数据上，香港股市规模约US$6.5 trillion（原文：$6.5 trillion）。在2,780家上市公司中，仅25%有分析师覆盖，约为695家（计算：2,780×0.25）；约三分之二为micro-caps，约1,853家（计算：2,780×2/3），定义为估值低于HK$2 billion（原文：HK$2 billion），折合US$256 million（原文：$256 million）。文中指出在牛市环境下，内部人更易借势进行pump and dump（保留原英文名称），而流动性不足的小型股更易出现flash crashes（保留原英文名称）。
监管层面，自2023年起，香港Securities & Futures Commission要求券商为客户分配Broker-to-Client Assigned Number（BCAN）并将其标记于所有证券指令，以追踪与遏制操纵；但作者引用Webb观点，认为资讯黑箱仍在于控股股东的share pledges（保留原英文名称）披露不足，可能加剧突发暴跌。文中举例2024-04-09 China Tianrui Group Cement Co.因保证金追缴与被迫抛售而几近归零，投资者在事后才得知关键细节。结论指向：技术（含generative AI）可降低维护资料库的门槛，但缺乏政治意志与利他动机使「下一位Webb-Keeper」成为未解命题。
Set on 2026-01-19 03:00 (UTC+8; original: 3:00 AM GMT+8), the column argues that activist investor David Webb, reported to have died “last week” at age 60, filled a rare watchdog role in Hong Kong by using only public information to trace ownership and governance. His core contribution was the free Webb-site database of shareholdings and board compositions, including the 2017 “enigma network” and the list “50 stocks not to own,” highlighting opaque cross-shareholdings among small caps.
The numeric backdrop is a Hong Kong equity market of about US$6.5 trillion (original: “$6.5 trillion”). Out of 2,780 listed firms, only 25% have analyst coverage—about 695 companies (2,780×0.25). Roughly two-thirds are micro-caps—about 1,853 firms (2,780×2/3)—defined as valued below HK$2 billion (original: HK$2 billion), or US$256 million (original: $256 million). The piece links thin liquidity to flash crashes and argues bull markets can mask pump-and-dump behavior by insiders.
Regulation has tightened: since 2023, the Securities & Futures Commission has required brokers to assign a Broker-to-Client Assigned Number (BCAN) and tag it on securities orders to deter manipulation. The author, citing Webb, says BCAN cannot solve information gaps around controlling shareholders’ share pledges, which may amplify sudden collapses. An example is 2024-04-09, when China Tianrui Group Cement Co. lost almost all value after margin calls and forced selling, with key details learned too late. The conclusion is that generative AI may make database upkeep easier, but political will and altruistic incentives remain the binding constraints.