How to spot a bubble bursting

Ray Dalio spied the dotcom bubble early. “We’re approaching a blow-off phase of the US stockmarket,” said the founder of Bridgewater, one of the world’s biggest hedge funds. Peter Lynch, the celebrated manager of Fidelity’s Magellan fund, thought “not enough investors are worried”. Howard Marks, a pioneering investor in junk bonds, very much was worried, since “every cocktail-party guest and cab driver just wants to talk about hot stocks and funds.” George Soros put his neck on the line and short-sold internet stocks outright. Warren Buffett refused to touch them, saying he could not “see what technology businesses will look like in ten years or who the market leaders will be”.

All were right…eventually. In March 2000 the tech-heavy NASDAQ index peaked, then fell by more than 80% over the following two and a half years. The trouble was that Messrs Dalio and Lynch were speaking in 1995, and Mr Marks in 1996. By 1999 Mr Soros’s short bets had lost his flagship hedge fund $700m and cost it billions more in withdrawals. Mr Buffett possibly felt the need to justify himself, also in 1999, since his investment vehicle had underperformed the NASDAQ by an average of 15 percentage points a year over the previous five. Between 1995 and March 2000, the index rose by nearly 1,100%.

Even for the very best investors, in other words, identifying a bubble is a good deal easier than judging when it will burst. Today there is no shortage of people worried that another is forming. The share prices of tech firms need only fall by a few percent—as they did in November—to send volatility leaping and make traders uneasy. Stocks related to artificial intelligence are the focus of their concerns; just look at Palantir, a data-analysis firm, with its bonkers valuation of over 200 times expected earnings for the coming year. But AI is not the only sector in nosebleed territory. Relative to underlying real earnings over the previous ten years, the S&P 500 index of big American firms has been priced higher only in 1999 and 2000. As a multiple of underlying sales, it is over 60% pricier than it was even at that boom’s peak.

So how would an investor know a crash was coming? High valuations are fairly good at predicting low long-run returns, but useless over the short run. Chart 1 shows how they have fared over the past few decades in forecasting share-price performance over ten years and over just one. Each dot represents a year between 1990 and 2024. The horizontal axes show the S&P 500’s valuation at the start of that year, measured by the cyclically adjusted price-earnings (CAPE) ratio popularised by Robert Shiller of Yale University. The vertical axes show the share-price index’s subsequent annualised return. With a ten-year horizon, the inverse relationship between starting valuations and returns is clear, and especially strong for high CAPE readings. Over a year, there is no correlation at all.

Investors might, therefore, have to turn to novel measures of market timing. Following Mr Marks’s lead, we looked for moments when every partygoer and cabbie was discussing stocks—or, more precisely, when Google searches for investing fads spiked. The logic is that a bubble is most likely to draw interest from lots of retail traders just as it reaches bursting-point. Chart 2 shows the results for a range of manias. They encompass cryptocurrencies (bitcoin and Dogecoin), baskets of once-trendy “thematic” stocks (cannabis, wearable tech and space) and the crazes, in 2021, for special-purpose acquisition companies (SPACs) and Cathie Wood’s “ARKK” investment fund.

Surges in Googling do a much better job than valuations at forecasting an imminent fall, as the third column of the chart shows. In each case the price of the stock, basket, fund or cryptocurrency dropped considerably over the 12 months following the peak in internet searches. Moreover, for the ARKK fund, bitcoin, GameStop and SPACs, prices spiked at almost exactly the same time as Googling did.

Naturally, such observations do not constitute a rigorous study. There will have been many instances of internet traffic concerning popular investments spiking with no subsequent fall in prices. In fact, searches for “AI stocks” hit their zenith in mid-August (see chart 3), and their prices continued to rise serenely for weeks.

It is nevertheless perturbing that Google searches for “AI stocks” have since fallen so dramatically, just as the stocks themselves are having a wobble. The share price of Nvidia, the world’s most valuable company and most important chipmaker, has fallen by 13% from its peak. The Philadelphia semiconductor index, which tracks firms in that industry around the world, dropped by 13% in the first three weeks of November, though it has since bounced back. The “AIQ” exchange-traded fund, a popular vehicle for investing in a basket of AI-related stocks, had a peak-to-trough fall of 12%. Since early October owners of bitcoin have suffered losses of more than 25%. Although not obviously linked to the AI theme, the cryptocurrency tends to appeal to similar groups of investors, and is closely tied to markets’ overall risk appetite.

That leads to another non-traditional measure. In the five years to March 2000, the NASDAQ suffered corrections of more than 10% on at least a dozen occasions, each time recovering and eventually rising nearly 12-fold. Even at the bottom of its subsequent plunge, the index was still twice as high as it had been at the start of 1995. Those who simply ignored both mania and crash, and held on throughout, were richly rewarded. The professionals who correctly called the bubble, meanwhile, often were not. Their experience was epitomised by Julian Robertson, another famed investor who over the two decades from 1980 handed his clients average returns of 25% a year, and in 1998 was overseeing $21bn. By March 30th 2000, withdrawals had forced him to close his fund, which had determinedly avoided the dotcom mania. As it turned out, the bubble had burst two days earlier.

Those trying to time the top of the present-day cycle should therefore look out for buzzkill types with big names going out of business. Such as, say, Michael Burry, who memorably bet against American mortgage-backed securities before they plummeted in value and set off the global financial crisis of 2007-09. This year Mr Burry has been busy shorting AI stocks, including those of Palantir and Nvidia. In late October, he wrote to investors to tell them he was closing his fund. ■