What Korean stocks did after each filing type 2026.08.07
Where this stands: of the 18 cells carrying a number, 15 have a 95% interval spanning zero. Those are not distinguishable from zero.
Every DART filing this pipeline has collected is joined to that company's own daily closes. The baseline is the first trading-day close on or after the receipt date; returns are then measured at +1, +5 and +20 trading days — rows forward in that stock's own series, not a calendar offset, because Korean market closures are irregular. From each return the stock's own index (KOSPI or KOSDAQ) over the identical window is subtracted. What you see is the median market-adjusted return per filing type across 706 filings, as of 2026.08.07. The largest sample, Supply contract (94 filings), has a five-day median of +0.79% (95% CI -1.80~+2.14, spans 0), with 52% of cases beating the index. This is a record of what happened, not a claim that filings move prices and not a statement about what comes next.
Machine-readable: disclosure_impact.json · summary only (smaller) · Hugging Face dataset
| Filing type | baseline day | +1 trading day | +5 trading days | +20 trading days |
|---|---|---|---|---|
| Supply contract | -0.31% (spans 0) · 95% CI -0.9~+0.7 · beat index 47% · n=153 | -0.26% (spans 0) · 95% CI -1.2~+0.4 · beat index 44% · n=153 | +0.79% (spans 0) · 95% CI -1.8~+2.1 · beat index 52% · n=94 | not elapsed yet |
| Preliminary earnings (consolidated) | +0.78% (spans 0) · 95% CI -0.5~+1.9 · beat index 56% · n=153 | -0.22% (spans 0) · 95% CI -1.4~+0.4 · beat index 48% · n=153 | +1.69% (spans 0) · 95% CI -1.5~+5.5 · beat index 55% · n=71 | not elapsed yet |
| Preliminary earnings (separate) | +0.78% (spans 0) · 95% CI -0.5~+1.6 · beat index 57% · n=88 | -2.22% · 95% CI -3.4~-0.7 · beat index 33% · n=88 | +1.29% (spans 0) · 95% CI -1.0~+3.8 · beat index 56% · n=32 | not elapsed yet |
| Dividend decision | -0.30% (spans 0) · 95% CI -1.3~+0.8 · beat index 45% · n=49 | -0.22% (spans 0) · 95% CI -1.3~+2.8 · beat index 47% · n=49 | +2.28% · 95% CI +0.2~+4.7 · beat index 69% · n=39 | not elapsed yet |
| Largest shareholder change | -0.62% (spans 0) · 95% CI -2.4~+2.6 · beat index 45% · n=29 | +1.11% (spans 0) · 95% CI -0.7~+3.4 · beat index 62% · n=29 | withheld, n=17 | not elapsed yet |
| Treasury stock trust contract | +1.88% · 95% CI +0.3~+4.1 · beat index 77% · n=22 | +2.05% (spans 0) · 95% CI -0.2~+6.6 · beat index 73% · n=22 | withheld, n=15 | not elapsed yet |
| Periodic financial report | +0.67% (spans 0) · 95% CI -2.0~+2.6 · beat index 64% · n=22 | -0.46% (spans 0) · 95% CI -2.3~+1.3 · beat index 46% · n=22 | withheld, n=10 | not elapsed yet |
| Paid-in capital increase | withheld by design, n=19 | withheld by design, n=19 | withheld by design, n=12 | not elapsed yet |
| Treasury stock acquisition | withheld, n=17 | withheld, n=17 | withheld, n=12 | not elapsed yet |
| Preliminary earnings | withheld, n=17 | withheld, n=17 | withheld, n=11 | not elapsed yet |
| New facility investment | withheld, n=15 | withheld, n=15 | withheld, n=11 | not elapsed yet |
| Treasury stock disposal | withheld, n=15 | withheld, n=15 | withheld, n=9 | not elapsed yet |
| Treasury stock trust termination | withheld, n=12 | withheld, n=12 | withheld, n=8 | not elapsed yet |
| Convertible bond issuance | withheld, n=12 | withheld, n=12 | withheld, n=9 | not elapsed yet |
| Merger | withheld, n=10 | withheld, n=10 | withheld, n=5 | not elapsed yet |
| Subsidiary paid-in capital increase | withheld, n=9 | withheld, n=9 | withheld, n=6 | not elapsed yet |
| Reverse stock split | withheld by design, n=5 | withheld by design, n=5 | withheld by design, n=4 | not elapsed yet |
| Paid-in and bonus issue | withheld by design, n=3 | withheld by design, n=3 | withheld by design, n=1 | not elapsed yet |
| Exchangeable bond issuance | withheld, n=2 | withheld, n=2 | withheld, n=1 | not elapsed yet |
| Bonus issue | withheld by design, n=2 | withheld by design, n=2 | withheld by design, n=2 | not elapsed yet |
| Material change in profit/loss structure | withheld, n=1 | withheld, n=1 | not elapsed yet | not elapsed yet |
| Bond with warrants issuance | withheld, n=1 | withheld, n=1 | withheld, n=1 | not elapsed yet |
The +20-day columns are empty on purpose. Filing collection started 2026-07-20, so that many trading days have not elapsed for any filing yet. It is time that is missing, not sample — the columns fill themselves. They are left visible rather than hidden.
Withheld by design — Paid-in capital increase, Reverse stock split, Paid-in and bonus issue, Bonus issue. These filing types break the price series itself: an ex-rights date or a share consolidation resets the quoted price mechanically. Closes here are not adjusted for corporate actions, so if that date falls inside the window the number measures the break, not a market reaction. No sample size will unlock these. The individual values stay in the JSON, flagged price_break — not hidden, just not averaged.
The columns do not cover the same filings. baseline day 2026-07-20 to 2026-08-04 · +1d 2026-07-20 to 2026-08-04 · +5d 2026-07-20 to 2026-07-29. A longer horizon excludes recent filings that have not had time to elapse, so its sample clusters earlier. If the market behaved differently across those windows, reading two columns side by side as “what happened N days later” is wrong — they are not the same set of events.
The baseline day column is the move from the previous trading day's close into the baseline close. DART accepts filings during the session and after it, and the receipt time is not in the public data, so the two cannot be separated. A filing made mid-session is already partly reflected in that day's close, and that part disappears into the baseline. This column exists to make the missing piece visible, not to remove it.
Of the 18 cells with a number, 15 have a 95% interval spanning zero — those are not distinguishable from zero. The intervals narrow as the sample grows. A bold figure is not automatically a finding.
About the baseline-day (h0) column as a whole. Receipt times were resolved for the 706 filings in this table (0 unknown). The whole day follows the filing only when it arrived before the open, and that is 22 of them (3.1%). 376 arrived intraday, so part of the day precedes them; 308 arrived after the close, so the whole day precedes them. h0 is not an estimate of a reaction — it is a diagnostic column that makes the window swallowed by the baseline visible.
About the cell that clears zero — Preliminary earnings (separate) · +1 trading day. The interval may move as the sample grows. n=88, among the smallest samples here. Do not read this cell as a finding.
About the cell that clears zero — Dividend decision · +5 trading days. The interval may move as the sample grows. n=39, among the smallest samples here. Do not read this cell as a finding.
About the cell that clears zero — Treasury stock trust contract · baseline day. Receipt times for this type break down as 0 before the open · 11 intraday · 11 after the close. A filing received after the close is preceded by the entire day’s move — that is not a reaction to it. An intraday filing is preceded by part of the day. Only a filing before the open has the whole day after it. n=22, among the smallest samples here. Do not read this cell as a finding.
One exception runs the other way: Treasury stock trust contract · +1 trading day has a median interval spanning zero, but its beat rate of 73% has an interval of 52–87%, which excludes 50%. The size is not credible; the direction is not a coin flip. The asterisk counts medians only.
The 95% CI shows how far the median moves with the sample (order-statistic based, no randomness, so it is identical on every publish). An interval spanning zero means that median is not distinguishable from zero — the table says so where that happens. Publishing a bold number without its interval would be the dishonest option.
Each cell is a median, not a mean — one extreme move would drag a mean around. “Beat index” is the share of cases that rose more, or fell less, than the stock's own index over the same window.
What "market-adjusted" means here — the stock's return minus its index's return, subtracted directly, which assumes beta = 1. Individual market sensitivity and size or sector effects are not removed. A KOSDAQ small cap is measured against all of KOSDAQ, so part of what is called market-adjusted here may be a size factor.
How it is computed
- Baseline = the first trading-day close on or after the filing receipt date.
- Return is taken to the close 1, 5 and 20 trading days later — trading days, not calendar days.
- The same-window return of that stock's own index (KOSPI or KOSDAQ) is subtracted. Rising 3% while the market rose 5% is underperformance, and should read as such.
- When one company files several documents on the same day, the baseline and the
price path are identical across them. Those count once per filing type.
Counting each receipt number separately inflates the sample and lets a single company's
day pull the median. The receipt count is published alongside as
n_filings. - Results are grouped by filing type and reduced to a median plus the share that beat the index.
- Amended filings are dropped — they duplicate the original. Filings whose horizon has not elapsed are excluded from that horizon only, which is why each cell carries its own n.
- Fewer than 20 observations gets no number at all. A median over a handful of cases turns coincidence into a statistic.
Questions you should ask
Did the filing cause this?
No, and nothing here claims it did. Earnings, sector moves and the market itself sit inside the same window. This is a record of what accompanied each filing type, nothing more.
Why the median instead of the mean?
One stock moving 40% would drag a mean far from the typical case. The median shows the middle case, which is what “what usually happened” should mean.
Is the stock's own index the right benchmark?
It is the honest available one, and it has a known weakness: a small-cap is compared against all of KOSDAQ, so part of what is called market-adjusted may be a size factor. Matching on size or sector would be better, but the buckets get thin at this sample size and no sector classification exists in this data. That question is open for discussion.
Can I check an individual case?
Yes. Per-filing values in the JSON are keyed by DART receipt number, so any single case can be joined back to the original document and verified.
본 페이지는 공개 데이터의 사후 집계이며 투자 권유가 아니고, 특정 종목의 매수·매도를 권하지 않습니다.
What is not here
- No real-time or intraday prices. Closing prices only, published the next business day (T+1) by the government source. If you need live quotes, this is the wrong dataset.
- No PER, PBR, target prices or analyst ratings. Those are brokerage-derived; this project publishes only public-sector data.
- No investor-type flows (foreign / institutional / retail).
- No sector classification yet.
- Filing explanations are written in Korean. Field names and structure are English-friendly; the prose is not translated.
This site publishes public data as fact. It is not investment advice and does not recommend buying or selling any security. Source: Financial Services Commission open data portal (prices) and DART, the Financial Supervisory Service filing system.
본 페이지는 공공데이터 사실 제공이며 투자 권유가 아닙니다. 특정 종목의 매수·매도를 권하지 않습니다.