Anatomy of the Korean Stock Market Bubble

The story everyone tells

Start with the version of events that is true and insufficient, because every big move gets one.

The fundamental story has two legs. The first is memory. Samsung Electronics and SK Hynix sit on either side of a duopoly in high bandwidth memory, the component every AI datacenter buys by the ton, and 2026 has been the year the market decided the memory cycle is not a cycle anymore but a secular repricing. The two of them dominate the KOSPI by weight, so when they run, the index runs. The second leg is governance. Korean equities spent decades trading at a structural discount to book value, the famous Korea discount, with the index average price-to-book hovering around 1 while developed peers sat at two or three times that. The standard explanations were chaebol cross-shareholdings, hostage minority shareholders, and treasury shares used as control currency rather than cancelled. Since early 2024 the government has been dismantling that discount on purpose: the Corporate Value-up Program first, then successive rounds of Commercial Code amendments, the third of which was promulgated in March this year and did the two things foreign investors had asked for since the nineties. It made cancellation of treasury shares the default and it extended directors' fiduciary duties to shareholders rather than to the company in the abstract. President Lee campaigned on KOSPI 5,000 barely a year ago and was mocked for it. The index went through 5,500 in February.

Both legs are real. Repricing a structural discount is a legitimate one-time event, and an earnings supercycle in your two largest constituents is a legitimate reason for the index to rise. If the KOSPI had finished the year up 40% on this news, nobody would be writing anatomy essays. The problem is arithmetic. Governance reform and HBM margins are stories about where the level of the index should settle. They are not stories that explain a 100% move in six months, followed by a 25% retracement in fifteen trading sessions, in the eleventh largest equity market in the world. Levels come from fundamentals. Paths come from flows. And the flow picture in Korea this year is one of the strangest I have ever looked at.

The buyer and the seller nobody mentions

Here is the strangeness in two numbers. Korean retail investors bought a net 97.4 trillion won of domestic stocks in the first five months of the year, about $70 billion, per Seoul Economic Daily. Foreign investors over roughly the same window sold a net 113.8 trillion won, the largest foreign exit in the history of the market by a factor of four. The previous records were about 25 trillion won in 2020 and 26 trillion in 2021. Estimates circulating this week put the running total above $100 billion. So the best-performing large index on the planet spent its entire melt-up being dumped by the investor class that usually gets credit for disciplining prices, and absorbed by households.

The retail side has deep roots. Korea's stock-owning population went from about 6 million people in 2019 to more than 14.5 million by the end of 2025, which is over a quarter of the country. The 2020 cohort called itself the Donghak ant movement, a name borrowed from an 1894 peasant uprising, ants being the local term for small individual traders. The ants who buy foreign stocks, the so-called Seohak ants, became famous for making Korean retail one of the largest foreign holder bases of triple-levered US semiconductor ETFs. I mention this not as color but as evidence of a revealed preference: this is a retail culture with a decade of practice at expressing views through leverage wrappers rather than through shares. When the domestic index finally gave them a reason to come home, they brought the habit with them. Brokerage accounts opened at record pace this year, margin lending at Seoul brokerages ran to records alongside the index, and the financial regulator was warning publicly about the margin-fueled character of the rally before the top, which is not a sentence you get to write about most bubbles.

The foreign side is stranger than it looks, and the strangeness matters. Some of the selling was a view: taking profits on a doubled market, or skepticism about how much of the HBM future was already in the price. But a large share of it, per CNBC's reporting in June, was not a view at all. As the KOSPI doubled, Korea's weight in global benchmarks swelled, and every fund that manages to a fixed country allocation or a tracking-error budget became a mechanical trimmer of the outperformer. The selling was a formula responding to the price, not an opinion about it. Goldman, for what it is worth, was still publishing upside targets in early June.

Put the two sides together and you get the actual microstructure of this market at the top: the marginal buyer was a levered household, and the marginal seller was a rebalancing rule. Neither of them trades on valuation. The participant who is supposed to anchor prices to fundamentals, the unconstrained professional with capital and a view, had largely left the building. Markets can run a long way in that configuration, and they can fall a long way too, because the thing that normally leans against the wind is absent in both directions. Short sellers, the other classical stabilizer, only returned to Korea in March 2025 after a blanket ban, and a market that spent years structurally long-only does not grow a deep borrow-and-lend culture back overnight.

A machine that trades the close

Now the part I actually want to write about, because it is the part the coverage keeps gesturing at without doing the arithmetic. The instrument at the center of this market is the single stock leveraged ETF. Products offering daily 2x exposure to Samsung and SK Hynix individually gathered about 7 trillion won, call it $5 billion, in roughly their first month. Seoul Economic Daily reported that number this morning. Local commentators are writing worried op-eds about it. They are right to worry, and the reason is a piece of arithmetic that fits on a napkin.

A daily leveraged fund does not hold leverage the way a margin account does. It targets a fixed multiple L of its underlying's return each day, which means it must rebalance to that target at every close. Take a fund with equity A running leverage L. Its exposure is LA. The underlying moves r on the day, so the exposure leg becomes LA(1+r) while the fund's equity becomes A(1+Lr). To reset tomorrow's leverage to L, the fund must hold L·A(1+Lr), and the difference between what it must hold and what it has is the trade it is forced to do:

trade = LA(1+Lr) − LA(1+r) = A · r · L(L−1)

Two things about that expression deserve a slow look. First, the sign. For any L above 1, the trade has the same sign as r. The fund buys after up days and sells after down days, every day, by construction, into the close. Second, the perverse symmetry: set L to minus one, an inverse fund, and L(L−1) equals 2, positive. The inverse product also trades in the direction of the day's move. Levered longs and levered shorts do not offset each other's rebalancing. They stack. Everything in the complex is momentum flow at the close, no matter which direction its holders are betting.

Now put numbers in. Seven trillion won of 2x single stock product, L(L−1) = 2, and a day where the underlying falls 6%, which is no longer a hypothetical in this tape. The forced trade is 2 × 7T × 0.06, about 840 billion won of selling, more than half a billion dollars, concentrated in the closing auctions of exactly two tickers. That flow does not care about HBM contract pricing. It cannot be argued with, postponed, or reasoned into averaging in tomorrow instead. And note the feedback: the selling worsens the close, the worse close enlarges r, and a larger r enlarges the next day's required trade if the move continues. The machine's output is proportional to its own recent output. I wrote about a different version of mechanical flows overwhelming a market's capacity to absorb them in the Jane Street Bank Nifty essay; the difference here is that no trader is even choosing this. The prospectus is doing it.

There is a second cost, slower and better hidden. Chain the daily resets together and the compounding drag on an L-times fund scales with L(L−1)σ²/2, where σ is the underlying's volatility. Run that for a 2x product on a single memory stock doing 50 vol, which is roughly where this complex lives now, and the reset grinds away something on the order of 25 percentage points a year relative to twice the buy-and-hold return. In a market that only goes up, nobody notices, because the leverage flatters every statement. In a market that chops sideways at high volatility, which is what a post-crash Korea is most likely to do, the products bleed even if the underlying goes nowhere. Five billion dollars of household savings is now parked in wrappers whose expected return in a flat, volatile market is meaningfully negative, held by investors who mostly believe they own Samsung with extra steps.

And Korea no longer has this machine to itself, pointed at its own stocks. SK Hynix listed an ADR on Nasdaq under SKHY, and in mid-July the first US-listed leveraged SK Hynix ETFs launched into the frenzy; the ADR jumped 19% on the day and dragged Micron, SanDisk and Western Digital along on pure sympathy flow. The same duopoly now gets force-rebalanced twice a day, once at the Seoul close and again at 4pm in New York. For the wider context: US-listed leveraged ETF volume is running around $45 billion a day, up 50% this year, aggregate leveraged ETF assets sit near $218 billion after growing 60% since March, and Benzinga reported traders warning that daily rebalancing flows across the complex have touched $50 billion, four times January's run rate. A SpaceX-linked leveraged product did $4 billion of volume in its first week, the biggest ETF debut in history. Korea is not an outlier. Korea is the preview.

The way down is the same machine in reverse

The proximate trigger for the break was almost comically small relative to the damage. A Chinese lab, Moonshot AI, shipped a model that read as close enough to the US frontier to put a question mark over the AI capex narrative, semis sold off globally, and the Philadelphia semiconductor index went into a 20% drawdown. Add some geopolitical noise and oil above $80. None of this is Korea-specific. The KOSPI's response was to fall roughly twice as hard as the global complex, which is exactly what the structure above predicts. An index whose marginal buyer is levered converts any external shock into internal forced flow: the down move triggers margin calls, margin calls are sales, the sales worsen the close, the leveraged ETFs sell 2Ar into that close, and the next morning opens with fresh margin calls priced off the marks the machine itself produced. KED Global's reporting on the July 13 and 14 sessions described exactly this, forced unwinds deepening the strain, Samsung and Hynix whipsawing while brokers liquidated undercollateralized accounts.

It is worth being precise about what the foreigners are doing on the way down, because the popular story has them fleeing. The benchmark mechanics that made them sellers on the way up flip sign below the peak: as Korea's index weight shrinks back, the same fixed allocation funds become mechanical buyers of the underperformer. Mechanical flow dampens nothing, but it does reverse, and this is one of the few stabilizing terms in the system. The destabilizing terms, margin and daily reset leverage, do not reverse. They only ever push with the move. That asymmetry, stabilizers that are weak or absent and amplifiers that are large and growing, is the whole diagnosis of this market in one sentence.

I would also gently note what the crash is not. It is not, so far, a fundamental repricing of the memory story; HBM contracts did not change in June. It is not a governance reversal; the Commercial Code amendments are law. The things that justified the rally are intact, and the index still sits enormously above January. What broke was the path, not the level, and paths break when the flow structure that produced them saturates. There is a reading of July in which the Korean market is doing something almost healthy: finding out, violently, how much of the last 3,000 index points was owed to borrowed money.

So is it a bubble?

Honest answer: the word does no work. If bubble means prices detached from any defensible fundamental story, Korea does not qualify; the discount-closing and the memory cycle are defensible, and even after the doubling the index's multiple is not exotic by global standards. If bubble means a price path manufactured by self-reinforcing leveraged flow that overshoots in both directions, Korea qualifies completely, but then so does every market where this product structure is metastasizing, including the one where SPCH just set a volume record. The valuation question and the mechanism question have different answers, and almost every argument you will read this week is two people answering different questions at each other.

The mechanism framing also does something the bubble framing cannot: it makes predictions. It predicts the crash accelerates into closes on big down days, because that is when the reset trades. It predicts volatility stays elevated after the fall, because the leverage stock has to be worked off through margin calls and product outflows rather than through one repricing. It predicts the leveraged wrappers underperform their naive multiple badly from here, because chop is where L(L−1)σ²/2 eats. And it predicts the index finds its footing not when sentiment improves but when the forced-flow terms shrink: margin balances down, leveraged ETF assets down, the close no longer trading like a machine is leaning on it. None of that depends on knowing the right price for a share of SK Hynix, which is convenient, because I do not.

What to watch

Three observables, none of which require an opinion about memory-chip fundamentals.

First, the closing auction signature in Seoul. The reset trade arrives in the last minutes of the session, so the tell for leverage-driven markets is amplification into the close on trend days: the final half hour extending the day's move rather than fading it, day after day. When that signature weakens, the machine is smaller. This is measurable from public minute data, and it is the single cleanest gauge of how much of this market is still mechanical.

Second, the two leverage stocks: margin lending balances at Korean brokerages, published regularly, and assets in the single stock leveraged complex. The rally cannot safely resume while both are at highs, whatever the news flow does, because the fuel for the next forced unwind is already loaded. Watch for the leveraged ETF assets to fall faster than performance alone explains; that is the sign holders have understood the decay math, and it removes the close-trading flow on the way out.

Third, the foreign flow flip. If the record foreign selling was mostly benchmark mechanics, it should reverse roughly in proportion as Korea's index weight falls back, and the reversal should be insensitive to headlines. If instead foreigners keep selling into a cheaper, lower-weight Korea, the mechanical story was wrong and something more deliberate is going on. The next few months of exchange flow data settle the question either way.

The recurring theme of these essays is that the wrapper matters more than the underlying, and Korea in July 2026 is the cleanest demonstration of it I have seen at national scale. The two companies at the center of this story did not become half again more valuable by June and a quarter less valuable by mid-July. What changed was the plumbing around them: how much borrowed money held the shares, through which products, with what reset rules, rebalanced at which closes, trimmed by which benchmarks. The price path of the world's most important memory duopoly is currently being written by prospectus formulas and margin desks, in two time zones, and most of the commentary is still arguing about price-to-book. Watch the machine, not the multiple. The multiple is just where the machine happened to stop.