Today's wind was stronger than yesterday's. Yesterday's decline was −10.8% by the close. Today, the wind bent the forest as far as −12.6% intraday — erasing a strong opening in the forest. For the first time, circuit breakers fired on two consecutive days. From the sky, only our forest shook. Tokyo −1.5%; Hong Kong +2.0%. Strong winds do not just topple trees. They reveal which trees had no roots.
01
Something Unprecedented
7/29 (Wed) close · Circuit breakers in both markets for 2 consecutive trading days — a first for Korean stocks
KOSPI
5,663.24
-5.98%
Open 6,089 → Intraday low 5,263(−12.6%)
KOSDAQ
662.68
-6.12%
Intraday low 630.99 · New 52-week low
SK hynix
1,401,000
-9.6%
Q2 revenue of KRW 79tn could not prevent a steep fall
Trading value
49.2tn
KOSDAQ KRW 6.5tn
Fear drove a surge in volume
A day that opened higher and ended with circuit breakers
Opened +1.1% → KOSDAQ circuit breaker at 12:19 → KOSPI circuit breaker at 12:38 → low −12.6% → partial recovery to a −6.0% close. Note: Minute bars are unavailable, so dots show only actual polling times while writing (not a continuous chart).
Circuit breakers — rules and records
Stage 1 activates when the index stays down −8% for 1 minute, halting all stocks for 20 minutes (1 time per day). Today they fired at 12:19 on KOSDAQ · 12:38 on KOSPI, marking the first simultaneous activations in both markets on two consecutive days. For KOSPI, this was the year's 9th.
02
Yet Only Our Forest Shook
The same news, the same time zone — today's changes across Asian markets
Asia today (closing prices)
Only Korea lost around −6%; Greater China rose — exactly the same order for a second day
Yesterday (7/28), the order was the same: Korea −10.8% · Taiwan −4.7% · Japan −4.0% · Hong Kong +0.4%.
the order of semiconductor concentration exactly.
Half the decline is "Korea-specific"
Excess declines are estimated by regression against an Asian semiconductor basket (Taiwan 40% · SOX 35% · Nikkei 25%). β is shown for both the latest 60-day and 120-day windows.
20-day cumulative excess decline: −12.6 to −13.3pp — not just today, but throughout the month, Korea fell further than Asia. Verdict: structural discount (both windows agree).
Yet we were also the ones who rose the most
2026 year-to-date (left) and from the year's high (right) · Today's closing prices
This year, KOSPI rose from 4,310 to 9,115: +111.5% in 6 months. Over the same period, the Nikkei gained +18.5%, while Hong Kong and China were negative. The present decline is the price of that rise — semiconductor concentration doubled the index on the way up and produces −37.9% on the way down. Two sides of the same mechanism.
Markets that rose less are falling less
Year to date, negative Hong Kong (−2.0%) and China (−2.5%) rose today as well. By contrast, SOX (+49.8%), Taiwan (+36.4%), and KOSPI (+31.4%), which rose the most, are retracing the most. "Only we are shaking" is also the reverse of "only we rode the wave". The difference is that KOSPI's decline from its high (−37.9%) is the deepest in Asia, while KOSDAQ (−46.0%) fell with it despite never having risen (−29.9% year to date).
1
Japan Had a Buffer
The Nikkei also holds semiconductors, but domestic-demand and content stocks such as Sony and Bandai Namco supported it. Yesterday, the two returned +2.4%·+2.3%, respectively. Today, Tokyo's decline was limited to −1.5%.
2
We Did Not
On KOSPI, 91 stocks rose and 805 stocks fell. On KOSDAQ, it was 141 versus 1,554. That is the difference between a market with somewhere to hide and one without.
03
Answer ① — A Forest Hanging From One Tree
Where trading value was concentrated
Semiconductors' share of trading value
60.8%
7/23: 45.6% → today
Samsung + SK hynix account for this share of KOSPI trading value: 6 tenths
Samsung Electronics
208,500
-5.2%
From the 2026 high: -42.5%
SK hynix
1,401,000
-9.6%
From the 2026 high: -52.0%
KOSPI advancers / decliners
91 / 805
No refuge
KOSDAQ 141 / 1,554
Major stocks — today, and from their 2026 highs
The damage to individual stocks is deeper than to the index
04
Answer ② — Earnings Did Not Break. The Narrative Did.
SK hynix reported quarterly revenue of KRW 79tn, yet fell −9.6%
Anatomy of a miss — why sell a record result?
Record earnings and a miss against consensus estimates can both be true
Operating profit of 60.5tn — a record, but against consensus estimates of 64.2tn , it fell 5.7% short of consensus. The miss was not "the end of growth," but "the end of acceleration" — DRAM price growth slowed from Q1's 60~70% to Q2's 30%. What the market sold was not earnings, but the second derivative. The company pushed back directly on its call: "Efficient models lower entry barriers and increase usage, which in turn increases memory demand" (translated). It pointed to long-term contracts with around 10 customers, the start of HBM4 mass production, and capex raised to the high KRW 40tn range. The company defended the narrative, but prices did not listen that day.
Physical indicators are not all pointing one way
Product prices are rising, while AI services' profitability is falling — we must separate the two
Left (rising) — DRAM spot prices rose through 7/28 for 8 consecutive trading days (DDR4 16Gb 81.2 → 85.0, +4.7%). Today: flat (−0.03%). Memory export prices were YoY +168.5%, with the internal verdict "accelerating — no sign of a peak" (June data, with a 3–4-week lag). Right (falling) — AI inference margins fell by −22% over ten days. GPU rental rates rose while token selling prices stayed flat.
View the table · Why use daily-average inference margins?
Date
H100 rental ($/h)
Inference margin ($/Mtok)
7/20–22 average
2.58
0.429
7/25
3.00
0.387
7/27
3.07
0.379
7/28
3.13
0.372
7/29
3.33
0.350
This indicator is collected 2 times a day, yet can swing by ±50% in half a day (on 7/29 alone, $2.668 → $4.000), because it takes spot GPU rental rates directly. We therefore judge direction from daily averages, not a single observation. The coefficient of variation across all 60 observations is 16.8%.
AI demand through electricity — overnight baseload YoY
Data centers run 24 hours a day, making minimum overnight demand a proxy for actual demand
Only Texas (ERCOT) jumps to +6.7% — where data centers are concentrated. Japan is actually negative. AI demand measured through electricity has not yet turned down. But we do not know how much electricity demand is needed to sustain today's market. Electricity shows only what already-built data centers consume now. It does not tell us how much more it must grow to justify current share-price multiples. "Growing" and "enough" are different questions.
So the "peak" argument is only half right
Memory product prices and electricity demand still point up. Meanwhile, AI services' unit profitability has fallen for ten days. Today's data do not say "AI is over," but "the economics of making money from AI are tightening". The market did not distinguish the two and marked down semiconductor multiples as a whole.
The trigger on both days was the outlook
Yesterday it was the fallout from CXMT's Monday listing (first-day close +466%, KRW 8.6tn raised → Micron fell sharply, US congressional investigation opened). Today it was Wall Street's "AI semiconductor peak" argument. Both concern future supply and demand. SK hynix's KRW 79tn quarterly revenue is an accomplished fact. What the market reversed was the multiple attached to those earnings.
05
Answer ③ — Debt-Driven Acceleration
Retail investors capitulated; institutions took the other side
Sellers — retail investors (KOSPI)
−KRW 1.9701tn
From yesterday's +KRW 4.3152tn in net buying to heavy selling in a day — classic capitulation
Buyers — institutions (KOSPI)
+KRW 3.1769tn
But of that, financial investment firms accounted for KRW 1.8563tn (58%) — program trading, arbitrage, and ETF liquidity provision
Inside institutions' KRW 3.2tn — directional or mechanical?
The total net buying by institutions looks like a bottom signal; breaking down the buyers reveals a different character
Verdict: [Mechanical buying] — financial investment firms' share: 58%. Pension funds accounted for just KRW 338.1bn. Mechanical money is not a directional bet: it can reverse the next day.
Retail losses on leveraged ETFs
56tn
Citi estimate
SK hynix-related alone: 8tn (brokerage estimate)
Samsung + SK hynix market cap
-530tn
Over two days
Value erased from just two stocks
KOSDAQ leverage intensity
1.045
+10% above the pandemic low's 0.948
Debt stayed flat; only the index fell, raising intensity
KRW 30 million minimum deposit
D-2
Effective 7/31
Target: single-stock leveraged products
How much liquidation has happened, and how much remains?
KOSDAQ margin loans have already fallen from a peak of KRW 11.05tn (4/29) to KRW 6.92tn (7/28): −37% (−KRW 4.1tn) liquidated. KOSPI's decline is limited to KRW 29.75tn (6/24 — the crash began at the peak) → KRW 26.27tn, −12%. Yet forced liquidation over these two days is not in the data: published margin figures lag by two days. In fact, Tuesday's (−10.8%) published balance rose +KRW 0.45tn (margin-financed averaging down on the crash's first day). Decomposing the last two days' increase in intensity: numerator (debt) +0.001 · denominator (index) +0.076 — the index collapsed; debt did not rise.
Possible paths (index at 663)
Intensity
Comparison
Forced liquidation cuts margin loans −10%
0.94
Below pandemic level — condition for a healthy bottom
Margin loans flat + index −10%
1.16
Near tightening bear (1.17) — danger
Margin loans flat + index rebound +10%
0.95
Back to pandemic level
Intensity's next destination is a race between the numerator (liquidation) and denominator (index). Thursday's and Friday's releases provide the first test.
Invisible mechanical volume — leveraged and inverse ETFs
Rules drive these trades, regardless of directional views — size not yet quantified (see limitations)
Leveraged ETFs reset their leverage at each close, generating mechanical additional selling late on down days. On a −6% day like today, this is a possible amplifier. On the other side, bearish bets accumulated in inverse and double-inverse ETFs become fuel for short covering when a rebound begins. These are precisely the products targeted by the 7/31 minimum-deposit rules. August's deposit and margin indicators must be read alongside the restructuring of these machines.
06
Where Roots Were Missing — KOSDAQ
It did not break today. It lacked support to begin with.
Distance from the 200-day average — KOSDAQ vs. KOSPI
"Deviation" measures how far an index is from its 200-day average. The further below, the more oversold. Two markets in the same country occupy completely different worlds.
The year-to-date gap is 61pp. KOSDAQ's −33.6% deviation from its 200-day average is at the 0.1% percentile over 10 years, almost touching the 2020 pandemic crash (−34.1%).
★
KOSDAQ Structurally Lacks the Safety Net of Institutional Bottom Buying
Today institutions put into KOSPI +KRW 3.1769tn, including KRW 338.1bn came from pension funds. Pension funds' KOSDAQ flow at the same time was -41.1bn — they were sellers. Effective from fiscal 2026, the pension funds' domestic-equity benchmark of KOSPI 95 : KOSDAQ 5 appeared exactly in the numbers. By design, KOSDAQ has no structural buyer to absorb stocks in a crash.
2
Only Foreign Investors Reached Out
In KOSDAQ, where pension funds are absent by design, foreign investors made +86.9bn in net buying yesterday. Today, +308.2bn: 3.6x as much — increasing their buying as the index collapsed −6%. That hand could be one of two things: stock-specific bargain hunting would be an early bottom signal; short covering would be liquidation that burns rebound fuel in advance. The tests are whether short balances (lagged by two days) fall alongside it and whether net buying continues (§08).
3
The Refuge Disappeared Too
Until yesterday, pharma and biotech were the only defense at −0.7%. Today, LigaChem −8.5%; EcoPro BM −9.0%. Money went into food, tobacco, and shipping — all KOSPI sectors. The refuge within KOSDAQ disappeared, becoming an exit from the market.
Margin loans per 100 index points — debt's weight relative to market size
KOSDAQ margin loans divided by the index level (green, left axis). Higher means heavier debt. Intensity moves through both the numerator (debt) and denominator (index), so we overlay KOSDAQ index (gray, right axis). If the index collapses, intensity soars even with debt unchanged.
Leverage intensity — Left axisKOSDAQ index — Right axisHistorical index-low dates
The published basis (7/28) gives 0.98; reflecting today's index gives 1.045. It is +10% above the pandemic low's level (0.948), with just 1% to go to the martial-law low's level (1.055). Note: Why low markers sit on peaks: indices can fall sharply in a day, while margin liquidation lags by several days, so intensity peaks on the day the index bottoms.
Historical low
Intensity then
Versus now (1.045)
2020-03-19 pandemic crash
0.948
+10.2%
2024-12-09 martial-law risk
1.055
-1.0%
2025-04-09 tariff shock
1.088
-4.0%
2022-10-13 tightening bear
1.170
-10.7%
Yesterday's report put this at 0.985, "almost equal to the pandemic low's level". It surpassed that in a day. Not because debt rose, but because the index collapsed — decomposing the change over the last two days, the numerator (margin loans) contributed +0.001; the denominator (index) contributed +0.076.
3
KOSDAQ/KOSPI Ratio: Bottom 1% Over 10 Years
Now: 0.1170. Throughout 2016–2023, it was 0.29~0.35. This is not an event confined to today, but a derating that has unfolded throughout 2026.
07
Meanwhile, Nobody Looked Out the Window
While eyes were fixed on semiconductors — events in the Red Sea
The bypass and its exit broke down in the same week
Yanbu is Saudi Arabia's Red Sea pipeline outlet for exporting crude while bypassing Hormuz. Bab el-Mandeb is the southern Red Sea gateway those ships must pass through. Physically, they are one system.
W29 → W30: Yanbu 5.88 → 2.16 mb/d(−63%) · Bab el-Mandeb 15.1 → 8.4 ships (−44%). Two independent indicators collapsed in the same week — the week of the Houthis' maritime-blockade declaration (7/20).
Bab el-Mandeb traffic
8.4 ships
−68% versus undisrupted 2023
Lowest in 61 weeks of observations(0% percentile)
Observed Gulf seaborne exports
5.72 mb/d
−76% versus 24 before the war
PortWatch (AIS) lower bound · Satellite (SAR) is supplementary
Market-implied S2 probability
0.0%
Our assessment: 35%
The market has priced out a Red Sea blockage
WTI
$82.48
7/24 $92.76
In four days, -11% · Buying easing tensions
Market-implied probability vs. our assessment
Probability inferred from today's price, compared with an assessment based on physical data
The S2 (Red Sea blockage) rail assumes Yanbu's 4.0 mb/d is exposed, yet Yanbu has already halved to 2.16. This is not an assumption but Ongoing.
Still — a blockage is not yet confirmed
This week's sharp drop (W30) is the first observation after the blockade declaration (7/20). The verdict comes from PortWatch's weekly update: if Yanbu and Bab el-Mandeb hold at these levels in the next update, S2 is confirmed; if they recover, it was temporary noise. Traffic data lag by several days, and satellite (SAR) snapshots are estimates based on calibration factors, used only as supplementary indicators.
08
Three Winds, Five Observations
Three kinds of wind — the five observations below will test a KOSDAQ rebound
Where
Kind of wind
Direction of the gap
KOSDAQ
Structural — a gap created by institutions
Already priced in, but no path to recovery
Semiconductors
Narrative — the story ran ahead of reality
Prices reacted too strongly.
Red Sea
Invisible — physical disruption leads, prices lag
Prices reacted too weakly.
1
Thursday and Friday margin-loan releases — executed forced liquidations
Forced liquidations over these two days first appear here. A sharp fall means numerator liquidation is progressing (the healthy path in §05's table); flat balances mean selling pressure remains. But a sharp fall after the 7/31 rules take effect may be a regulatory effect, so reading it at face value can mislead.
2
Leverage intensity at 0.95 — which path gets it there?
A numerator-driven decline (liquidation) is a condition for a bottom. A denominator-driven rise (further index declines) reaches the tightening-bear level (1.17). This tests the race in §05.
3
Continuity of foreign KOSDAQ net buying + short balances
From yesterday's net buying of +86.9bn · Today +308.2bn (3.6x). If short balances fall too, it is short covering (one-off); if net buying continues with balances stable, it is bargain hunting (structural). This reveals the identity of §06's "only hand."
4
Defense of the 0.0994 KOSDAQ/KOSPI ratio
The 10-year low set on 6/25. Now 0.1170. A break below would mean another step down in derating.
5
Restoring the biotech defense — a refuge within KOSDAQ
Stability in LigaChem and EcoPro BM would be the first sign. Does the "exit from the market" (§06) stop? A refuge must return within KOSDAQ for a rebound to lift the index.
The calendar ahead — when we can check
The dates when the observations above can actually be tested
7/30 (Thu) 03:00 KST
FOMC statement — tonight's first variable
The 7/28–29 meeting's outcome arrives on the 29th at 14:00 US Eastern (03:00 on the 30th in Korea). The current rate is 3.50~3.75%. Markets lean toward a hold (CME FedWatch hold probability around 65%), but expectations for a September hike have reached 82%. This meeting has no dot plot (SEP). Why it matters for Korea — the BOK already pivoted toward hikes on 7/16 (2.75%). A hawkish Fed reading would tighten §05's leverage axis (cooling × hiking) further; a dovish reading would counter today's sharp fall. Remember, though, that the trigger for this decline was not rates but AI multiples — more than USD 1 trillion was erased from global semiconductor market cap this week.
7/30 (Thu) Intraday / after close
Margin-loan release — executed forced liquidations
KOFIA margin-credit data lag by two days. Forced liquidations on 7/28–29 first become visible here. How much of KOSDAQ's KRW 6.9tn is cut gives the first verdict in §05's "numerator vs. denominator" race.
7/31 (Fri) Takes effect
Single-stock leveraged products: KRW 30 million minimum deposit increase FSC announcement (7/28), effective immediately. The target is retail leveraged ETFs, directly tied to Citi's estimated KRW 56tn loss. A sharp fall in investor deposits or margin loans after this date may reflect "regulation," rather than "sentiment", so reading it at face value can mislead. Regime judgments are suspended for ±20 business days around implementation.
7/31~8/1 Before the weekend
PortWatch traffic update — will Suez turn down?
Current observations run through 7/26. Ships passing Bab el-Mandeb (8.4 ships, a 61-week low) take around a week to reach Suez. If Suez falls from 15.4 ships, the Red Sea blockage is real; if it holds, this is temporary noise. The market assigns this risk 0.0%.
Early August Monthly
July semiconductor export prices · Contract-price negotiations The currently cited YoY +168.5% is based on June, with a 3–4-week lag. Whether "CXMT oversupply" has affected actual contract prices first appears in July data. This tests whether §04's "product prices are still rising" holds.
Ongoing lag by two days
KOSDAQ short balances — the nature of foreign investors' +KRW 308.2bn
If balances fall alongside buying, it is short covering (one-off); if net buying continues with balances flat, it is bargain hunting (structural). This distinguishes the character of §06's only buyer. Note: We could not include the figures in this edition — the public access path was blocked (see methodological footnote).
Why We Do Not Write "Oversold, Therefore a Rebound"
Over 10 years, KOSDAQ's deviation from its 200-day average has fallen below −25% only twice: the pandemic (2020-03) and now. Of those, only one episode is complete. One case cannot support a probability. Relaxing the threshold to −20% gives 46 days, but these are overlapping samples clustered in a few episodes, not independent observations. A directional reference is all these data allow.
