Home / Event study statistics
Event study statistics
Base rates for common Taiwan stock filing events: for each category, the mean abnormal return over several holding windows after the event historically occurred. CAR and BHAR benchmarks are shown side by side, and flagged when they disagree. Every number comes from this site's event study engine, not hand-picked.
Which filings move prices, and for how long
As of 2026-07-09. Each cell is the mean abnormal return for that category over that holding window, tagged with the benchmark it came from (BHAR is shown where both exist; categories whose benchmarks disagree are flagged in the cards below). Cells in bold with a dagger clear the multiple-testing correction; those that do not are still printed without a mark — a null result is a result, and hiding it would be selective disclosure. A dash means no statistic was produced. Rows are ordered by event count, not by effect size — this is not a ranking.
| Event category | Events | Event day | After 5d | After 20d | After 60d |
|---|---|---|---|---|---|
| Earnings call held | 36,498 | +0.12%BHAR | +0.27%BHAR | +0.77%†BHAR | +2.57%†BHAR |
| Share buyback announced | 3,974 | +2.71%†BHAR | +4.15%†BHAR | +6.37%†BHAR | +9.08%†BHAR |
| Share buyback completed | 3,943 | -0.51%†BHAR | -0.63%†BHAR | +0.68%†BHAR | +2.70%†BHAR |
| Shareholder meeting filings | 1,987 | +0.34%†BHAR | — | +1.63%†BHAR | — |
| Non-routine material information | 1,391 | +0.56%†BHAR | +1.79%†BHAR | +5.40%†BHAR | — |
| Dividend policy announced | 916 | +0.52%†BHAR | +2.28%†BHAR | +5.01%†BHAR | — |
| Trading restriction lifted | 896 | -0.80%†CAR | -0.74%†CAR | +2.75%†BHAR | +5.56%†BHAR |
| Senior management change | 884 | +0.32%†BHAR | +1.62%†BHAR | +4.11%†BHAR | — |
| Other material information | 765 | +0.35%†BHAR | +1.57%†BHAR | +5.90%†BHAR | — |
| Corporate governance filings | 764 | +0.40%†BHAR | +1.82%†BHAR | +3.39%†BHAR | — |
| Investor conference announced | 501 | +0.94%†BHAR | +2.38%†BHAR | — | — |
| Market surveillance notice | 282 | — | -2.57%†CAR | — | — |
| Board resolution | 214 | — | +4.10%†BHAR | +8.54%†BHAR | — |
| Asset acquisition | 186 | +0.66%†BHAR | +1.95%†BHAR | +6.24%†BHAR | — |
| Other pooled events | 162 | — | — | +6.86%†BHAR | — |
| Capital increase | 147 | — | +4.16%†BHAR | +9.47%†BHAR | — |
| Other material matters | 145 | +0.55%†CAR | — | — | — |
| Asset disposal | 114 | — | — | +8.13%†BHAR | — |
| Endorsement and guarantee | 113 | — | +2.43%†BHAR | — | — |
| Merger and acquisition | 94 | — | — | +6.58%†BHAR | — |
| Clarification statement | 91 | +0.99%†BHAR | — | — | — |
| Buyback-related material information | 73 | +1.31%†BHAR | +4.25%†BHAR | — | — |
Dagger: clears the BH-FDR multiple-testing correction (q < 0.1)
All 27 event categories
Earnings call held
Share buyback announced
Share buyback completed
Shareholder meeting filings
Non-routine material information
Dividend policy announced
Trading restriction lifted
Senior management change
Other material information
Corporate governance filings
Investor conference announced
Market surveillance notice
Debt issuance announced
Board resolution
Asset acquisition
Other pooled events
Capital increase
Other material matters
Insider share pledge change
Asset disposal
Endorsement and guarantee
Capital reduction
Insider share pledge released
Merger and acquisition
Clarification statement
Buyback-related material information
Earnings call scheduled
How these numbers are computed
Day 0 is the public-availability date, and entry is the day after
Every event is timestamped to the day the filing could actually be looked up on the official public source, not the day the company resolved on it internally. Entry is then set to the following trading day: assuming a fill during the session the news broke is not realistic for most readers, and returns computed on it are returns nobody could have taken.
Two benchmarks are computed independently, with no cherry-picking
CAR is estimated with a market model and calibrated by a placebo test — event dates are kept, but the tickers are drawn at random from the same day, to see whether the engine manufactures returns from nothing. BHAR is measured against control stocks matched on size and liquidity. Neither path consults the other, and where they disagree both are printed.
Every cell carries a multiple-testing correction
Look at several hundred cells at once and chance alone will surface a batch of significant ones. BH-FDR is applied across all of them to control the false discovery rate. Cells that clear it are bolded and daggered; the rest still print their figure.
Too little sample means no output, not a zero
Where the history is too thin, the engine emits no window at all and the category simply has no row here. A blank means no reliable figure exists, not that the effect is zero — those are very different statements and they do not share a symbol on this site.
Four terms to read this page
Frequently asked questions
Can I trade off this table?
No — we publish no signals and no recommendations. These are historical base rates: how prices moved on average after past events of each kind. An average is not a forecast. The individual events behind any one cell are widely dispersed in both directions, and when an event happens you do not get to know in advance which one you drew. Treat a base rate as where understanding starts, not as a reason to enter.
Why does each cell show only one benchmark?
The cell is tagged with whether the figure came from CAR or BHAR. Where both exist we show BHAR here, because it covers the widest set of windows. To see the two side by side, open the category's detail page, where every window lists CAR and BHAR row by row. Categories whose benchmarks disagree are flagged in the cards below, so you are never shown only the flattering one.
Why print figures that are not significant?
Because withholding them turns into selective disclosure. If only significant cells were printed, every blank would read as 'not tested' rather than 'tested and failed', which is an illusion built out of layout. So everything is printed, cells that clear the correction are marked, and the rest stand as they are. A row that is null throughout is itself the most useful thing to know about that kind of filing.
Why are some categories missing from the table entirely?
Five categories have too little history for the engine to produce any window at all, so they would be a full row of dashes. They remain in the cards below, labelled as still accumulating sample. Existing with insufficient data is not the same as not existing, and we do not quietly drop them from the page.
Is the multiple-testing correction really necessary?
At this scale, yes. The scan covers dozens of categories times four windows times two benchmarks — several hundred cells. On chance alone, that many tests will throw off a batch with p below 0.05. The BH-FDR correction controls exactly that: how many of the results called significant are false positives. An event-study table with no correction looks far better and is far less trustworthy.