This is part 6 of Crisis Money, a Money Outsider series about what financial crises do to our minds, and what our minds do back. It stands on its own, but part 5 covered the pull of the herd.
Herding runs from Isaac Newton, who lost a fortune buying back into the South Sea Bubble in 1720, to the fund managers whose careers reward losing money the same way as everyone else. Two ideas matter here: the information cascade, in which rational people suppress their own doubts because the crowd looks better informed, and the neuroscientist Vasily Klucharev's finding that the brain treats disagreement with a group as an error. The herd has since moved onto your phone.
The digital herd
On 11 January 2021, shares in GameStop, a struggling American video game retailer, were trading at around $20. Seventeen days later, on 28 January, the stock hit an intraday high of $483. Then it collapsed. By mid-February it was back below $50.1
What happened in between was the most visible demonstration of digital herding in financial history. Members of the Reddit forum WallStreetBets, a community of roughly two million amateur traders at the time, had identified that several large hedge funds held significant short positions in GameStop. A short position is a bet that a stock will fall. If you can force the price up instead, the short sellers face theoretically unlimited losses and may be forced to buy shares to close their positions, pushing the price higher still. This is a short squeeze, and WallStreetBets decided to engineer one.
What made GameStop different from previous short squeezes was the mechanism of coordination. There was no central organiser. Nobody was issuing instructions. The subreddit functioned as a real-time information cascade, with users posting screenshots of their positions, celebrating gains, mocking hedge funds, and shaming anyone who suggested selling. The social dynamics were as important as the financial logic. Selling wasn't just leaving money on the table. It was betraying the group. The language of the forum, "diamond hands" for holders, "paper hands" for sellers, made conformity a badge of identity.
Long and colleagues, in a 2023 study in The Financial Review, analysed 10.8 million Reddit comments and found that sentiment on WallStreetBets directly affected GameStop's intraday returns at 5-, 10-, and 30-minute frequencies.2 A 2024 study found that positions initiated at peak WallStreetBets attention produced average holding-period returns of negative 8.5 per cent.3 The people who arrived at the peak of the herd's enthusiasm, which is to say the majority, lost money.
The GameStop saga was American, but the dynamics it demonstrated are running constantly in UK markets. The FCA's 2022 cryptoasset consumer research found that 4.97 million UK adults, roughly 9 per cent of the adult population, owned cryptocurrency.4 Of those who held crypto, 34 per cent had first learned about it from friends or family. Fewer than half said they understood the underlying technology. And 45 per cent reported losses on their holdings. Bitcoin, the most widely held crypto asset, had fallen by roughly two-thirds from its November 2021 peak by the time the FCA survey was conducted in mid-2022.
Robert Cialdini described the mechanism in 1984 in his book Influence.5 Social proof, the tendency to look at what other people are doing when deciding what to do yourself, operates most powerfully under two conditions: when you are uncertain what to do, and when the people you observe are similar to you. Both conditions were met perfectly during the UK crypto boom. People who knew little about blockchain watched their friends and colleagues making returns they couldn't ignore. Social media made the gains visible and the losses invisible. The cascade formed fast.
What separates digital herding from its historical predecessors is speed and transparency. In 1720, Newton watched the South Sea Bubble develop through conversations in coffeehouses and letters from friends, over weeks and months. In 2021, retail investors watched GameStop's price move in real time on their phones, with a live commentary thread explaining (or misexplaining) what was happening. The time between seeing the herd move and joining it has compressed from months to minutes. The information cascade that Bikhchandani described as a sequential process, one person watching the next, now happens simultaneously across millions of screens. The dynamics are the same. The speed is different. And speed, in financial markets, kills.
There is also a structural change that deserves attention. In Newton's time, and even in 2007, joining a financial herd required some friction. You had to contact a broker, or at least walk to a bank branch. That friction was, in a quiet way, protective. It gave people time to reconsider. Trading apps have eliminated that friction entirely. A 22-year-old in Birmingham can see a cryptocurrency surging on TikTok and buy it within ninety seconds without speaking to another human being. The time gap between herding impulse and herding action has collapsed to nearly zero. The brain's error-correction mechanisms, the ones Klucharev identified, don't have time to operate. The herd moves, and you move with it, before you've finished thinking about whether you should.
When the herd is wise, and when it isn't
There is a version of the herding story that treats all crowd behaviour as pathological. That version is too simple.
In 2004, the journalist James Surowiecki published The Wisdom of Crowds, arguing that under certain conditions, the collective judgement of a group reliably outperforms the judgement of any individual member.6 His conditions were specific: the group needed diversity of opinion, so that members brought different information and perspectives; independence of decision, so that each person formed their own view without pressure to conform; decentralised knowledge, so that different people had access to different local information; and a mechanism for aggregation, so that individual judgements could be combined into a collective answer.
The jelly-beans-in-a-jar experiment is the classic illustration. Ask a hundred people to estimate the number of jelly beans in a jar. Most individuals will be wildly wrong. The average of all their guesses will be remarkably close to the true number. Prediction markets work on the same principle: when diverse, independent bettors trade on the probability of an event, the market price tends to be more accurate than any individual forecast.
The problem is what happens when any of Surowiecki's conditions breaks down. Herding, by definition, destroys independence. When people observe and mimic each other's choices, the diversity of information in the group collapses. You no longer have a hundred independent estimates. You have a hundred copies of the same few estimates, amplified by repetition.
In 2006, Matthew Salganik, Peter Dodds, and Duncan Watts at Columbia University ran an experiment that demonstrated this with uncomfortable precision.7 They created a website called MusicLab, where 14,341 participants could listen to songs by unknown bands and choose which ones to download. Some participants were assigned to an independent condition: they could see the songs but not how many times each had been downloaded. Others were assigned to social-influence conditions, where download counts were visible.
In the independent condition, quality predicted success. Better songs got more downloads. In the social-influence conditions, the same songs existed, but the results were profoundly different. Popular songs became more popular and unpopular songs became less popular, regardless of quality. Inequality between successful and unsuccessful songs increased. And, most disturbingly, the outcomes became less predictable. A song that topped the charts in one social-influence world might languish in obscurity in another. The initial random advantage of a few early downloads snowballed into permanent dominance, or didn't, depending on what the first few people happened to do.
A 2008 follow-up by the same team went further. They inverted the download counts, telling participants that the least popular songs were the most popular and vice versa. Even the songs with fabricated popularity benefited. Social influence was powerful enough to override quality signals entirely. The herd followed fabricated signals because the signal of "other people chose this" was more compelling than their own ears.
The parallel to financial markets should be obvious. When stock price movements become the primary signal that other investors watch (and they do), price itself becomes social proof. A rising price says "other people are buying this." A falling price says "other people are selling this." And those signals trigger the same conformity mechanisms that Klucharev identified in the brain, the error signal that fires when you disagree with the crowd.
This is where contrarian investing gets interesting, and where the data confounds intuition. If herding pushes prices away from their fundamental values, then going against the herd should, on average, produce excess returns. The evidence, broadly, supports this. Werner De Bondt and Richard Thaler showed in 1985 that stocks with the worst recent performance subsequently outperformed stocks with the best recent performance, a pattern they attributed to overreaction by herding investors.8 Lakonishok, Shleifer, and Vishny found in 1994 that value strategies, which are inherently contrarian, produced higher returns not because they were riskier but because they exploited predictable errors in crowd behaviour.9
But contrarian investing requires a psychological tolerance that most people, including most professionals, do not possess. You are buying what everyone else is selling, or selling what everyone else is buying. Your brain is treating every trade as an error. Your colleagues think you've lost the plot. Your clients are calling. And if the position goes against you before it comes good, which it often does, the career risk identified by Scharfstein and Stein kicks in with full force. Being right eventually is cold comfort if you've been fired in the meantime.
This is the genuine difficulty at the heart of the herding problem. The crowd is sometimes wise and sometimes catastrophically wrong, and the conditions that determine which one you're in, Surowiecki's four criteria, are exactly the conditions that are hardest to assess from inside the crowd. If you could always tell whether you were in a wisdom-of-crowds situation or a bubble, there would be no bubbles.
We have built markets that systematically disable the conditions under which crowd wisdom works, and then we express surprise when the crowd turns out to be wrong.
What Salganik's experiment really showed is that the same group of people, with the same information and the same preferences, can produce wildly different outcomes depending on whether they can see each other's choices. Remove social visibility and quality wins. Add social visibility and randomness wins. Financial markets, by design, are environments of maximum social visibility. Every price is a signal of what other people are doing. Every Bloomberg terminal, every trading app, every fund performance table is a mechanism for transmitting social information. We have built markets that systematically disable the conditions under which crowd wisdom works, and then we express surprise when the crowd turns out to be wrong.
The herd is global and the platforms are American, but a crowd still runs with a national accent. The next part turns to the British herd, and how a nation of queuers behaves when the running starts.
Next in this series: The British herd. If this was useful, subscribing gets you the rest of the series as it lands.
GameStop/r/WallStreetBets episode, January 2021. Shares rose from approximately USD 17 to intraday high of USD 483, then collapsed below USD 50. SEC (2021) Staff Report on Equity and Options Market Structure Conditions.
Long, C. et al. (2023) 'Reddit sentiments and GameStop intraday returns', Financial Review, 58(1), pp. 19-37. 10.8 million comments analysed; 5-30 minute price effects.
International Review of Financial Analysis (2024) 'WallStreetBets attention and uninformed trading'. Positions opened at peak attention produced negative 8.5% average holding-period returns.
Financial Conduct Authority (2022) Cryptoassets Consumer Research. 4.97 million UK adults held cryptoassets; 9% ownership rate.
Cialdini, R. (1984) Influence: The Psychology of Persuasion.
Surowiecki, J. (2004) The Wisdom of Crowds. Random House.
Salganik, M.J., Dodds, P.S. & Watts, D.J. (2006) Experimental study of inequality and unpredictability. Science, 311(5762), pp. 854-856.
De Bondt, W.F.M. & Thaler, R.H. (1985) Does the stock market overreact? Journal of Finance, 40(3), pp. 793-805.
Lakonishok, J., Shleifer, A. & Vishny, R.W. (1994) Contrarian investment extrapolation and risk. Journal of Finance, 49(5), pp. 1541-1578.