When a stock price suddenly plunges, investors face a familiar dilemma. Is this a temporary swoon caused by a broader market panic, meaning the stock is likely to rebound? Or does the drop reflect a real problem inside the company, suggesting more pain ahead? The old Wall Street sayings capture the stakes: should you “buy the dip” or avoid “catching a falling knife”?
A new investigation published in the Journal of Behavioral Finance tackled this question using a database of 2.7 million news articles processed by artificial intelligence. The researchers found that the tone of company-specific news during a crash can help separate stocks that will recover from those that will keep falling.
The question behind the research
Flurin Jurt, Matthias W. Uhl, and Santiago Walliser of the University of Zurich wanted to know whether news sentiment data could predict which companies bounce back after extreme price drops. Their work builds on a long tradition in behavioral finance, which studies how emotions and psychology push markets away from rational pricing.
Earlier research had established some puzzling patterns. Stocks often overreact to sudden shocks when no major company news is driving the move, then reverse course. But when bad news genuinely concerns the firm itself, the decline tends to continue. The trick is telling the two situations apart in real time.
To understand the study, it helps to know what a “tail event” means. Daily stock returns usually cluster around an average, but occasionally a stock will swing far outside its normal range. These rare extreme moves live in the “tails” of the distribution of returns, and they matter enormously. As the authors note, an investor who missed just the ten best days of the S&P 500 between 2001 and 2020 would have seen their long-term performance cut in half. Many of those best days happened within two weeks of the worst days.
Building a sentiment thermometer
The team focused on companies in the S&P 100 index between 2003 and 2021. They pulled news sentiment scores from Refinitiv News Analytics, a system that uses natural language processing (a branch of AI that interprets human writing) to classify articles as positive, neutral, or negative based on their wording about specific firms.
Rather than looking at raw sentiment scores, the researchers created a relative measure. For each company on each day, they calculated how the tone of recent news about that firm compared to the tone of news about the broader market. To smooth out noisy daily fluctuations, they compared each stock’s 1-month average sentiment against its 12-month average, a technique borrowed from price momentum analysis.
They then identified negative tail events by building a rolling statistical model of each stock’s return distribution over the previous 252 trading days (roughly one year of trading). A tail event was any day when the stock dropped more than two standard deviations below its recent average return, a threshold that captures the bottom 2.5% of typical daily returns. They further narrowed the focus to days when the overall market was also down, isolating cases where a broader downturn might be dragging the stock along.
What the numbers showed
Using a statistical technique called fixed-effects regression, which controls for differences between individual companies and across time periods, the team analyzed 5,161 negative tail events during a training period from 2004 through 2014.
The pattern was clear. Companies that experienced a negative tail event while their relative news sentiment was positive tended to post higher returns over the following month compared with firms whose sentiment had turned negative. In plain terms, when a stock crashed but the news coverage about the company itself remained comparatively upbeat, the drop often reversed. When the company’s news tone was souring, the decline was more likely to continue.
Turning the finding into a strategy
The researchers translated their findings into a trading strategy. The rule: after a negative tail event that coincides with a down market, buy the stock only if its relative news sentiment is positive, and hold for 20 trading days. They added safeguards including an equal volatility-weighting scheme and a 10% cap on exposure to any single stock.
Between 2004 and 2014, the strategy outperformed the S&P 100 index, generating a Sharpe ratio of 0.27 and an information ratio of 0.36 (both are common measures of risk-adjusted returns).
But when the team tested the same approach on data from 2015 to 2021, the edge weakened. They suggest that markets have sped up. Increased news reporting, more investors reacting to headlines, and the rise of high-frequency trading firms may all be compressing the time between a news event and its reflection in prices.
In response, they propose a faster version of the signal, comparing 1-week sentiment against 1-month sentiment and holding positions for just five days.
Takeaways and caveats
For investors and analysts, the study offers a way to think about crashes. A broad-market sell-off that catches a fundamentally sound company may represent a reversal opportunity, while a price drop accompanied by deteriorating company-specific news coverage deserves more caution.
Several caveats deserve attention. The analysis covers only large-cap U.S. companies in the S&P 100, so the results may not transfer to smaller stocks or international markets. Transaction costs, liquidity, and tax consequences could erode real-world returns. And the weakening performance in recent years suggests that as markets evolve, yesterday’s profitable signal may not be tomorrow’s.
