Herding, narratives and bubbles

The temptation is to explain bubbles with mass foolishness. That explanation is both unkind and inadequate, because it cannot account for the participation of careful, well-informed professionals — and they always participate.

Why following others is usually correct

Start by taking herding seriously as a strategy. If a hundred people have examined a question and reached a conclusion, their conclusion is evidence. Ignoring it in favour of your own single judgement is not independence; it is discarding information.

This is why herding is not a bias in the way the disposition effect is. It is a reasonable rule that fails in a specific circumstance, and the circumstance is worth stating exactly.

Informational cascades

Suppose each person has a little private information and can see what others did, but not why.

The first person acts on their own information. The second sees that action and weighs it against their own. By the tenth, the observed behaviour of the previous nine outweighs almost any private signal — so the tenth person rationally ignores their own information and follows.

From that point, no new information enters the sequence. Everyone after is copying a chain whose entire informational content came from the first few participants, who may have been wrong. The crowd looks like a hundred independent confirmations and is actually one or two, repeated.

That is the mechanism, and nothing in it requires a single irrational actor. It requires only that people cannot see each other's reasons.

Agency herding

A second mechanism, independent of the first, and in professional markets probably stronger.

A manager's outcome is judged relative to peers and against a benchmark. That creates an asymmetry:

• Wrong alongside everyone — a bad year, shared, survivable
• Wrong alone — client withdrawals, perhaps a career ending
• Right alone, eventually — possibly preceded by years of underperformance during which the clients leave

So the penalty for being wrong with the crowd is far smaller than for being wrong against it, and the reward for being right early is often negative. A manager who maximises their own survival will therefore stay close to the consensus even while believing it mistaken. This is not a failure of character; it is the incentive working as built.

Narratives

A bubble needs a story, because a price alone cannot be repeated at dinner. Good bubble narratives share features:

• Partly true. The internet did change commerce. The best narratives are not fantasies; they are real insights with the valuation consequences overstated.
• Hard to falsify in the short run, because they concern a distant future.
• Self-confirming through price. Rising prices are taken as evidence for the story that is causing them.
• Dismissive of measurement. The appearance of the argument that traditional valuation no longer applies is the single most reliable late-stage marker, precisely because it removes the only instrument that could contradict the story.

The lottery preference

There is a related and separate finding worth holding alongside this: investors show measurable preference for stocks with high idiosyncratic volatility and high idiosyncratic skewness — payoffs that are mostly small losses with a small chance of something enormous. The same demographic characteristics that predict buying lottery tickets predict the strength of this preference in stocks.

That is a demand for the shape of a payoff rather than for its expected value, and it explains why speculative manias cluster in assets that can plausibly be described as going up a hundredfold. India's derivatives data show the pattern at scale: participation rising to 96 lakh individual traders while 91% of them lost money. People were not buying an expected return. They were buying a distribution.

Working the problem

The irrationality account. The manager believes the asset is overpriced and buys it, which is acting against their own stated beliefs — the textbook definition of irrationality. If enough professionals do this, prices stop reflecting informed opinion and the market's main social function fails.

The rational account. The manager is not investing their own money; they are running a business whose survival depends on not diverging from peers for long. Being right in three years is worth nothing if the fund closes in one. Given the contract they operate under, buying is the correct decision for them — and the mispricing persists not despite professionals but because of how they are paid.

Which is more persuasive. The second, and strongly, because it explains more. Mass irrationality cannot account for sophisticated participants who correctly identify a bubble and participate anyway, and such participants are documented in every episode. Agency herding predicts exactly that, and it also predicts the timing: the herd breaks when withdrawal risk flips, so that holding the asset becomes the thing clients punish. That is why unwinds are abrupt rather than gradual.

The uncomfortable corollary is that "the professionals are still buying" carries much less information than it appears to, since their buying may be an artefact of their incentives rather than their judgement.

The honest limit of this chapter

Bubbles are far easier to identify afterwards. At any moment there are many assets that have risen a great deal, of which some are bubbles and some are correctly repricing. Nobody has a reliable advance test, and people who claim one have usually been predicting a crash continuously.

So this chapter cannot be converted into timing. What it gives you is narrower and still useful: recognising the mechanisms means not mistaking a cascade for a hundred confirmations, and not mistaking professional participation for professional endorsement.

The point

Following the crowd is usually sensible because other people's conclusions are evidence, and it fails when observed behaviour drowns out private information — after which a cascade of a hundred people carries the information of the first two. A separate and stronger mechanism operates among professionals, who are punished for being wrong alone far more than for being wrong together, so they stay with a consensus they disbelieve. Both mean a bubble requires no irrational participant, and neither yields a test you can apply in advance.