This post was first published in Korean on November 7, 2026. Korean version
KOSPI closed at 5,059.69 today, November 6.
It fell 6.0% in a day and 11.0% over yesterday and today.
That is 44.5% below the June 22 peak of 9,114.55.
KOSDAQ closed at 644.27, down 47.5% from its peak.
SK hynix is down 59.7% from its peak, and Samsung Electronics is down 46.6%.
The drop started in the US the night before.
The Philadelphia Semiconductor Index fell 12.25% in a single day, its biggest drop since March 2020.
Should I cut my losses here, hold, or buy more?
I worked through it using two sources I trust, 86th Street and KB Securities’ Egzion, plus backtests I ran with Claude.
The short answer
I won’t sell. I’ll hold, and if the market falls further, I’ll buy in three steps.
Here is why, one line each.
Today’s price already assumes chip earnings shrink to a third or less next year.
It can still fall further, and history says by about 20%.
The upside if things go right (+52%) is bigger than the downside if they go wrong (-30%).
1) Why did it fall this hard?
Three things hit at once.
First, political fear.
The Democrats won both the House and the Senate in the US midterms.
Fear spread that “if the Democrats block data centers, chip demand dries up.”
Second, crowded money rushed out at the same time.
There were reports of heavy selling in leveraged ETFs and by several hedge funds.
Foreign investors sold ₩3.4 trillion of KOSPI stocks on November 5 alone.
Third, Korea has its own plumbing problem.
KOSPI swings harder than other countries’ indices that hold nothing but chip stocks.
The reason is flows.
On the way up, single-stock leveraged products poured fuel on the rally.
On the way down, margin-lending rules poured cold water on it.
Margin debt has fallen from about ₩39 trillion a few months ago to ₩33–34 trillion.
The fear of “I’d better sell before someone else’s margin call forces a sale” makes the fall worse.
That is not a rational decline.
But an irrational decline is still a decline.
2) Can it fall further?
Yes, it can.
In 36 years, KOSPI has fallen 44.5% from a peak three times.
November 1997 (Asian financial crisis): it fell another 36%
September 2000 (dot-com bust): another 19%
October 2008 (global financial crisis): another 17%
All three times, it kept falling.
Applying the middle value, -19%, to today puts KOSPI at about 4,100.
The shape of the bottom points the same way.
In an October 22 post, Egzion (KB Securities) noted that after KOSPI fell more than 25%, the only quick V-shaped rebound was roughly 2020.
Most of the rest were W-shaped, and the second bottom was often lower than the first.
That is exactly the shape we have now.
First bottom, October 26: 5,469.47 (-40.0% from the peak)
Rebound, October 28: 6,012.44 (+9.9%)
Second bottom, November 6: 5,059.69 (7.5% below the first bottom)
Counting days when the market fell 10% or more in two days gives a similar picture.
Before now, there were eight times when KOSPI fell 10% in two days while already 30% or more below its peak.
Further decline within a year: middle value -21%
Share higher a year later: half (4 of 8)
Return a year later: middle value +17%
The best case was March 2020: that day was the bottom, and a year later the index was up 108%.
The worst was November 1997: it fell another 44%, and a year later it was down 37%.
3) How cheap is the market now?
I used two yardsticks.
The first is earnings.
It asks “how many times next year’s expected earnings is the share price?” and is called the forward P/E.
KOSPI is at about 4x today.
In other words, four years of next year’s earnings would buy the whole index.
I recalculated BNK Securities’ October 30 figure of 4.6x using today’s close (I couldn’t compute the forward P/E directly).
Past bear-market bottoms looked like this.
2008 financial crisis: 6.1x
2020 COVID: 7.8x
2022 rate shock: 8.5–9x
August 2024 yen carry unwind: 8.0x
On this yardstick, the market is cheaper than at the financial-crisis bottom.
But there is a catch.
The number only means something if “next year’s expected earnings” turn out to be right.
2018 shows why.
In October that year, the forward P/E was 8.7x and looked cheap.
By the following January, stocks had fallen further, yet the forward P/E had risen to 10.6x.
Chip earnings forecasts were being cut faster than share prices were falling.
The second yardstick is assets.
It asks “how many times the company’s net assets is the share price?” and is called the forward P/B.
When earnings forecasts are shaky, this yardstick is more reliable.
KOSPI trades at a P/B of 1.12x today.
Compared with past bottoms:
1.0x (book value): 4,512 (-11%)
0.9x (August 2024 bottom): 4,061 (-20%)
0.83x (September 2022 bottom): 3,745 (-26%)
0.78x (October 2008 bottom): 3,520 (-30%)
Put in numbers, “too cheap to fall much further” means a P/B of 0.8–0.9x, or KOSPI at 3,600–4,100.
Of course, companies earn far more now than they used to.
Return on equity is high.
A company that earns more deserves to trade above book value.
But what the market doubts is exactly whether those earnings will last.
4) How much bad news is already priced in?
I split KOSPI into the two chip giants (Samsung Electronics and SK hynix) and everything else.
The two chipmakers are about 45% of KOSPI’s market value.
But they account for about 76% of next year’s expected earnings.
KOSPI’s next year effectively depends on chip earnings.
Everything else trades at 10.6x next year’s earnings, around its historical average.
Neither cheap nor expensive.
So what about the chip stocks on their own?
If the rest of the market sits at an average valuation, today’s KOSPI price assumes chip earnings fall 66–80% next year.
Here is how much earnings actually fell in past memory downturns.
2018 to 2019: Samsung Electronics -53%, SK hynix -87%
2022 to 2023: Samsung Electronics -85%, SK hynix swung to a loss
In other words, “a memory downturn next year” is already almost fully in the price.
To fall further from here, the downturn would have to arrive and investors would also have to pay less for each won of chip earnings.
5) When was it most like this before?
The closest match by size of fall is 2002–2003.
From 937.61 in April 2002 to 515.24 in March 2003, KOSPI fell 45.0%.
A credit card crisis, the SK Global accounting scandal, the Iraq war and the North Korean nuclear standoff all hit at once.
86th Street drew the same comparison in its November 5 post.
Global markets have not even fallen 30%, yet only Korea is plunging, which it says recalls the early 2000s.
Back then, Korea was singled out by the credit card crisis (2002) and the China GDP shock (2004).
Afterwards, though, it reaped the rewards of China’s rapid growth and the commodity supercycle.
I checked the numbers.
On March 5, 2003, KOSPI hit -40% from its peak.
It fell another 8% before bottoming.
A year later it was up 52%.
The closest match by structure is 2018.
Stocks turned down first, while memory prices and earnings were still at their peak.
Here, the DRAM contract price is what Samsung Electronics and SK hynix charge big customers such as PC and server makers.
Most of a memory company’s revenue is set at this price.
Market research firm DRAMeXchange publishes it at the end of each month.
The DRAM contract price hit a record $8.19 in April 2018 and stayed there through September.
It first fell in October ($7.31).
Stocks turned before that.
Samsung Electronics: peaked in November 2017
SK hynix: peaked in May 2018
At the end of September 2018, with the price still at its record, Samsung Electronics was already 19% below its peak and SK hynix 23% below.
Stocks move on where prices will be a year from now, not where they are today.
Today, the DRAM contract price is still forecast to rise 10–15% in the fourth quarter.
That is why “prices are still rising” is not enough to feel safe.
Of course, there is a worse comparison too.
Since January this year, Egzion has compared SK hynix with HMM, the container shipping line, during the pandemic.
Both, it argues, are “over-ordering cycles,” where demand floods in and orders overflow.
HMM’s share price peaked at ₩50,600 on May 27, 2021.
Container freight rates (SCFI) hit their record of 5,109.6 seven months later, on January 7, 2022.
That day, HMM traded at ₩26,400, already 48% below its peak.
It bottomed at ₩13,950 in October 2023, down 72%.
SK hynix is now 59.7% below its peak.
If it follows HMM’s path exactly, chip stocks could fall further.
But HMM’s supply (ships) flooded in two to three years later, while most memory volume for 2027 is already locked into long-term contracts.
The closest match by trigger is August 2024.
Crowded money rushed out with no earnings problem behind it.
The market fell another 6%, and a year later it was up 29%.
Putting the three together:
If this is only a flows problem, the fall stops around 4,500.
If earnings forecasts start getting cut as in 2018, the floor is 4,000–4,200.
If the market leaders collapse as in 2000, it is 3,500.
6) If it rises, which scenarios get it there?
I checked six scenarios against current evidence, one by one.
The probabilities are my own judgment of how likely each is to play out within 12 months.
1. Rate cuts (oil has to come down first)
Probability of a cut: 20%.
The Fed raised rates in September and paused in October.
The odds of another hike in December are 73% (CME FedWatch, November 6).
In its November 2 weekly, 86th Street split the conditions for a December pause into two.
Lower inflation or falling oil would make it a “good pause.”
Weak jobs or a weak economy would make it a “bad pause.”
Oil is falling.
WTI is about 20% below its April peak.
JPMorgan estimates the oil market was oversupplied by 1.4 million barrels a day in October (cited in 86th Street’s November 4 post).
Egzion looks the same way.
In an October 22 post, it named two things weighing on the market: memory worries and inflation worries.
Of the two, it saw easing inflation as the key to a rebound.
In a November 2 post, it wrote that right now “resolving the Iran issue” matters more than the election or AI spending.
The key is refined products such as diesel, more than crude oil.
According to Egzion, KB Securities expects refined product prices to stabilize from as early as year-end, as long as Iran causes no new trouble.
So the realistic good news is not a “cut” but “no more hikes because inflation is under control.”
2. AI investment (capex) continues
Probability: 70%.
Big Tech has not cut next year’s investment plans.
Alphabet and Amazon said they will spend “significantly more” in 2027.
86th Street, though, pointed out something interesting in its October 30 post.
This earnings season, capex estimates were almost unchanged before and after the results.
Until now, estimates had jumped every earnings season.
86th Street sees this as a good thing.
If spending rises too much, people worry about “where the money comes from”; if it falls too much, they worry the cycle is ending.
3. Data center rules turn out milder than feared
Probability: 70%.
The Democrats won, but not by a landslide.
As 86th Street noted several times, they fall short of 60 Senate seats (to override a filibuster) and two-thirds of both chambers (to override a presidential veto).
President Trump has already said he will be “Mr. Veto.”
The idea that Democrats hate data centers is not quite right either.
Pennsylvania, one of the states with the most data centers planned, already has a Democratic governor (Shapiro).
Companies adapt.
Google said it will build a data center in Wyoming designed to solve its power, water and carbon problems.
Nvidia talked about using old telecom company buildings.
In 86th Street’s words, reality is likely to settle “somewhere between 0 and 1.”
I agree.
Still, with the new governors of Texas and Ohio having campaigned on moratoriums, the pace could slow.
4. AI companies keep growing earnings
Probability: 75%.
Google Cloud revenue rose 103% year on year in the third quarter, and Microsoft Azure 48%.
US labor productivity grew 3.6% in the third quarter.
86th Street noted productivity growth over the last four quarters climbing from 1.2% to 1.4%, 2.1% and then 3.6%, and said the AI effect is starting to show up in the numbers.
The worry is money.
Free cash flow has turned negative at several Big Tech companies.
According to forecasts cited by 86th Street, the first half of 2027 is the worst point, with a recovery after that.
5. Memory prices keep rising
Through the first half of 2027: 60%. Through the end of 2027: 35%.
TrendForce expects the DRAM contract price to rise 10–15% in the fourth quarter, and NAND 15–20%.
Micron said more than 75% of its 2027 volume is already sold under long-term contracts.
The US economy has not turned down yet either.
In a November 2 post, Egzion pointed out that the ISM manufacturing index signals SK hynix’s operating margin about four months in advance.
October’s ISM manufacturing index was 55.5, above the 54.5 expected.
Past peaks have generally been around 60.
There are opposing signals too.
Qualcomm lowered its profit outlook because memory costs have risen so much.
That means expensive memory is starting to eat into demand.
6. Chipmakers keep investing, which supports equipment, materials and substrate makers
Probability: 75%.
Samsung Electronics plans ₩84.4 trillion of capital spending this year and has signaled a “significant increase” next year.
SK hynix has also announced a major increase next year.
Chip equipment makers should do well next year too.
7) Which scenario matters most?
For KOSPI, scenario 5, memory prices, matters most.
Chips account for 76% of next year’s earnings, and memory prices set those earnings.
Next is scenario 2, AI investment.
It is the money that buys the memory.
Scenario 3, regulation, was the trigger for this drop.
But it reaches the index only later, through scenario 2.
The new governors take office in January, and actual permitting decisions come after that.
Scenario 1 depends on the time frame.
Where KOSPI stands a year from now will be decided by memory earnings.
But the direction over the next quarter or two will likely be set by oil and inflation.
As Egzion says, of the two things weighing on the market, inflation is the one that could ease soonest.
Two variables could matter even more than the six scenarios above.
First, memory supply.
Samsung Electronics and SK hynix are raising capital spending sharply next year, and China’s CXMT is catching up.
Memory stocks move on supply two years out, not on today’s prices.
That is why stocks turned before prices in 2018.
It also means scenario 6 is good news for equipment and materials makers but could be bad news for scenario 5, memory prices.
Second, flows.
In an October 22 post, 86th Street likened today’s market, where everyone is bracing for tail risk, to watching a movie while pressed up against the cinema exit.
Hedge funds have already cut their AI net positions sharply, and trading in Big Tech CDS (default insurance) more than quadrupled in a month.
If everyone has already bought insurance, there are fewer people left to sell.
There is support at home too.
86th Street estimates the National Pension Service’s domestic equity weight fell from 25.8% at the end of August to about 22% now.
If it uses about 25% as its benchmark, as Japan’s public pension fund does, it has room to buy more.
Over a few weeks, these flows matter more than fundamentals.
8) Combining the scenarios: how much upside?
Good case (25% probability)
AI investment holds + rules ease + memory prices hold + rate hikes pause
KOSPI about 7,700 (+52%)
Base case (40% probability)
AI investment holds + regulation limited to cost burdens + memory prices flat
KOSPI about 5,700 (+13%)
Bad case (25% probability)
State moratoriums + slower AI investment + falling memory prices (2019-style)
KOSPI about 4,200 (-18%)
Worst case (10% probability)
Memory bust + economic slowdown (2001- and 2023-style)
KOSPI about 3,500 (-30%)
Weighting these four possible futures by their probabilities gives an expected one-year return of about +11%.
Even in the good case, the gain does not come from chip earnings growing further.
Confirmation that “chip earnings are not collapsing” is enough for chip stocks to return to normal valuations.
The first checks are Samsung Electronics’ preliminary results in January and Big Tech’s fourth-quarter investment plans.
So what am I doing?
First, I’m not selling.
Selling now would lock in “next year’s memory downturn” at today’s price.
Even in the worst case, the remaining downside is around 30%.
Second, I’m checking position size.
The test is whether I could hold through another 30% drop.
If not, I trim.
That is sizing, not stop-loss selling.
Third, I’ll buy more in three price steps.
4,500: P/B 1.0x
4,100: P/B 0.9x, the middle value of past further declines
3,600: P/B 0.8x, the 2008 and 2022 bottoms
Fourth, if these signals come first, I won’t wait for the price.
(1) Confirmation that inflation is easing, through an Iran deal or lower diesel prices (starting with October CPI, released November 10)
(2) A pause at the December FOMC
(3) Big Tech holding or raising investment plans in January
(4) A forecast of rising DRAM contract prices for the first quarter of next year
Egzion also said in a November 3 post that once the market gets past November’s turmoil (earnings shocks, funding worries, the Anthropic IPO), opportunities will return from year-end as the economy expands.
That is why I’m buying gradually, not in a rush.
Fifth, I’ll stop buying if these signals appear.
(1) Any Big Tech company officially cuts its 2027 investment plan
(2) The DRAM contract price forecast turns negative
(3) Texas or Ohio actually puts a moratorium into effect
(4) The ISM manufacturing index falls below 50 (a sign chip margins will turn down four months later)
Of course, as in 2018, the cuts to earnings forecasts may only be starting.
In that case, I have to allow for a drop to 4,100.
Egzion also warned of a signal in an October 19 post.
It is the combination of rising expectations of rate hikes on inflation worries, while the 10-year Treasury yield actually falls.
It means the market is starting to fear that “tightening will eventually break the economy,” and it appeared in the first quarter of 2000, just before the dot-com bubble burst.
The 10-year yield has come down from 5.3% to 4.84%.
Still, Egzion sees no clear sign yet that the economic cycle has turned, and unlike 2018, today’s price has already absorbed much of those cuts.
Sources
86th Street (Naver premium content) posts cited:
Concerns about Democratic policy (November 5)
Midterm event over (November 4)
Midterm worries (November 3)
A sudden thaw? (November 2)
86th Street Weekly, first week of November
Worry after worry (October 30)
86th Street Weekly, fourth week of October
Fears of extreme scenarios (October 22)
A bond market hunting for scapegoats (October 16)
PPI beats expectations (October 15)
Egzion (KB Securities, Naver blog) posts cited:
ISM manufacturing: why only orders stalled (November 3)
How our view on chip stocks has evolved (November 2)
What to check in tonight’s ISM manufacturing release (November 2)
What a W-shaped bottom needs (October 22)
What “rising inflation plus falling long rates” warns of (October 19)
DRAM contract price: DRAMeXchange monthly releases, as reported by Yonhap, Edaily, iNews24, News1, Electronic Times and others. SCFI: Shanghai Shipping Exchange, as reported by the Korea Economic Daily, Edaily, Money Today, Financial News and others.
Market data: Naver Finance daily prices for KOSPI, KOSDAQ, Samsung Electronics, SK hynix and HMM.
KOSPI forward P/E and P/B: Kuki News (BNK Securities, 2026-11-01). Their 4.6x and 1.27x as of the October 30 close of 5,730.7, recalculated at today’s close.
Past forward P/E and P/B bottoms: Yonhap Infomax (2008), Korea Economic Daily (Daishin Securities) (2022), and other broker reports as cited.
Causes of the sell-off: Forbes (2026-11-05), Hankook Ilbo (2026-11-06).
Rates: PANews (CME FedWatch 72.8%).
Election: CNN, CBS.
Memory prices: TrendForce (2026-09-30).
