360
Productivity & Workflow355
Software Development249
Automation & Workflow238
Marketing & Growth207
AI Infrastructure & MLOps197
Photography & Imaging170
Writing & Content Creation195
Design & Creative173
Data & Analytics165
Customer Support157
Education & Learning145
Voice & Speech130
Sales & Outreach134
Operations & Admin119
Researchers analyzed 170 years of The Economist using AI and found bubble warnings appear every month, often overstating the chance of a crash.
In short: Researchers used AI to read 170 years of financial articles and found that bubble warnings are constant, but strong spikes often line up with real market bubbles.
A new research paper looks at how often the media warns about market “bubbles,” meaning prices that may be unrealistically high and could fall fast. The authors studied 170 years of articles, using back issues of The Economist.
They found that bubble worries show up all the time. On average, articles implied about a 20 percent chance of a major market crash within six months, which the researchers say is around four times higher than what actually happened historically. They also found the implied chance of a crash never dropped below 12 percent, with bubble warnings appearing in every month of the 170-year period.
To go beyond simple counting, the researchers used LLMs, which are large language models (AI systems trained to read and write text, like an autocomplete on steroids). The AI rated each article for things like tone and how “bubble-like” it sounded. The team then combined those ratings into scores that aimed to estimate when markets were in a bubble, based on 22 major historical bubbles across 10 countries.
The study says the strongest signals matched well-known moments like the dotcom bubble around 2000 and the lead-up to the 2008 financial crisis. When the scores rose above certain levels, they often lined up with periods that later ended badly.
The researchers also extended their method to recent coverage and found a high level of bubble-like language around today’s AI boom. A key question is whether the AI is truly spotting warning signs, or whether it is partly “remembering” history from its training data, like a student who has already seen the answers.
Source: Financial Times