Introducing Price-Move Context Cards for Taiwan Stocks

What a card answers

• How big was the move — a magnitude bucket: 5–7%, 7–9.5%, or ≥9.5% (the ≥9.5% band is the price-limit zone).
• Did it hit the daily price limit — an era-aware flag: ±7% before 1 June 2015, ±10% on/after (method 'approx' until an exact per-day limit table lands).
• How much was excess over the market — the move minus the TAIEX move that day (a −10.23% stock on a −1.78% market is −8.45% excess, not just 'down with the market').
• Was there an official event — a no-official-event flag when no filing/corporate action is on record for that stock-day.

The real numbers (dry run over 266,097 rows)

The candidate build over Taiwan history produced 266,097 big-move card rows. The magnitude split: 5–7% moves were 43.9% (116,915), 7–9.5% were 11.8% (31,339), and ≥9.5% (limit zone) were 17.6% (46,903). A further 26.7% (70,940) were 'relative-only' hits — under a 5% absolute move but ≥4% excess over the market. By track, 55.7% (148,154) qualified on both the absolute and the relative test.

The headline finding: most big moves have no filing

Of the 266,097 rows, 240,681 (90.4%) carried no official event on record for that stock-day. That is the point of the card: a large move is not the same as a news event. If your pipeline assumes every big move maps to a disclosure, this says otherwise nine times out of ten — the card lets you separate event-driven moves from the far larger set that are flow- or market-driven.

Example cards (from the deterministic sample)

move = -10.23%   market = -1.78%   excess = -8.45%   limit-flag = true    event = none
move =  +5.14%   market = +0.24%   excess = +4.90%   limit-flag = false   event = none
move =  +7.00%   market = -1.07%   excess = +8.07%   limit-flag = false   event = none

Honest limits (stated, not hidden)

• Market-relative is blind on some days: the TAIEX reference is null before 2009-01-05 and on data gaps, so those rows carry only the absolute track — they are honestly marked, not guessed.
• The limit flag is method 'approx': it fires within about 0.5% of the era's band; an exact per-day limit (rounding, special sessions) is a later refinement. A single fixed 9.5% flag would miss every pre-2015 7%-limit hit, so the flag is era-corrected.
• The factor evidence behind it is marked unverified: cross-sectional IC for the candidate factors was measured over a sample of 160 liquid ordinary equities across 7,498 trading days (as of 2024-12-31), and is labelled 'unverified' until an out-of-sample check confirms it — the numbers are shown with that caveat, not as settled alpha.

Coming soon

The price-move context dataset is still being built — the numbers above are from a dry run, not a live endpoint. When it lands it will be a derived, insert-only daily dataset you can query per symbol. Until then, the era-aware limit logic it relies on is already usable today through the stock-price-limit-daily dataset.