The performance of the US market over recent decades has been defined by long periods of expansion punctuated by rapid, but relatively rare, downturns.
Even during moments of extreme volatility, recovery has typically followed quickly. The behaviour of the S&P 500 during the onset of the COVID-19 pandemic offers a clear example.
LSEG Data & Analytics recently delved into how to predict market downturns with news.
After falling by around 30% in the early stages of the pandemic, the index began climbing again within weeks and had regained its lost value in six months. By the end of that year, it was more than 10% higher than its pre-pandemic level. This raises an important question for analysts and investors alike: can news sentiment be used to anticipate these sharp shifts in market direction?
Financial researchers have explored whether news cycles display patterns that could help signal market stress before it appears in asset prices. Earlier work from LSEG found that its Machine Readable News and News Analytics services can highlight stock-specific developments, often identifying surprise events that emerge with little build-up.
The most prominent examples tend to relate to high-profile companies such as Tesla and Amazon, where a constant flow of stories can create sharp, sudden sentiment changes. But when the focus shifts away from individual equities to the wider US market, the news environment looks very different.
Rather than being sparse or unpredictable, aggregated news sentiment across the US market is exceptionally dense. Tens of thousands of articles are published on any given trading day, LSEG explained. The challenge becomes less about identifying rare signals and more about dealing with strong, persistent autocorrelation. Analysis shows that the daily sentiment score for US-relevant articles often carries over from one day to the next, even across longer time lags. This persistence can make it difficult to use sentiment alone as an indicator of rapid shifts in market conditions.
To address this, researchers examined the change in daily aggregate sentiment rather than the sentiment value itself. This adjustment creates a more responsive indicator with far less autocorrelation. Except for a noticeable negative correlation on the first lag—a sign of mean reversion—the revised measure behaves much more dynamically, it said. News coverage in stable conditions tends to revert around an average level, so sudden swings in sentiment followed by a snap-back the next day form part of a broader pattern.
To smooth this behaviour and create a more reliable signal, the daily changes can be averaged over a short look-back period. In this case, a simple five-day moving window was applied. Another approach might have been to only flag extreme deviations, such as movements two standard deviations below the average daily change.
For more insights into predicting market downturns, read the story here.
Read the daily FinTech news
Copyright © 2025 FinTech Global
