Korea's 2026 stock plunge came after its biggest boom in decades, not a global panic

Every fall had its own mix of causes

Korea's twelve sharp falls since 1997 share no single cause. Most came with a jump in Wall Street's fear gauge to about double its usual level. Two came with a collapse of the won, when the dollar rose by more than half against it, and two, in 2004 and 2026, with neither⁠^{Source 1,Source 2,Source 10}. It helps to picture four stages: the fuel that built up, the spark that set a fall off, the wind that made it worse, and what came before calm. What follows shows timing, not proof of cause.

We picked the falls on daily closes of the KOSPI: every fall of a fifth or more since 1997, except three short swings in the rebound from the 1998 low. The slide from July 2024 to April 2025, a fall of a fifth on daily closes, appears as two shorter falls, July–August 2024 and February–April 2025. The long slide from 2018 to 2020 counts as two falls, split at its August 2019 low.

To measure them we use the OECD's monthly average of Korean share prices, which smooths daily swings and so makes falls look shallower, most of all falls that happen within weeks⁠^{Source 1}. The deepest were 59.2% in 1997–98 and 48.8% in the 2000–01 IT bust; 2007–08 fell 46.4%⁠^{Source 1}. The 2026 fall, 21.6% from June to August so far, ranks mid-sized for now in monthly averages: six of the eleven earlier falls were deeper⁠^{Source 1}, though on daily closes it is already deeper than the 2020 and 2021–22 falls.

In ten of the eleven earlier falls the VIX, an index of how much US shares are expected to swing, closed at about double its median of 18.1 or more⁠^{Source 2}. It reached 79.1 during the 2007–08 fall and 82.7 in 2020⁠^{Source 2}. The exception was 2004: Korean share prices fell 17.7% from April to July in monthly averages, yet the VIX never closed above 20.0, US shares slipped 2.4% and won per dollar rose only 0.5%⁠^{Source 1,Source 2,Source 4,Source 10}. The won collapsed twice, when the dollar rose by more than half against it: by 54.3% in 1997–98 and 52.9% in 2007–08⁠^{Source 10}.

Some sparks were regional. US shares rose 10.8% during the 1997–98 fall, an Asian crisis rather than a global one⁠^{Source 4}. Korea's political shock on 3 Dec 2024 came with a VIX of only 13.3; the fall in our table began in February 2025, and its VIX high, 52.3, came on 8 Apr 2025⁠^{Source 2}.

Like 2004, the 2026 fall came without a global panic: the VIX closed no higher than 20.7, US shares rose 4.8% from June to August, and won per dollar fell 8.3%, a stronger won⁠^{Source 2,Source 4,Source 10}. Apart from swings in the flows themselves, global fear and US share returns explain far more of the money moving into and out of emerging-market funds than domestic factors do, in an IMF study that does not include Korea⁠^{Source 11}. Yet from June to early September 2026 emerging-market shares as a group gained⁠^{Source 5}. The table below sets the twelve falls side by side.

The mood was upbeat at the top

No data measure whether people felt a bubble: none of the surveys we could use asks whether people think stocks are overpriced. News mood about the economy, the closest measure, swings widely: in the year before each peak it dipped below its long-run average at some point, yet on each peak day it was above it⁠^{Source 3}.

The Bank of Korea's news sentiment index reads the tone of economic news each day. In all eight falls since 2007 it was above its long-run average on the day stocks peaked. In the year before each peak it had been higher at some point, and also lower: every time it fell to 93.0 or below, and before the 2020 peak to 76.5⁠^{Source 3}. In 2026, mood went from 125.7 on 3 Nov 2025 down to 87.5 on 9 Mar 2026, then back up to 108.1 on 22 Jun 2026, the day stocks peaked⁠^{Source 3}.

In each fall mood then sank below its dip of the year before, sometimes far below: in the global financial crisis to 57.7 on 8 Oct 2008, against 79.7 in the year before the peak⁠^{Source 3}. In 2026 the gap is small: mood reached 85.0 on 18 Oct, its lowest since 11 Apr 2025, against 87.5 in March⁠^{Source 3}. The chart below compares mood on each peak day with its lowest reading from the peak to three months after the low.

The consumer sentiment index, a monthly survey of households, fell to 67.7 in December 2008 but barely moved during the 2026 fall, with a low of 104.5 in August 2026⁠^{Source 12}.

Sensing a crash is common, so it is no warning by itself: US investors in long-running surveys put the chance of a crash at more than ten times its historical frequency⁠^{Source 13}. An IMF study finds that money flowing into funds tends to peak just before large falls, a quantity worth watching alongside mood⁠^{Source 9}. In Korea's data, mood was above average on each peak day; its lowest readings came during or just after the falls, though in 2024 and 2026 only a little below its dips of the year before.

2026: a run-up resting on two chipmakers

Before the 2026 fall, Korean share prices had their biggest two-year rise since the boom of 1987–89, resting on two chip stocks and leveraged bets⁠^{Source 1,Source 5}. The spark was a global repricing of chip shares, not a fall in Korea's chip exports⁠^{Source 5,Source 6}.

Prices rose 209% in the two years to June 2026, the most since December 1988 but short of the record: 282% in the two years to February 1988⁠^{Source 1}. In data back to 1981, only two earlier run-ups doubled prices in two years: the boom of 1987–89, and the run-up of 1998–99, when prices rose 152% in the two years to December 1999⁠^{Source 1}. Each ended in a fall of almost half: 48.6% from March 1989 to August 1992, though the first took more than three years, and 48.8% in the IT bust that followed⁠^{Source 1}. Two cases are a reason for caution, not a rule.

Shares of Samsung Electronics and SK Hynix, Korea's dominant memory chipmakers, more than tripled in the first half of 2026, and the wider market followed them⁠^{Source 5}. Bets piled up on top: assets of leveraged products on the major memory-chip stocks, Samsung, SK Hynix and the US firm Micron, rose from under a hundred million dollars in mid-2025 to more than thirty-eight billion in June 2026⁠^{Source 5}. In June the Bank of Korea Governor said the surge had come with "a significant increase in leveraged equity investment"⁠^{Source 14}.

Research shows why such a run-up matters but cannot time a fall. In US industries, the chance of a fall of two-fifths or more within two years rises from one in five to more than half as a two-year gain over the market goes from fifty to a hundred per cent⁠^{Source 8}. Yet in run-ups that did crash, prices rose about another thirty per cent on average before the peak⁠^{Source 8}.

The spark came from abroad. At the end of June, tech stocks began to falter amid the strain of a prolonged conflict in Iran, and by early September the price-earnings ratios of US-listed semiconductor makers had fallen by two-fifths⁠^{Source 5}. The BIS says Korea plunged furthest because of its exposure to AI-linked firms and turmoil in a few leveraged investment vehicles⁠^{Source 5}.

The chart below shows that chip export values, after a dip in July, were well above their June level by September and 268% up on a year earlier, while Korean share prices were 19.0% below June⁠^{Source 1,Source 6}. The fall repriced chip shares; it did not follow a fall in chip sales.

Selling that follows prices, not feelings

The 2026 fall was among the fastest in the data, and some of what looks like panic may be selling that borrowing or fund rules force on investors⁠^{Source 1,Source 5}. The data we can use include no daily figures on forced sales, so they cannot tell the two apart.

Prices fell 21.6% in two months, by August. Timed from each fall's high month to its low month, only 2020's 19.0%, also in two months, matched that pace. Every earlier fall of a fifth or more in monthly averages took at least 11 months to reach its low⁠^{Source 1}. Steeper two-month drops did come inside the 1997–98, 2000–01 and 2007–08 falls⁠^{Source 1}. July 2026 alone was 17.6% below June⁠^{Source 1}.

When the stock behind a leveraged ETF falls, the fund's leverage rises above its target, so it must sell⁠^{Source 5}. For SK Hynix, the trading such funds needed after a ten per cent move reached almost five billion dollars by the June peak⁠^{Source 5}. Against daily turnover of under ten billion, the BIS author estimates from a model, not observed trades, that this could add two to four points to such a move⁠^{Source 5}.

Hedging autocallable notes can add large trades near their barrier prices⁠^{Source 5}. The author stresses concentration: "a handful of instruments referencing two stocks that dominate an index, in a market where retail participation and leverage are both unusually high"⁠^{Source 5}.

Borrowed money works the same way, through the margin call. In China's 2015 crash, margin investors sold heavily as their leverage neared its limit, and the stocks they held fell abnormally and later recovered, the mark of a fire sale⁠^{Source 15}. The authors add that the market might have corrected eventually even without that trigger⁠^{Source 15}.

Herding, investors copying each other, is the other suspect. Before Korea's 1997 crisis foreign investors chased recent returns, but over a day or a few days the authors find nothing that singles them out as destabilising⁠^{Source 16}. Herding measures also cannot tell copying from reactions to the same news⁠^{Source 17}.

A fast fall need not be irrational: when investors expecting bad news cannot sell short, their information tends to come out during declines⁠^{Source 18}. The speed of 2026 fits forced selling but does not prove it: 2020 was as fast, in a global panic.

What came before the market steadied

Recoveries took from two months to 72 months. Most followed rate cuts, but the 2007 high was regained in 2011 as rates rose, and the 2022 low turned while the rate was still rising⁠^{Source 1,Source 7}.

Prices took seven months to regain their earlier high after the 2020 peak, 39 months after 2007 and 72 months after 2011⁠^{Source 1}. Three months after the low, the 2020 market had regained 83% of its fall, the 2008 market only 7%⁠^{Source 1}.

The Bank of Korea cut its base rate during or near several falls' lows; the first cuts came on 8 Feb 2001, 9 Oct 2008 and 17 Mar 2020⁠^{Source 7}. But the rate rose 2.00 pp from the 2021 peak to the 2022 low, and that market still turned, if slowly⁠^{Source 1,Source 7}. Cuts are neither the only route to calm nor a guarantee of it.

In June 2026 the Governor said the Bank needed to raise its rate before it was too late, to keep prices stable⁠^{Source 14}. It did so on 16 Jul 2026 and 27 Aug 2026; by 15 Oct the rate was 0.50 pp higher than at the June peak⁠^{Source 7}. The chart below shows the base rate through each fall.

Research on other falls shows what else came before calm. In August 2024 US shares regained all of a week's losses by the Friday, and authorities did not need to intervene⁠^{Source 19}. In 2008 a US ban on short selling did little to slow the fall in financial shares⁠^{Source 20}.

The 2026 fall is still open: monthly average prices edged up 3.3% from August to September⁠^{Source 1}. But prices fell again in October, and the October average is not out yet. The VIX was 20.0 on 16 Oct⁠^{Source 2}. We declare no low and make no forecast.

How a crash-risk model for Korea could work

A useful model would estimate how fragile the market is, as a probability or a state, from measures of fuel, spark and wind, tested on Korea's twelve falls; it cannot date the next crash⁠^{Source 8,Source 9}.

For the fuel, the US industry study turns a two-year run-up into a crash probability, and rising volatility and share issuance set apart the run-ups that crashed⁠^{Source 8}. An IMF framework adds quantities, such as trading and fund flows, to prices⁠^{Source 9}. The table below sets out the design.

For the spark, it would track the won, global fear and US share returns. In an IMF study that does not include Korea, the last two explain far more of the money moving into and out of emerging-market funds than domestic factors do, apart from swings in the flows themselves⁠^{Source 11}. For the wind, it would measure borrowing, leveraged-fund trades against turnover and one-sided option bets⁠^{Source 5}. Korea's own record is thin: only two earlier run-ups doubled prices in two years, too few cases to estimate from Korea alone⁠^{Source 1}.

The output would be a probability of a large fall, a forecast of how lopsided returns will be, or a state such as calm, fragile, falling or recovering⁠^{Source 18}. It would be fitted on some falls and checked on others, knowing that any warning threshold trades missed falls against false alarms⁠^{Source 9}.

Much of the data such a model needs is unpublished or not cleared for reuse, as the table shows. And as an IMF paper puts it, "bubbles can only be identified with complete certainty ex-post"⁠^{Source 9}.