Taiwan Holder Brackets — a bracket is a definition, not our opinion
The standard misreading is “the big holders went up, so somebody is accumulating”. Three unjustified leaps hide in that sentence. First, the unit is an account, not a person: one person may hold several accounts, and an account may hold on behalf of others. Second, “large holder” is only the name of a size bracket; it describes no identity — the band may contain founders, a passive fund, or one individual who happens to concentrate holdings in a single account. Third, the series is weekly, so anything that moved in and out within the week is invisible.
What the statistic actually counts
Each week we return, per size bracket, the number of accounts, total shares held and share of outstanding, together with the data week and the bracket boundaries. The unit is an account — not a person, not an investor, not an institution. One person may split holdings across several accounts and drop out of a higher band, and one account may hold on behalf of many beneficiaries. A change in account count is therefore not a change in the number of holders.
The boundaries are the published definition, not our line
Bracket boundaries come from the public statistic itself. We copy them verbatim and neither merge, re-cut nor invent a “large holder” threshold of our own; changing them would break comparability with the historical series and with every other implementation. Grouping several brackets together and calling the result “large holders” is likewise a choice, not a property of the data, so we return every bracket as published and let you decide.
Weekly limits, and no forced alignment
One observation per week means a buy and a sell inside the same week cancel and look like nothing happened. Drawing it daily would require inventing four unobserved days, so we do not interpolate, resample or force alignment with daily series; comparisons happen only on dates where both have observations. Weeks are gated at publication time, and a missing week is marked undetermined rather than zero-filled or carried forward.
No chip score, no inference about who is trading
Compressing bracket movements into one number needs weights with no basis, so there is no chip score and no main force direction field here — those words appear only to say we do not provide them, because nobody could recompute such a construct from the public statistic. A shift in the distribution is also insufficient to identify who bought or sold: we infer no institutional activity, issue no signal, publish no ranking and no target price. Use it to start asking which bracket moved and whether accounts and shares moved the same way.