Korea Now Has Two Closing Prices: Why Your Data Feed May Show the Wrong One

On 14 September 2026 a Korean chip-equipment maker closed at two different prices. The Korea Exchange settled it at $31.53 (₩43,400). A widely used market-data library, queried four days later, returns $31.31 (₩43,100) for the same stock on the same date. Neither number is a typo, and neither feed will tell you which one you are holding. This is what changed in Korea’s end-of-day data when a four-hour evening session opened — and what it means for every collateral ratio, index weight and backtest that reads it.

🕓 What Are Korea’s Two Closing Prices?

A Korean stock has one official closing price — the settled close struck in the 15:30 KST closing auction — and, since 14 September 2026, a second, unofficial last-traded price formed when the 16:00–20:00 after-market ends at 20:00. Both are real prints in real sessions. Only the first is the exchange’s closing price.

The mechanics of the new session are covered in full in our guide to what actually changed on 14 September: the old 16:00–18:00 batch auction was abolished, and a continuous order-by-order session — the same matching engine as the regular market — now runs to 20:00. This article is about the consequence nobody labelled: that session prints prices after the close, and data vendors have to decide what to do with them.

Day one was not a trickle. The KRX after-market turned over 72.24 million shares worth ₩1,817.7bn ($1.32bn) in its first session, while Nextrade’s competing after-market in the same window traded 12.17 million shares worth ₩1,789.6bn ($1.30bn) — KRX figures as reported by Korea JoongAng Daily and Asia Economy, 14–15 Sep 2026. Retail investors accounted for 93.0% of that KRX after-market flow (Source: KRX tally as reported by Money Today and Herald Business, 14 Sep 2026). So the price that may be sitting in your “close” field was set by a thin, overwhelmingly retail evening tape.

📊 Did Data Feeds Actually Change on September 14?

Yes, and the break is clean enough to date to the session. We took 60 Korean stocks — the 30 largest by market capitalisation plus 30 mid- and small-caps — and compared, for each of eight consecutive trading sessions, the KRX Open API settled close against the end-of-day close returned by FinanceDataReader, a widely used open-source market-data library. Both series were pulled on 18 September 2026, so neither is a stale snapshot.

Four sessions of perfect agreement, then four sessions of near-total disagreement, with the break falling on the launch date. Across the four post-launch sessions the gap is usually small — median 0.56%, with 54 of 221 divergent stock-days at 1% or more — but the tail is not small. The largest single-day gap in the sample was 3.00%, and outside the sample we found Bumhan Fuel Cell closing on 14 September at $18.23 (₩25,100) on the exchange against $16.96 (₩23,350) in the feed: a 6.97% spread between two numbers both labelled “close.”

One detail deserves attention because it complicates the obvious story: the volume field is identical on every stock-day, before and after launch. If the feed were simply consolidating the regular and evening sessions into one bar, its volume would have risen. It did not. The price fields moved and the volume field did not — which is a property of the record, not something we can explain from outside.

🔁 Does It Change the Number, or the Direction?

Both — and the direction flips often enough to break a screen. A percentage change is a ratio of two closes, so an error in each end compounds. On 25 of 180 stock-days (14%) in the three sessions after launch, the daily move computed inside the feed carried the opposite sign to the change the exchange published.

These are not obscure names. Hyundai Motor, Hyundai Mobis, SK hynix and Samsung Electro-Mechanics are among the most widely held stocks in Korea, and on 15 September 2026 a feed-driven screen would have sorted every one of them onto the wrong side of zero. A momentum filter, a “top decliners” alert, a stop-loss rule keyed off the daily change — each reads the sign before it reads the magnitude.

Scale matters too. SK hynix closed on 17 September 2026 at $1,267.62 (₩1,745,000) on the exchange and $1,282.87 (₩1,766,000) in the feed. Against 730,492,365 listed shares (Source: KRX Open API, 17 Sep 2026) that 1.20% difference is $11.1bn of market capitalisation — the gap between two numbers that both call themselves the closing price.

🏦 Which Close Is the Official One?

The 15:30 settled close — and you can prove it from the exchange’s own arithmetic rather than taking anyone’s word for it. The exchange publishes a daily percentage change alongside each close. If the evening price were the official close, the next day’s change would be measured from it. It is not.

Four for four, to the basis point. Bumhan Fuel Cell is the clearest case: the exchange records the stock as unchanged on 15 September, while a feed anchored to its own 14 September close would show it up 7.49%.

This matters beyond cosmetics, because the settled close is load-bearing in Korean market plumbing. It is the reference price (기준가) from which the next session’s ±30% daily price limit is measured — a point we set out in our 2026 Korea market calendar and trading-hours guide. It is also the price at which the 15:40–16:00 closing-price session (시간외종가매매) executes every trade. Feed an evening price into either and the error propagates silently.

📐 Why the Two Closes Can Now Drift Much Further Apart

Because the evening session lost the price collar that used to tether it to the close. That collar is visible only in the deleted text of the old regulation — a paragraph we decoded in English just over two weeks before launch, in our guide to the new session. What follows is the other half: what that deletion did to the data.

The abolished session, 시간외단일가매매, was governed by Article 34-2(3) of the KOSPI Business Regulation (유가증권시장 업무규정). That paragraph required an order price to satisfy both of two conditions: item 1, “a price within 10% above and 10% below the day’s closing price” (당일 종가를 기준으로 10% 높은 가격과 10% 낮은 가격 이내의 가격); and item 2, a price within the day’s limit-up and limit-down. The June 2026 amendment deletes the article outright — 제34조의2를 삭제한다 — and inserts Article 34-4 for the new continuous after-market. Article 34-4 has two paragraphs, and neither of them sets a price range. It specifies the matching method and delegates the rest to the enforcement rules; the words 가격제한폭, 상한가 and 하한가 appear nowhere in it. (Source: KRX rule pre-announcement, KOSPI Business Regulation partial amendment, 29 Jun 2026, effective 14 Sep 2026 — published on the KRX legal portal.)

Be precise about what this does and does not change. The joint ±30% daily band, measured from the previous settled close and shared across the regular and evening sessions, still applies — it did before, through item 2. What disappeared is the additional ±10% collar around the day’s close. So the evening band did not tighten; it widened, to whatever room is left inside the day’s limits.

One brake did arrive in the same amendment, and it matters for reading the rest of this article. The pre-announcement’s summary of principal contents states that the volatility interruption (VI) — a brief single-stock cooling-off pause — is extended to the after-market, on the reasoning that alongside the daily price limit it is the per-stock price-stabilisation device for continuous matching; it can fire even before the session has printed a price, in which case it references the regular-session close. So the evening is not unbraked. The anchor the deleted ±10% collar used is still doing work — as a pause rather than a price cap. How that pause behaves is set out in our guide to the new session, under “What Stops a Runaway Price at 7pm?” (Source: KRX rule pre-announcement, KOSPI Business Regulation partial amendment, 29 Jun 2026, summary of principal contents.)

The practical consequence still stands. Under the old batch auction the two end-of-day prices could never be more than 10% apart, because the rule said so. That hard ceiling is gone: divergence is now bounded only by VI interruptions and whatever room is left inside the day’s band, which on a quiet-open stock is most of ±30%. The 6.97% Bumhan Fuel Cell gap on day one would have been legal under the old rule with room to spare — but a gap wider than 10% was impossible by rule, and it no longer is.

🔎 How Can You Tell Which Close Your Feed Is Showing?

Run one arithmetic check: take any liquid Korean stock, divide your feed’s close by your feed’s prior close, and compare the result to the percentage change KRX publishes for that session. If they disagree, your feed is not carrying the settled close. This takes under a minute and needs no vendor documentation.

Three further diagnostics, in order of how quickly they fire:

• Close outside the regular-session range. Compare your row’s close against the high and low the exchange published for that session — not against the high and low sitting in your own row. On that test, 6 of 60 sample stocks fired on 14 September 2026, and 12 of 240 stock-days across the four post-launch sessions. The distinction is the whole point: the library we tested builds each row to be internally self-consistent, so checking the close against its own high and low finds nothing — that version of the test did not fire once in the 479 stock-days we examined across all eight sessions. Our own error, described below, was the rarer variant: the row had already mixed two sources, so it failed even against its own low.
• Low below the regular low, high above the regular high. The range fields move too, not just the close. Techwing’s feed row for 14 September 2026 shows a low of $31.16 (₩42,900) against a regular-session low of $31.53 (₩43,400).
• Prices move but volume does not. In our sample the volume field was identical to the exchange’s on every single stock-day. A record whose price fields changed on 14 September while its volume field did not is internally inconsistent, whatever the cause.

Who needs to care: anyone whose process reads a Korean daily close. That includes margin and collateral ratios, index and ETF weightings, net asset values, factor screens, stop-loss and trailing rules, and every backtest whose price history was refreshed after 14 September 2026. None of these fail loudly. They just quietly use a different number than the exchange did.

🪞 We Walked Into This Ourselves

On 15 September 2026 we recalculated a collateral ratio for Techwing using ₩43,100 and called it the settled close. It was the after-market price. The correct figure was ₩43,400.

The error was caught inside our own drafting process, before anything was published, when a second pass pulled the exchange’s Open API directly and got a different number. What made it catchable was the exchange’s own arithmetic: it reported Techwing down ₩2,950 that session at −6.36%, and ₩46,350 − ₩2,950 = ₩43,400, not ₩43,100. The ratios we would have published were wrong by less than one percentage point and the conclusion would have survived. That is luck, not method.

The structural version of the same failure was in our stored price history, where 422 rows for 14 September 2026 contradicted their own OHLC — 154 with a close below the row’s low, 268 with a close above the row’s high. The two trading days before launch had zero such rows. The cause was mundane: our collector fills the newest session from a fast provisional source and overwrites it later with the official exchange values, and for a few hours the provisional close and the official high–low sat in the same row. It self-corrected once the official values landed; the diagnostic that surfaced it was simply close < low OR close > high.

We are describing this because the diagnostic transfers and the embarrassment does not matter. If you keep Korean price history in a database fed by more than one source, that one-line check is worth running today. (Techwing is the same company we examined in Techwing: Why a 173% Profit Jump Ended in a Loss — a reminder that a stock can be well understood fundamentally and still be mispriced in your own database.)

📚 Lingo Check

🎯 Why It Matters for K-Export Stars

This site exists because Korean primary sources are written in Korean and Western coverage translates the press release rather than reading the rule. The deleted ±10% collar in Article 34-2 is a clean example: it is one paragraph of struck-through text in a Korean pre-announcement PDF, and it is the reason the two closes can now drift further apart than the old rule permitted. We decoded that paragraph in English just over two weeks before the session opened, in our guide to what changed on 14 September. This article is the other half of the job — what the change did to the data, which is the part no rulebook tells you.

The measurement side is the same argument in a different register. We maintain daily settled prices for the full KOSPI/KOSDAQ universe from the Korea Exchange‘s Open API, so we could date the break to a session rather than speculate about it — and we found it because it bit us first. For investors reaching Korea through a data vendor rather than a Korean brokerage account, this is the kind of gap between the official record and the delivered record that no one has an incentive to flag.

If you are working out which venue your stock even trades in during those evening hours, our companion pieces cover the structure: Nextrade Explained for how Korea’s alternative trading system works, and Nextrade Eligible Stocks for why a given name may be absent from it.

Disclaimer: This article is for informational and educational purposes only and does not constitute investment advice, a recommendation, or an offer to buy or sell any security. Figures are point-in-time as dated and may since have changed; all won-to-dollar conversions use ₩1,376.6/$, the rate our price history records for 16 September 2026. Do your own research and consult a licensed financial adviser before making investment decisions.