Writing / Market structure
KOSPI fell 8%. What happens to your orders when trading stops?
The 15 October KOSPI halt explains why order permissions, liquidity, and reopening mechanics matter to an automated trading system.
At 12:38:12 p.m. in Seoul on 15 October 2026, the KOSPI reached 5,528.05, about 8% below its previous close. The Korea Exchange activated a first-stage circuit breaker, interrupting trading in its KOSPI market for twenty minutes. Those figures appear in the exchange halt notice reproduced through the Koscom disclosure feed.
For someone with an outstanding order, the percentage decline is only part of the problem. The next decisions depend on what the exchange still permits, what has happened to that order, and how prices will be established when trading returns.
One detail matters immediately: during a first-stage halt, new exchange orders are blocked, but existing orders can still be cancelled. KRX’s circuit-breaker explanation
That makes this a useful case for the kind of order-book-first analysis I want to publish. My ongoing work on autonomous trading systems starts with market mechanics. A system reading bids and offers also needs to understand when those orders can actually trade.
What the 8% trigger measures
The first-stage condition is a decline of at least 8% from the previous trading session’s closing index, maintained for one minute. It does not mean an 8% decline must occur within a single minute. The distinction is explicit in the halt notice.
The numbers on 15 October were:
| Reference | Index level |
|---|---|
| Previous close | 6,008.81 |
| Previous close multiplied by 0.92 | 5,528.1052 |
| Reported activation level | 5,528.05 |
The middle row is my calculation from the published previous close. It explains the threshold; it is not a replacement for the exchange’s official calculation or trading-status messages.
The interruption changes the permitted actions in the market. It does not supply a new valuation for the companies in the index, or identify which sellers were acting on information, leverage, or a change in risk appetite.
Three different moments in the trading day
The day’s notices show a sequence of distinct interventions. All times below are Korea Standard Time, UTC+9.
| Time | Announced event | What it means |
|---|---|---|
| 09:07:42 | Sell-side sidecar | A five-minute suspension of the effectiveness of relevant program sell orders |
| 12:38:12 | First-stage circuit breaker | Twenty-minute interruption of KRX KOSPI-market trading |
| 12:58:12 | Circuit breaker lifted | Start of a scheduled ten-minute order-collection period for single-price matching |
Sources: reproduced exchange notices for the sidecar, circuit breaker, and resumption.
The morning sidecar used a designated KOSPI 200 futures contract as its trigger. Its price had fallen 5.27% from its reference price; the notice specifies a decline of at least 5%, sustained for a minute.
Its scope was narrower than the later circuit breaker. It affected program sell orders, rather than stopping the entire cash market. “Program trading” is a regulatory category here; it does not mean every order submitted by software.
These notices also show why a twenty-minute countdown is an incomplete model of reopening. The announced 12:58:12 resumption began order collection. The announced process returned to continuous trading after the auction.
Trading stops; orders still need managing
KRX’s first-stage rules create an asymmetry: participants cannot submit fresh exchange orders during the interruption, but they can withdraw existing ones. That permission creates the possibility of a changing resting book while executions are paused.
It does not tell us how many orders were actually cancelled on 15 October.
It also does not mean the halt automatically wiped every outstanding order. An order owner still needs an authoritative answer about the remaining quantity and whether a cancellation was accepted.
That distinction matters around any interruption. The FIX order-state specification, pages 5 and 13–14, distinguishes a pending cancellation from a completed one and documents cases where executions occur before cancellation succeeds.
Those are general order-handling scenarios, not evidence that trades occurred during this KRX halt. A fill reported late may belong to an earlier execution. The practical lesson is to reconcile messages and quantities rather than treat an instruction sent by a bot as an instruction already completed.
The cancellation permission discussed here concerns the first-stage interruption. Later stages have different conditions and permissions; an implementation needs the rules for the actual market state.
What a liquidity snapshot can—and cannot—tell you
KOSPI is an index, not a single executable order book. To investigate execution, we would need to specify an instrument: a constituent stock, an ETF, or a futures contract, together with its trading venue.
For a chosen instrument, a depth snapshot answers a conditional question: what quantity is displayed at these prices at this moment?
A simple illustration makes the consequence of changing depth visible. The following quantities and prices are invented. They describe two hypothetical books in continuous trading, not Korean market observations or a simulation of the reopening auction.
| Bid price, in arbitrary currency units | Shares before cancellations | Shares remaining afterward |
|---|---|---|
| 100 | 1,000 | 0 |
| 99 | 2,000 | 1,000 |
| 98 | 7,000 | 7,000 |
| 96 | 2,000 | 2,000 |
Suppose a seller immediately executes 10,000 shares against each snapshot, with no other orders arriving or disappearing.
Against the first book, the average sale price is 98.40. Against the second, it is 97.70, because the order reaches the bids at 96. Relative to the original best bid of 100, the price concession increases from 1.6% to 2.3%, before fees.
The example isolates one mechanism: removing quantity near the top of a book can worsen the execution available for a fixed order size. It does not predict the next price, explain why the orders were cancelled, or show what happened in Korea.
Displayed depth also has limitations in the other direction. Hidden or replenishing orders can make visible quantity an incomplete measure of available liquidity. BIS discusses both short-lived quotes and iceberg orders in its analysis of electronic fixed-income trading. Those concepts are useful here; its bond-market findings are not measurements of KOSPI.
Volume answers another question: how much has already traded? In their investigation of 6 May 2010, SEC and CFTC staff documented intense turnover alongside deteriorating buy-side depth. That historical episode illustrates why high volume alone can be a poor indicator of capacity to absorb the next order. SEC/CFTC staff report, printed pages 3–4
For an order-book-first system, the useful unit of analysis is therefore specific: an instrument, an order size, a price tolerance, and a time horizon.
Reopening is an auction process
The 15 October resumption notice describes ten minutes of order collection, single-price matching, and then continuous trading.
During collection, orders accumulate for the auction rather than executing through ordinary continuous matching. Eligible orders are then matched at a single price for each instrument under the auction rules. The index itself does not receive one market-wide execution price.
The practical distinction is between being allowed to submit an order and being able to execute immediately in a continuous market.
Auction indications, where available, need to be interpreted in that context. A displayed indicative value expresses an evolving auction outcome; it should not be treated as an ordinary immediately executable quote. As a separate venue example, Coinbase Exchange’s auction-feed documentation explicitly describes its indicative auction values as nonfirm. KRX’s own rules govern the Korean process.
The nominal ten-minute interval also does not prove the exact timestamp of the first post-halt execution in a particular stock. That requires the relevant exchange records or trade data. I would not label 1:08:12 p.m. as an observed first-trade time merely by adding thirty minutes to the halt.
What this means for an autonomous system
My design takeaway is that a trading system needs to track three separate conditions:
| Condition | What the system needs to establish |
|---|---|
| Market state | Which order actions and matching process are currently permitted? |
| Data integrity | Is the instrument’s book synchronized and sufficiently current for the decision? |
| Order and position state | What remains open, what has filled, and what exposure exists? |
A halt is different from a broken feed. If trading pauses while the feed remains synchronized, the absence of trades does not itself require rebuilding the entire book. Conversely, an “open” market status cannot make a book trustworthy if updates were missed. Exchange data specifications model these separately; CME, for example, documents market-state messages and book-recovery procedures.
For a strategy designed around continuous matching, I would review the transition back to trading in this order:
- Confirm the venue’s current state and permitted actions.
- Reconcile existing orders, cancellation responses, fills, and positions.
- Establish book integrity, recovering missed data where necessary.
- Confirm that the strategy supports the current auction or continuous phase.
- Re-evaluate the signal and execution conditions before submitting a new order.
These are engineering recommendations to test. They are not claims that my project already implements every control, trades Korean equities, or would have performed well during this event.
A pre-halt signal belongs to the conditions that produced it. Waiting for a timer to expire does not establish that those conditions still hold.
What we can conclude from this event
The notices establish the trigger, interruption, and announced reopening process. They do not establish whether cancellations surged, market makers withdrew, forced selling dominated, or the circuit breaker improved liquidity.
A proper investigation of those questions would need timestamped trades, order additions and cancellations, market-state messages, and auction records for selected instruments. It would also need to distinguish conditions around the halt from the wider shock already affecting the market.
For now, the strongest practical conclusion is narrower: an order-book-first system must interpret depth together with market phase, data quality, and outstanding exposure.
The KOSPI halt makes that requirement concrete. At the next interruption, the first useful output from a trading system may be an accurate account of what it can do, what exposure remains, and what it knows about its orders.
This research note uses intraday exchange notices dated 15 October 2026. The order-book example is hypothetical. It contains no reconstruction of actual KOSPI order flow or claim about trading performance.
Get the next market-mechanics research note by email. Subscribe to The Bagholder—free, periodic, and easy to unsubscribe.
Educational research. Not financial advice.