💰 Money

Why AI counted 170 years of bubble warnings

2 min read Tiny Why Newsroom · By Curio, Martian correspondent

Words
bubble

A market state where prices rise far beyond what the evidence supports.

indicator

A number built to show a change or warning.

S&P 500

A US stock index tracking 500 large companies.

What happened

Researchers at Cambridge University, a UK university, and London Business School studied financial news with artificial intelligence, or AI. Their working paper was dated November 15, 2026. A MarketWatch report said one method gave September 2026 a bubble probability as high as 97%.

That headline needs care. The number described how much the market resembled a bubble. It did not mean a 97% chance of an immediate crash.

Why count warnings at all?

Financial news often warns that markets are too hot. Such warnings can be useful, but they can also create many false alarms. The researchers compared the warning level with what actually happened later.

News articles implied an average 20% chance of a major market collapse within six months. The historical frequency was only 5.7%. In other words, the news sounded much more worried than history justified, on average.

Yet the warnings were not empty noise. They became stronger before several important bubbles. The examples included the dot-com bubble, the US savings-and-loan crisis, and the housing boom before the global financial crisis.

How the AI study worked

The researchers used 795,964 articles from The Economist, a British economics magazine. The archive covered about 170 years, from 1843 through 2013. They compared the articles with 22 recorded bubble episodes across 10 countries.

The AI examined financial sentiment, the strength of speculative stories, and the stage of a possible market cycle. The researchers turned those clues into 14 monthly indicators. Each indicator was compared with its own earlier 20-year history. This helped them search for unusual changes, rather than simply count scary words.

What the 97% signal means

One method assigned up to a 97% probability to a bubble in September 2026. Other indicators gave about 40%. That gap matters. It shows that the result depends on the model and the way the signals are combined.

The market had also weakened by the time of the report. The S&P 500 was down 6% from its October 6 record close. The Nasdaq Composite was down 12%. Those falls add context, but they do not prove that a bubble exists.

What the evidence supports

The study supports a careful conclusion. Financial media often exaggerates near-term crash risk. However, sharp changes in media warnings can still contain information about growing market excess.

The lesson is not to trust every warning. It is to watch whether several signals strengthen together, and whether later events confirm them.

What remains unknown

The September 2026 reading has not yet received a final real-world verdict. A bubble can be identified only more clearly after its outcome unfolds.

The study also has important limits. It relies mainly on one publication. The model may have indirect access to information about how historical episodes ended. The researchers acknowledge that this creates a possible look-ahead problem. The paper is also a working paper, not a final judgment.

The next useful test is simple: do other news sources and other indicators show the same pattern? The market’s later performance will matter more than the headline number.

💰 Money

Does a 97% bubble signal predict a crash?

📰 Full story: Why AI counted 170 years of bubble warnings

AI read old financial news for signs of an overheated market.

1 min read Tiny Why Newsroom · By Curio, Martian correspondent

Words
bubble

A market that may be too excited and overpriced.

working paper

A research report that is not a final judgment.

S&P 500

A US index that tracks 500 large companies.

💡 The gist

  • AI read old business news for signs of overheated markets.
  • Warnings often predicted more crashes than actually happened.
  • A 97% bubble signal did not mean a 97% crash chance.

What researchers did

Cambridge University is a UK university. London Business School is a UK business school. Their researchers wrote a working paper. The paper is not a final answer.

They used AI to study 795,964 articles. The articles came from The Economist, a British economics magazine. The archive covered about 170 years. It ran from 1843 through 2013.

Researchers compared the articles with 22 bubble episodes. Those episodes came from 10 countries. They measured news tone and excited speculation. They also studied the market’s possible cycle stage. Then they made 14 monthly indicators.

Why warnings can mislead

News articles suggested a 20% collapse chance within six months. Actual collapses happened 5.7% of the time. So news often sounded more frightened than history supported.

But warnings sometimes rose before real bubbles. They rose before the dot-com bubble. They also rose before other major financial crises. The change in warning levels mattered more than the total number.

What 97% means

For September 2026, one method gave a 97% bubble likelihood. Another method gave 40%. These numbers were different model readings. They were not guarantees.

The 97% described a bubble-like market state. It did not predict a 97% crash chance. The S&P 500 had fallen 6% from its October high. The Nasdaq Composite had fallen 12%. Those falls added context, but proved nothing alone.

What comes next

The study used one publication’s archive. The model may also know how old events ended. That can create a look-ahead problem. The paper is still a working paper.

Readers should watch other news sources and indicators. They should also watch what markets do later. That evidence will test September’s warning.

💰 Money

AI looked for a too-hot market

📰 Full story: Why AI counted 170 years of bubble warnings

AI is a computer helper that reads many words.

1 min read Tiny Why Newsroom · By Curio, Martian correspondent

Words
AI

A computer helper that reads many words.

bubble

A price that has grown too high.

stock

A tiny piece of a company.

What happened

Cambridge University is a school in Britain. London Business School is another school in Britain. Their researchers studied old money news.

They used AI to read 795,964 articles. The articles covered 170 years. They came from The Economist, a money magazine.

Researchers compared the news with 22 old bubbles. Those records came from 10 countries. A bubble is a price that grows too high. A stock is a tiny piece of a company.

The warning

The news guessed a 20% big fall within six months. Big falls happened 5.7% of the time. So the warnings were often too loud.

But warnings sometimes came before real bubbles. That made some warnings useful.

The 97% number

For September 2026, one test gave a 97% bubble chance. Another test gave 40%. The number described a bubble-like market. It did not promise a 97% crash chance.

Nobody knows the final answer yet. People must watch what happens next.

Sources