The Branch That Broke Under Its Own Weight

The six-day path was as follows (KOSPI / KOSDAQ): 7/31 +17.91% / +11.63%(the KOSPI's biggest-ever percentage gain), 8/3 −5.12% / +2.44%, 8/4 +1.62% / +5.88%, 8/5 +3.76% / +2.42%, 8/6 −4.58% / +0.26%, 8/7 −0.60% / −0.36%.

The KOSDAQ rose for all 5 trading days, then paused on day 6. The KOSPI suffered two large setbacks. The KOSDAQ is now 11.0% above the surge day's close (7/31); the KOSPI is 5.1% below it.

An index returning to its old level describes price. Beneath it, the market's structure differs completely from 6 weeks ago. Let us examine each change.

① Leverage vanished — below the pandemic low

The current value is 21.9% below the pandemic crash low (0.948). It is also below the trade-dispute value (0.878). In the 10-year series, only the range-bound market of 2016 was lower.

② It began after debt drained away — then debt returned from day four

On August 4 alone, investor deposits rose KRW 7.58 trillion to KRW 110.41 trillion. Margin loans fell KRW 40 billion that day. Cash accumulated while nobody borrowed to buy: the typical aftermath of liquidation.

The very next day, the direction reversed. On August 5, when the KOSPI rose 3.76%, deposits fell KRW 7.20 trillion to KRW 103.21 trillion and margin loans rose KRW 1.02 trillion. Investors used stored cash and borrowed again. It continued on August 6: deposits +KRW 0.86 trillion (KRW 104.07 trillion), margin +KRW 0.37 trillion (KRW 28.79 trillion). Margin loans rose for two consecutive days.

③ Concentration did not unwind — trading merely dried up

Here is the commonest misreading. Samsung Electronics plus SK hynix's share of total KOSPI trading value fell to 7/31 68.3% → 8/5 47.3%. On the surface, concentration has eased. It has not. The share still swings between 47–54%.

The share fell without other stocks growing. Compared with 7/31, total KOSPI turnover fell −46%, but semiconductors shrank faster at −58%. The rest also dried up, down −19%.

Then it rebounded to 53.8% on August 6, before falling to 47.5% on August 7. Concentration has not unwound; the share fluctuates with news about the two stocks. Within the 17-stock watch list, it went from 85.3% on 8/5 to 88.0% on 8/6 to on 8/7 to 85.2%. Concentration built over 14 months has not unwound. A brief drying-up of trading should not be mistaken for deconcentration.

In June 2025, the two stocks' daily turnover was 2.20tn; by June 2026 it was 23.72tn, or 10.8x. In May it changed magnitude (KRW 9.29 trillion → KRW 22.33 trillion), then kept consuming over 90% of this basket. Today's KRW 14.24 trillion is −31% below July's KRW 20.64 trillion, but 6.5x last June's level.

④ But single-stock concentration in SK hynix really did ease

SK hynix turnover ÷ Samsung Electronics turnover was 1.69 before the crash, meaning SK hynix traded 70% more. On 7/31, the record surge day, it fell to 1.18.

But on August 6 it was 1.37 and on August 7 1.45, moving back up. On the 6th, SK hynix fell −10.37%, much more than Samsung Electronics (−6.30%). On the 7th, SK hynix −4.88% and Samsung Electronics +0.22% went in opposite directions. The market is focusing on SK hynix alone again. It is too soon to declare the single-stock dependence over.

⑤ Money moved to the KOSDAQ

KOSDAQ turnover ÷ KOSPI turnover is 11.1%(7/31) → 24.9%(8/5) → 21.5%(8/7), holding around twice its previous level. In absolute terms, on 8/5 it rose to 6.38tn before falling to KRW 5.23 trillion. While KOSPI turnover halved (KRW 48.91 trillion → KRW 24.30 trillion), this was the only place to hold its ground.

⑥ The buyers changed completely

Foreigners bought KRW 368.8 billion on the way down and sold more than twice as much, KRW 863 billion, on the way up. In August they were KOSDAQ net sellers on all five trading days(−KRW 252.2 billion on the 7th too). Investment trusts did the reverse, selling KRW 211.4 billion in the decline and buying 520.7bn in the rebound. They kept that direction on the 7th too, at +KRW 35.3 billion.

The two markets differ. In the KOSPI, foreigners sell (−KRW 5.96 trillion), while retail investors and institutions absorb it. But of institutions' KRW 3.14 trillion of buying, 58% is financial investment (+KRW 1.82 trillion) — program-like activity. In the KOSDAQ, both retail and foreigners sell while institutions buy(+KRW 1.10 trillion).

The link between the two is Investment trusts . Nearly neutral in the KOSPI (−KRW 57.8 billion), they bought 446.1bn in the KOSDAQ. Managing others' money, they cannot simply hold cash, but they can choose where to buy. They chose the KOSDAQ. This is the only flow in this period that reflects a discretionary judgment.

Over the last 10 trading days in the KOSDAQ, the correlation of foreign net buying with index moves was −0.32. For KOSPI foreigners over the same period, it was +0.91. The same foreign-investor category drives the KOSPI index but has no relationship with the KOSDAQ index. KOSDAQ selling reflects a by-product of reducing semiconductor exposure rather than a judgment on the KOSDAQ (§05).

⑦ Sector leadership changed — away from semiconductors

Over the latest 5 trading days, sector relative strength versus each sector's own market index is led entirely by non-semiconductors: Electrical/electronics +19.23 percentage points(LG Electronics), biotech +17.51(Alteogen, 20-day +31.52), power equipment +16.51(Hyosung Heavy Industries +22.27, HD Hyundai Electric +16.51, LS ELECTRIC +13.87), steel +12.51, internet +11.79 (20-day +32.74). Semiconductors over the same period: 5-day −6.90 and 20-day −2.74. Power equipment's 20-day relative strength was −2.28 on August 3; four days later it was +16.41. At the top of our scorecard's neglect index now sit Wonik IPS, SK hynix and Jusung Engineering — all semiconductors or their equipment.

A figure has become visible for the first time: semiconductor equipment concentration within the KOSDAQ is 30.6%(as of 8/7; it varies between 30–40% daily). It is below the KOSPI's 89.2%, but the impression that the KOSDAQ has no semiconductor connection is false.

Power equipment is rising because of the data-center power bottleneck. Our power data show the backdrop. PJM, serving Virginia's Data Center Alley, recorded minimum overnight load of 94.0GW in July 2026, +0.8% year on year, effectively flat. Texas ERCOT, by contrast, was at 56.0GW, +7.5%. When power is the bottleneck, transformers and heavy electrical equipment command the price.

⑧ The won strengthened instead

USD/KRW was KRW 1,481.8 on July 20. On August 7 it was KRW 1.4174 thousand, a KRW 64.4 (4.3%) appreciation. Over just the latest 10 trading days (7/27–8/7), foreigners sold 5.96tn in the KOSPI while the won strengthened KRW 44. Even over the 3 trading days of the stock-market collapse (7/28–30), the won strengthened KRW 22.3.

In the last week of July, Korean stocks collapsed with an unprecedented 2 consecutive trading days of circuit breakers. An event that large seems to demand an equally large cause. So we looked outside and investigated two suspects.

Suspect ① Dollar liquidity — reserves were lower in April

On July 29, US bank reserves were $2,944.5bn, below $3 trillion. They fell $155.9bn in 2 weeks, while the Treasury General Account (TGA) rose $174.5bn. Correspondence: 0.894. Treasury absorption explains most of the reserve decline.

So far, liquidity was indeed drained. The problem is that this was not the first time.

Reserves on April 22 were below today's level. June 17 was also near today's level; the KOSPI hit a record the very next day. The ON RRP buffer had already reached 0 in March. Reserves below $3 trillion without that buffer are not new.

There is another decisive counterexample: who was hit, and by how much, over those 5 days.

Bitcoin fell only 1.86%. In a real dollar squeeze it would be the first and hardest-hit asset. Nasdaq fell −4.86%, Nikkei −6.86%. Korea took 4.4x the Nasdaq's hit and 11x Bitcoin's.

The plumbing indicators were quiet too. SOFR 3.65% vs effective federal funds rate 3.63% — gap +2bp. Fed Standing Repo Facility (SRF) usage over the latest 60 days was USD 0. In a real cash shortage, these two would scream first.

Suspect ② Capital outflows — the won contradicts them

Over the three days of July 28–30, foreigners net sold in the KOSPI KRW 4.4 trillion, and over the latest 10 trading days (7/27–8/7), KRW 5.96 trillion. If capital outflows caused the crash, the won should have collapsed. The won strengthened KRW 44 (3.0%) over the same 10 trading days.

The sale proceeds did not leave as dollars. They may have moved into won deposits or bonds; the flows may instead have unwound derivatives or arbitrage positions without actual capital movement. Either way, the exchange rate does not support the story that foreigners left Korea and caused the crash.

Nasdaq −4.86%, Nikkei −6.86%: the wind was real. But the KOSPI fell −21.18%. The difference was the weight on the branch, not the wind.

The weight had two layers.

The first weight is concentration. When over 90% of a market's trading rests on two stocks, diversification fails as their story wavers. Those two are what investors can buy and what they can sell.

The second weight is debt. KOSDAQ margin loans peaked at KRW 11.05 trillion on April 29. A falling index erodes collateral, triggers forced selling, and pushes the index lower again. This explains the time from the July 28 sidecar (09:06) to the circuit breaker (10:13): 67 minutes.

The trigger came from abroad. Broadcom's early-June earnings warning opened the first crack in the AI semiconductor story, and China's CXMT's successful DDR5 mass production and major IPO targeted Samsung and SK hynix's position directly. But a trigger is only a trigger. The same news took 4.86% off Nasdaq.

The amplifier works both ways. Once forced liquidation ends, the people who must sell disappear. That is exactly what happened on July 31.

SK hynix closed at its +29.80% daily limit. Since the daily price limit widened to 30% in 2015, this was the first . Samsung Electronics rose +27.05%. Foreigners net bought in the KOSPI KRW 7.5 trillion, with over KRW 5 trillion concentrated in SK hynix alone. The KOSPI closed +17.91%(+1,001.89p), its biggest percentage gain ever.

After that, their paths split. The KOSDAQ rose all 5 trading days and passed 800. The KOSPI suffered two setbacks and fell below the surge day's close. Liquidation ended in both markets. Why did only one stand up?

The KOSPI's β is 1.47. When Asian semiconductors move 1%, the KOSPI moves 1.47%. Its α is −0.25, negative.

If the same semiconductor exposure can be obtained in Taiwan or Japan at β 1.0 with less lost alpha, cutting Korea's weight is rational. Our pipeline labels the KOSPI “structural discount — persistent 20-day cumulative excess losses”, with the same conclusion from 60-day and 120-day fitting windows.

The KOSDAQ differs. Its β of 0.919 is nearly the basket's, and the verdict is neutral. Its 20-day cumulative excess return is +10.62, positive. The KOSDAQ carries no discount.

This fits the flows in §02⑥. Over the latest 10 trading days, foreigners sold in the KOSPI 5.96tn and in the KOSDAQ 614.9bn. KOSDAQ selling reflects a by-product of reducing semiconductor exposure rather than a judgment on that market, hence its −0.32 correlation with the index.

§03 established dollar liquidity's alibi. But that investigation revealed a fact about the future: July's absorption was a scheduled event, rather than an accident.

In its May 6 quarterly refunding statement, the US Treasury wrote: “We expect the end-July cash balance to reach around $1 trillion (±$50 billion).” (translated) The actual TGA on July 29 was $970.4bn, almost exactly as planned.

The next-two-quarter plan announced on August 3 points the other way.

Then the August 5 quarterly refunding statement made the path clearer. It is not a steady decline.

The TGA exceeds $1 trillion again in October. Mid-September corporate taxes replenish cash, then October bill issuance rises to meet seasonal fiscal outflows. Perli warned that “large net bill issuance in coming months could tighten money markets” (translated). That pressure now has a date: October .

The forward guidance was unchanged. The Treasury again said it would keep nominal coupon and FRN auction sizes steady “for at least the next several quarters” (translated). August refunding totals $125bn (3-year $58 / 10-year $42 / 30-year $25), unchanged from May. All the increase in $739bn of net borrowing goes into bills.

The year-end target of $850bn is $79bn below today, returning that amount to bank reserves. The Fed side also leans toward expansion. Quantitative tightening (QT) ended in November 2025. From the December FOMC, reserve management purchases (RMP) have bought around $40 billion of bills a month. Total Fed assets rose from $6,655.9bn on March 18 to a July 29 level of $6,738.2bn. In a July 9 speech, New York Fed markets head Roberto Perli said reserves were “ample range”and described April tax-day pressure as “modest and short-lived.” (translated)

Around 85% of recent issuance matures within 1 year. One-fifth of federal debt matures within 4 months and one-third within 1 year. Higher rates feed directly into refinancing costs. FY2026 net interest costs are $1.0 trillion, or $2.8 billion a day.

On July 31 (US time), the US and Japan conducted joint foreign-exchange intervention, selling dollars and buying yen, at an estimated scale of JPY 6–7 trillion. It was their first joint action in 15 years. Japan's finance minister, Satsuki Katayama, said it addressed the yen's “excessive volatility and disorderly movements” (translated) and that “we will not hesitate to intervene again.” (translated) President Trump said “Japan needed help because of the weak yen.” (translated)

The effect was immediate: from around JPY 164 to the JPY 155 range. USD/KRW fell alongside it to the KRW 1,420 range (a 9-month low). July's KRW 125.4 fall in USD/KRW was the largest since March 2009.

What intervention can and cannot change is answered by the structure we saw in #08.

Companies, rather than securities, drive yen weakness. Portfolio investment actually showed $265bn of net inflows. Most outflows reflected real-economy decisions: companies building overseas factories and lending to subsidiaries.

Intervention acts on L1. Sell dollars in the spot market and speculative positions hit stops and unwind. But intervention does not bring L2 back. Companies do not reverse overseas investment decisions because FX moves for a few days.

The move is already reversing. From the intervention low of 156.98 (8/3), it reached 158.36 on August 7 — JPY 1.4 over four days.

The picture is this.

The branch broke, the wind passed and the tree stands, but with one branch missing.

Are new buds emerging on another branch, or is the broken one regrowing? Today's data cannot distinguish them. The six points above record what has changed so far, rather than promising it will continue. The next chapter sets out what would tell us.