How big was it
Headlines called Thursday the worst day since April 2025, and we expected that date to be further back, as it is only 19 months ago. A points record says little, because the index is far higher than it was. Black Monday’s 20.47% fall in 1987 would take 1,574 points off the index at Wednesday’s close price.
Falls of this size are rare and they bunch together, with 13 in 2008, eight in 2020 and none from 2012 to 2019.
A 6% drop from a high is ordinary. The S&P 500 has fallen at least that far at some point in 66 of the 76 calendar years from 1950 to 2025, and it is still up about 7% this year.
Chip shares fell hardest
The iShares Semiconductor ETF fell 12.15% on Thursday, while the equal-weighted S&P 500, in which every company counts the same, fell 2.89%. Reports tied the selling to doubts about the cost of financing the AI build-out and possible limits on data centre construction after Democrats’ midterm wins.
The gap is wider over the past month. From 7 October to Thursday’s close, the chip fund lost 32% and the Nasdaq 100 lost 14.5%, while the equal-weighted S&P 500 rose 3%, the FTSE 100 rose 3.1% and India’s Nifty 50 rose 8.6%. The chip fund is still up 31% on the year.
Stock markets in South Korea and Japan, which have large weightings in chip and AI-related companies, were hit too. Both fell more than 6% in early trading on Friday, and Korean trading was halted for 20 minutes after the Kospi index dropped more than 8%.
If you hold a global fund priced at the London close, its Thursday valuation may not show all of the fall, because London closes four and a half hours before New York.
What happened after similar days
Treating falls of this size less than three months apart as one episode, there were 15 before Thursday. In 11 of them, the index ended the first day more than 10% below its high. In ten, no fall as large followed in the next three months. In the other five, including 2008 and 2020, more did.
A year after the first day, the S&P 500 was higher in ten of the 15 episodes. Picking any day since 1950 would have given a higher share, about 75%, so a bad day on its own has been no signal to buy. The lowest point during that year was typically about 8% below the first day’s close, and in five episodes it was more than 18% below. The table shows the last six.