Stock Market Circuit Breaker Explained: The Halt Stops the Exchange, Not the Price

A circuit breaker does not stop a stock from moving. It stops one venue from quoting it, and on 28 July 2026 the Korea Exchange demonstrated exactly how narrow that difference is. At 10:13 in the morning, after the KOSPI had held losses of more than 8 percent for over a minute, Korea's Level 1 breaker fired and the market went dark for twenty minutes. Samsung Electronics closed down 13.39 percent. SK Hynix closed down 14.65 percent. The index finished at 6,023, off 10.84 percent, its fourth-largest single-day fall on record. The next morning it happened again, and Korea booked the first back-to-back market-wide halts in the exchange's history.

Inside those twenty minutes, a Samsung perpetual future on an offshore venue kept quoting. It kept charging funding. It kept liquidating positions against a reference price that nobody in Seoul was permitted to update.

This is a Kodex walkthrough with Eunha, who works the seam between what a market structure does and what it feels like it does. The gap here is unusually wide. A halt reads as protection, and mechanically it removes the deepest book on the asset while every wrapper built on top of that asset keeps trading.

Eunha starts where the confusion starts, which is the word itself.

What a circuit breaker actually stops

"Read the rule as an address," she says. "Not a property of the stock. An address."

A circuit breaker is a threshold written into one venue's rulebook. Cross it and that venue stops matching orders for a fixed period. Korea's Level 1 breaker triggers when the index falls 8 percent from the prior close and holds there for a minute, and the halt runs twenty minutes. On 29 July the tech-heavy Kosdaq tripped its own breaker at 12:19 p.m. and the Kospi followed within minutes, both frozen for twenty, the benchmark trading at 5,535 by 12:47.

America's version runs on the S&P 500 at three levels. A 7 percent fall buys a 15-minute pause, 13 percent buys another, and 20 percent closes the market for the day. Individual names carry their own mechanism, Limit Up-Limit Down, which pauses a single stock when it travels too far too fast against its own recent range. The thresholds differ by market. The architecture does not.

What every one of those rules shares is an owner. It belongs to a market operator, it governs the order books that operator runs, and it expires at the boundary of that operator's authority. Korea's breaker had complete power over the KRX book in Samsung Electronics and no power at all over any other book in the world quoting the same company.

That is not an oversight. The World Federation of Exchanges, the industry's own trade body, put it in writing in 2021: "Since the same stock can be traded across different venues and jurisdictions, and can underlie the price of derivative products, some markets lend themselves to establishing a certain level of coordination. In other cases, this may not be desirable since, for example, the reasons for a circuit breaker being triggered in one venue, may not apply in another, even for the same stock."

When the WFE surveyed its member exchanges on whether coordination should be extended across venues and jurisdictions, none of them ranked it as a priority. The prevailing view in the survey was that each trading venue should be free to run whatever volatility control suits its own market structure, participants and trading mechanism. A handful favoured more alignment on dual-listed names to close small arbitrage windows. The rest treated the boundary as a feature.

The fragmentation is deliberate. It was designed in, by the people who run the venues, on the reasoning that a rule calibrated for one book is the wrong rule for another.

Why the price kept moving anyway

Eunha pulls up the second book, the one that stayed open.

On 2 June 2026, Binance Futures listed USDT-margined perpetual contracts on three Korean blue chips: SAMSUNGUSDT, SKHYNIXUSDT and HYUNDAIUSDT. Up to 20x leverage. Funding settled every eight hours. Trading twenty-four hours a day, seven days a week, on a venue that has never adopted a Korean market rule and was never going to.

So when the KRX book froze, the Samsung perpetual did not freeze with it. There was no mechanism by which it could. The contract references the share price, and referencing something is not the same as being bound by the rules of the place it trades.

A perpetual has no expiry and no delivery. What keeps it tethered to reality is the funding rate and the mark price, and the venue computes both, continuously, whether or not the underlying market is open. On Kodex that distinction has its own piece: do perpetual futures expire works through what happens when a venue itself can end. Here it merely pauses, and the others carry on without it.

Leveraged ETFs on Samsung and SK Hynix sit in that category too. So do tokenized shares. Each is a separate instrument on a separate venue, referencing a price that had stopped being produced.

What does a perpetual price when the reference feed stops?

It is 10:20 in Seoul and the Samsung Electronics line on the KRX screen has not moved in seven minutes. Eunha has the SAMSUNGUSDT chart open beside it, and that one is still printing candles.

A market maker on the offshore perpetual has one job during normal hours: quote both sides, hedge the resulting exposure in the underlying. During a halt that hedge is gone. The stock cannot be bought or sold on its primary book, so the maker holds inventory it cannot offset. In a name that just fell 8 percent in a minute. That might reopen anywhere.

The rational response is not to stop quoting. It is to widen, hard, and price the reopening gap into every fill.

For someone holding a leveraged position, that is where the damage lives. The liquidation engine does not pause because the cash market paused. The mark price keeps getting computed from a book that has just lost its deepest participant and its hedging channel at the same moment, and margin is measured against that mark, continuously. A position can be closed out at a price the actual shares never traded at, because during the halt the actual shares did not trade at all.

Funding is the quieter half. A perpetual holds its tether by paying one side of the market every eight hours, and that clock runs on the venue's calendar, not Korea's. It runs through the halt. It runs through the Korean night, and through a Seoul weekend with the KRX shut for two days, priced off a book whose participants are quoting a company nobody can currently buy or sell where it actually lists.

Half the price-forming machinery is switched off. The cost of holding does not notice.

The 1987 crash is the proof, running in the opposite direction. As the market broke, the Chicago Board Options Exchange suspended its stock-index derivative products at 11:45 a.m. and the CME followed at 12:15 p.m., and the Federal Reserve Bank of Chicago recorded why: although the NYSE was officially open, more than 20 percent of the underlying stocks were not trading. The derivatives venues shut themselves down because they could not price against a stalled reference.

In 2026 the offshore venue simply did not.

Which venue could still fill you, and who was allowed to use it

"So which book would have filled you?" Eunha asks. If the halt only reaches one venue, and another venue is still quoting Samsung Electronics, then the practical question was never whether the market was open. It was who could actually reach the book that stayed open.

Binance restricted the Samsung, SK Hynix and Hyundai perpetuals from South Korean users, a compliance decision driven by the Financial Services Commission's licensing regime for virtual asset service providers. It is not a technical limit and not a liquidity problem. The contracts exist, the book is deep, the screen is live, and a Korean account is not permitted to touch it.

Read that against the halt and the result is stark. The people with the most exposure to Samsung Electronics on 28 July, Korean retail holding actual shares in the market where the breaker fired, were locked out of both books at once. The primary market was closed by rule. The synthetic market was closed to them by nationality. Protection was domestic and so was the exclusion, and both landed on the same people.

Everyone else kept trading their shares.

Two of those rows describe the same contract. The only variable between them is where the account holder lives. A halt is a venue rule; access is a jurisdiction rule; and they are decided by different people for different reasons, which is why they can strand you in the intersection. Kodex has argued the parent version of this before, that the rulebook a venue sits under decides what protects your position rather than the interface you clicked. The equity side of that argument shows up in tokenized stocks and US retail access, where availability turns out to be a property of the venue rather than of the market.

None of this makes circuit breakers a bad idea. They exist because 1987 happened, and pausing so information can reach participants is a defensible answer to a disorderly market. The honest version is just narrower. Inside its boundary the rule does what it was built to do. Outside that boundary it does nothing at all, and your exposure may be sitting outside.

When the derivatives venue is the one that halts

The split does not always run in the same direction, and assuming it does is its own expensive mistake.

On 7 April 2025, tariff news tore through Asian markets. Nikkei 225 futures hit their lower price limit within minutes of the open, and the Osaka Exchange, which operates Japan's derivatives market, paused stock futures for ten minutes at 8:45 a.m. Tokyo time. That venue publishes its own price limits and circuit breaker rules, in its own rulebook, on its own schedule. It halted on its own trigger.

So the pattern is not "old markets stop, new markets run." Halts attach to venues and to products, and coordination between them is a choice each operator makes alone. Sometimes the cash market stops and the derivative keeps quoting. Sometimes the derivative stops first.

The chain breaks at either end. It also breaks where there is no cash equity within a thousand miles: an L2 with a sequencer escape hatch poses the same structural question in a different domain. When the operator stops, does a path out still exist?

Eunha's read is that the asset was never the unit of analysis. The venue was.

Can you sell during a trading halt?

On the halted venue, no. Orders can be entered and they queue, they do not execute, and they sit there absorbing whatever happens to the price in the meantime. You can cancel them. In markets where listed options on the name are still open, positions can sometimes be exercised, though the options venue frequently halts alongside the stock for the hedging reason above.

The reopening is where queued orders get their answer, and it is rarely the answer they were entered for. Trading resumes through an auction that collects everything accumulated during the pause and clears it at one price. So a market order placed the moment the halt hit executes into wherever twenty minutes of one-sided pressure moved the book. The sale happens, at a level set by the thing you were trying to escape.

Off the halted venue, the answer stops being about the rule and starts being about you. Is the same exposure listed somewhere the halt does not reach? Are you permitted to trade there? Is there depth on that book at 3 a.m. local time, or a two-sided quote so wide that exiting costs more than the gap you were trying to avoid? And if the position is a wrapper rather than the share itself, does the venue keep marking and liquidating while the reference market is dark, which is the case for every perpetual described here?

Eunha treats those four as a pre-trade checklist, not a crisis procedure, because a crisis is the one moment they cannot be run. Each one is a lookup rather than a forecast, and all four are cheaper to answer than the position they protect.

Every one is answerable on a quiet afternoon. None is answerable in the twenty minutes when it matters. So the habit worth building is not reacting faster during a halt. It is knowing, before any of it starts, which book can fill you and whether your account is allowed to stand on it.

Somewhere inside those twenty minutes, a Korean retail holder learned that both of their exits belonged to someone else. One closed by rule, the other by passport.

Take a position in one of the 32 tokenized stocks in the Kodex simulator on a Friday afternoon. Hold it through the weekend, while the exchange behind that share sits shut and nothing sets the price the way it does midweek. Simulated capital, real habit: you find out where your exit lives before the session you need it.

Start free →