector nobody is interested in simply does not gather enough subscription to launch at all.

The consequence for a reader is a mismatch between two records that get read as one. The category shows a strong trailing return; the scheme itself shows none, because a new scheme's own track record starts on the day it launches, and the run that made it saleable sits entirely before its history begins. What is being bought on the strength of the sector's past is a scheme with no past.

The launch price adds a second layer, and this one is arithmetic rather than opinion. Units in a new fund offer are conventionally issued at ₹10, which is an accounting starting point and not a discount — the value of one unit says nothing about whether what the scheme will hold is cheap. A scheme priced at ₹10 that is about to buy recently-run-up shares is a scheme that holds recently-run-up shares.

The specific error, stated so it is recognisable: treating the number of people subscribing to a launch, or the volume of coverage around it, as information about the assets the scheme will go on to buy. Those are facts about the crowd. And the trailing return that assembled the crowd is the same thing that raised the prices the scheme now has to pay — which is one reason a broad-market holding and a sector bet get compared on the wrong axis.

The feed is a selected sample, twice over

Social media does not so much lie about individual trades as show you a sample that was chosen twice before it reached you.

The first selection is made by the poster. A position that worked gets a screenshot; the one that did not gets no post at all. Nobody has to be dishonest for this to bite, because the selection happens per post rather than per person — someone can report every trade they choose to report with complete accuracy and still produce a feed with almost no losses in it.

The second selection is made by the platform, which shows what earns engagement, and a large gain earns more of it than a small loss. You are looking at a selection of a selection, and the two compound rather than cancel.

What is missing from all of it is the denominator. A feed can tell you that a hundred people made money on something; it cannot tell you how many tried. Ten thousand attempts with a hundred successes produces exactly the feed in front of you — and so does a hundred attempts with a hundred successes. Those two worlds imply opposite things and generate identical timelines.

The same filter runs offline at lower volume. A neighbour mentions the flat that doubled and not the one that would not sell; the colleague who did well in a stock raises it and the one who did badly changes the subject. It is the same property that lets holdings that never print a loss sit unexamined for decades — without a visible negative number, nothing prompts the question.

Separating evidence from reflection in your own decisions

None of this argues for ignoring other people, which would throw away most of what anyone knows. It argues for asking one thing about a crowd before treating it as evidence: how much of it decided independently?

Five questions that separate the two, each answerable without holding any view on where the market goes next:

• How many independent readings are in here — how many of these people reached the conclusion before watching the others reach it?
• What each of them knew — did they have access to something about the asset, or only to each other?
• Is the quoted figure a price or an opinion about a price — a cleared trade and a circulated estimate are different objects wearing similar clothes.
• Does the argument survive without the count — strike the number of participants out of the case and see whether anything is left standing.
• What is the denominator — of everyone who tried this, what share is visible to me?

And one that catches the regret channel specifically: would this still be worth holding if nobody I knew held it? The question works because it removes the comparison term without altering a single fact about the asset. If the answer moves, the comparison was doing the deciding.

The honest limit on all of it: none of these questions tells you whether a price is right, and a crowd echoing itself can be correct by accident. What they establish is narrower and still useful — whether the confidence you are borrowing from other people is backed by anything, or is just the sound of the copying.

Reading a run from every start date instead of one

Everything above concerns the evidence behind a decision rather than any particular asset. One piece of it is testable rather than arguable: whether the run that is drawing the crowd has already happened.

FNOTrader's Mutual Funds app computes rolling returns across every available start date rather than the single window ending today, alongside the worst peak-to-trough fall. It runs on the daily per-unit values — net asset values, or NAVs — published by AMFI, the mutual fund industry body: around 34 million rows of them. A trailing number that everyone is quoting reads differently once the same holding period is measured from every start date instead of the flattering one.

The same history answers the other half. An entry made near the top of a run is judged by the drawdown along the path, which is what decides whether a holding survives being held — and that is knowable from the record in a way the next move never is. Past performance is a record of what happened, not an indication of what will.

FNOTrader is not a SEBI-registered investment adviser and this is not investment advice.

Common questions

Continue reading

More in Behavioural Finance · App: Mutual Funds · Definitions: glossary · Free tools: calculators · All: every article