Retail–Mega Divergence — who is accumulating? Computed from official weekly filings
"Big holders quietly buying while retail sells" is one of the most popular narratives in Taiwan investing — usually backed by a screenshot. We compute it: rolling N-week deltas of the mega-holder (1,000+ lot) share vs the retail share, straight from the official weekly holder-distribution filings, flagging divergence when the two move in opposite directions. Open formulas, deterministic, availability-date stamped.
How it is computed
The official filing buckets holders into 15 lot-size brackets weekly. We define 1,000+ lots as mega holders and small brackets as retail, compute rolling N-week (default 12) deltas for both, and flag divergence when directions oppose. Every row carries its availability-date stamp — filings publish the week after settlement, so backtest signals only activate after publication. Honest limit: bracket shifts can come from CB conversion or corporate actions, not just accumulation.
What we deliberately do not do
We do not claim divergence is predictive. Shifts in the top brackets can come from share pledges, trust transfers or odd-lot redistribution after ex-dividend — all indistinguishable from trading in the distribution table.
We also do not ship divergence rankings or alert pushes. Raw bracket shares and weekly deltas are all preserved; you set the threshold, rather than us picking a number that happens to look convincing.