Chip Stocks Fell 20%. Profit Forecasts Rose 15–25%. Reading the Sell-Off With Three Numbers

On November 5 (US time), the PHLX Semiconductor Index, which tracks a basket of chip-company stocks, fell 12.25% in a single day. Over that week, Micron fell 22.0% and SK hynix fell 20.2%. Compared with their highs in June, the chip index is down 36% and SK hynix is down 61%. In less than five months, SK hynix lost more than half its value.

There was no shortage of explanations. Investors were said to be selling AI stocks after the US midterm elections. A Chinese memory maker, CXMT, was reported to have a new technology. Some warned that memory chips could be in oversupply by around 2029. Over the weekend, Nvidia CEO Jensen Huang admitted to an extramarital affair in an email to staff, alongside reports that US intelligence suspects the other person of working for Chinese intelligence. Nvidia's board started an independent review and said it had no reason to believe company secrets were leaked.

This post does not try to judge which of those stories is right. Instead, it checks three things with numbers.

1. How scared are people right now?
2. While share prices fell, did forecasts of how much these companies will earn fall too?
3. When chip stocks fell this hard in the past, what happened next?

As of the November 6, 2026 close. “Top 16 of 100 days” means that since 1990 the fear index (VIX) was higher than this on only about 16 days in 100.

Five words to know first

• VIX (the "fear index"): A number for how much US stocks are expected to swing over the next month. The more nervous people are, the higher it goes. VIX3M is the same idea measured over the next three months.
• EPS (earnings per share): A company's yearly profit divided by its number of shares. Here we use analysts' forecasts of next year's EPS, since next year hasn't happened yet.
• P/E (price-to-earnings ratio): How many years of profit the share price equals. Share price ÷ EPS. For the same company, a lower P/E means the shares got cheaper compared with what it earns.
• Median: Line up all the numbers from smallest to largest and take the one in the middle. One very large or very small number can't pull it around much.
• Percentile: Where something ranks out of 100. "Lower on 84% of days" means that on 84 days out of 100, things were calmer than today.

1. How scared are people? About the top 16 days in 100

Line up every trading day since 1990, and 84.4% of them were calmer than November 6. Only about 16 days in 100 were scarier.

The third row stands out. Normally, "one-month fear" is smaller than "three-month fear," because the far future looks more uncertain. Right now the ratio is above 1. People are more afraid of the next month than of the next three. Since 2006, that has happened on about one day in ten.

Look at chip companies one by one, and it gets more extreme.

Bold names had a drop that shows up 3 times in 1,000 or less: very rare.

For Samsung, SK hynix and Micron, a one-month drop this big shows up only 1 to 3 times in 1,000 one-month periods since 2015.

2. Did profit forecasts fall? No, they went up

People being very scared doesn't mean they're wrong. If a company really is going to earn less, its stock should fall. So I checked how analysts' forecasts of next year's EPS changed over the same three months that share prices fell.

When the two bars point opposite ways, the shares got cheaper relative to profits (a lower P/E). Price as a multiple of next year’s profit: Nvidia 17.5→11.7 · Micron 5.3→3.3 · Samsung 3.4→2.4 · SK hynix 3.1→2.1. Broadcom’s forecast stayed flat.

"Next year" follows each company's fiscal year, so it differs a little (fiscal 2028 for Nvidia and Micron, calendar 2027 for the two Korean companies).

For six of the seven companies, the forecast of next year's profit went up while the stock went down. Nvidia's and Micron's forecasts rose by almost a quarter in three months. In the last month alone, far more analysts raised their forecasts than cut them.

That is why P/E ratios dropped so much. Three months ago, Nvidia's share price was 17.5 times next year's expected profit. Now it is 11.7 times. The profit forecast got bigger while the price got smaller.

The one exception is Broadcom. Its forecast barely moved, but its stock fell 31%.

But this table only shows one fact: forecasts went up and prices went down. It can't tell us which side will turn out right. A forecast is a guess, not a report card. And some analysts have already started cutting. On November 3, Korea Investment & Securities lowered its Samsung Electronics price target from 570,000 won to 440,000 won, saying memory prices may rise more slowly. On November 2, Melius Research cut its Nvidia target from $420 to $315.

In my earlier post on the chip outlook, I looked at 22 past cases where the chip index fell more than 20%. What decided the outcome was whether the drop came at the same time as the peak in company profits. The catch is that you can only see a profit peak after the forecasts start to turn down.

3. When chip stocks fell this hard before, what happened next?

Finally, I looked at the past. I picked days when two things were true at once: the stock was far below its highest price of the past year, and it had also fallen a lot in the past month. Each company swings by a different amount, so I set the cutoffs per company so that today's situation counts. To avoid counting the same fall many times, cases had to be at least three months (63 trading days) apart.

Then I always compared with "buying on any day." Over long periods, stocks tend to go up anyway. So "it went up after I bought on a crash" isn't enough. The real question is whether it went up more than usual.

All numbers are medians (the middle value when lined up). Bold = worse than buying on any day. Colors follow this site’s Korean convention: red = up, blue = down. The two panels use different scales. Case counts are for one year; SK hynix and Samsung have few cases.

One year later

Three months later

How much further it fell within six months of buying (buy-day price vs. the lowest price after)

Three things stand out.

1. One year later, buying after a crash beat buying on any day. All five medians are higher. But for Nvidia, the share of cases that rose was actually lower (68% vs. 75%).

2. Three months later, that wasn't always true. For Micron, SK hynix and Nvidia, the three-month median was lower than buying on any day. For SK hynix, four out of five cases were lower three months later.

3. Prices often kept falling after the buy. For Micron and Nvidia, in half the cases the stock fell more than 24% further within six months. The worst cases came when the early-2000s internet bubble burst and during the 2008 financial crisis. Point 1's "better after a year" only holds if you sat through those extra drops.

The number of cases is small, and that matters. SK hynix has only 4–5 cases, and Samsung only 5–6. Samsung's +81% comes from a few big rebounds in 2001–2002 and 2008. Even the chip index's 16 cases include 7 bunched in 2000–2002. They are less like 16 separate events and more like the same period counted several times.

What the numbers tell us, and what they don't

• What they tell us: Fear in the market is high by historical standards. Yet for six of seven companies, next year's profit forecasts went up, the opposite of the fear. After similar drops in the past, results one year later were better than usual.
• What they don't tell us: where the bottom is, whether prices will rise or fall over the next three months, and whether the analysts' forecasts will turn out right.

The fear index only tells you "people are very scared right now." Whether that fear is too much is decided by how much these companies actually earn. So the next numbers to watch are on the profit side.

What to check next

• Nvidia's earnings report (November 17, US time): Nvidia's own forecast for this quarter's revenue is $108 billion, plus or minus 2%. After the result and its outlook for next quarter, I'll check whether analysts keep raising next year's forecasts.
• Which way forecasts move: I'll recalculate the table in section 2 two weeks from now. The key question is whether any company sees more analysts cutting than raising.
• 1-month fear ÷ 3-month fear: whether it drops back below 1. That would be the first sign that short-term fear is easing.
• In early December, I'll also post the follow-up on 14 stop-loss exits that I promised in the fact check on post-crash recoveries.

Data sources. Prices and indexes were calculated directly from Yahoo Finance daily adjusted closes for ^SOX, ^VIX, ^VXN, ^VIX3M, NVDA, MU, 005930.KS, 000660.KS, TSM (ADR), AVGO and AMD, as of November 6, 2026. A percentile is the share of all trading days, from the start of each index's data to that date, with a lower value than that day. One month is 21 trading days; three, six and twelve months are 63, 126 and 252 trading days. Next-year EPS forecasts and the counts of raised and cut forecasts are Yahoo Finance consensus data (eps_trend, eps_revisions, retrieved November 9, 2026). The three-month price change uses the close on that date and the close 90 days earlier. P/E is that day's price ÷ that day's next-year EPS forecast. In section 3, the drop is measured from the highest close of the previous 252 trading days (about one year), not the all-time high. Using the all-time high would make entire long bear markets meet the condition. A case is the first day the condition is met; days that meet it again within 63 trading days count as the same case. Cases without enough time before the cutoff date (July and November 2026) are left out of that period's statistics. "How much further it fell" is the lowest close within 126 trading days, compared with the case-day close. "Bought on any day" is the N-day return on every day across each stock's full data history. Price-target changes come from The Economist Korea (Nov 3, Korean) and Yahoo Finance's rating-change records. The Nvidia CEO report is from CNN (Nov 8). Nvidia's revenue forecast is the figure the company gave with its previous quarterly results. To reproduce: scripts/factcheck_semis_sentiment.py.

How this was written. A large language model (LLM) was used for research, calculation and drafting, and figures were recalculated from the data above to check them. This is the English edition of a post first published in Korean on November 9. It is not investment advice.