Research Lab研究实验室
Research Lab.研究实验室。
Long-form analyses and post-mortems from the desk. Every figure is from our own data on the date shown, every fix it led to is pinned by a test, and each post ends with a way to check it yourself. 来自投研台的长文分析与事后复盘。每个数字都来自我们自己在文中日期的数据，每一处由此而来的修正都有测试锁定，每篇文章结尾都告诉你如何亲自验证。
-
7 min read约 7 分钟
Rotation happens a level down轮动发生在更细的一层
The S&P 500's eleven sectors hide their 25 industry groups. We rebuilt all 25 from the index's own constituents, checked them against the official indices, and found this year's tech rally was two groups, not one sector.标普 500 的 11 个板块掩盖了其下的 25 个行业组。我们用指数自身的成分股重建了全部 25 个行业组，与官方指数逐一核对，发现今年的科技行情其实是两个行业组的行情，而不是整个板块。
-
8 min read约 8 分钟
A forward multiple needs a forward history前瞻倍数，需要前瞻的历史
Ranking today's forward P/E against five years of trailing P/Es marks every growing company cheap — for half of our 22-name test, as cheap as a percentile can go. Rebuilding a forward history fixed the ranking. Here is how, and what it still gets wrong.把今天的前瞻市盈率放进过去五年的滚动市盈率里排名，会让每一家成长型公司都显得便宜——在我们 22 只股票的测试中，一半直接被排到了百分位的最底端。重建一段前瞻历史修正了排名。本文说明做法，以及它仍然存在的偏差。
-
6 min read约 6 分钟
TSM at a P/E of 1.01市盈率 1.01 倍的台积电
A reader spotted TSM trading at one times earnings. It was three incompatible bases from one data feed — dollars per ADR, Taiwan dollars per ADR, and ordinary shares — and every figure they touched came out cheaper than the truth.一位读者发现台积电的市盈率只有 1 倍。原因是同一个数据源给出了三种互不兼容的口径——每份 ADR 的美元价格、每份 ADR 的新台币收益、以及普通股股数——而每一个受影响的数字，都比真实值更“便宜”。
-
6 min read约 6 分钟
Grading the desk without flattering it给投研台打分，而不美化它
Before any hit rate, a ledger: one row per decision, measured from a close the call could actually have traded, counted in sessions rather than days — and deliberately storing no score of its own.在谈命中率之前，先有一本账：每个决策一行，从决策真正可以成交的那个收盘价开始计算，以交易日而非自然日计数——并且刻意不保存任何自评分数。
-
6 min read约 6 分钟
Every look-through number is a floor穿透后的每个数字都是下限
Yahoo discloses a fund's top ten holdings — 37.6% of VOO, 13.0% of VXUS. Grossing the rest up would invent concentration at the moment you are judging it. So every figure is a lower bound, with its coverage beside it.Yahoo 只披露基金的前十大持仓——VOO 的 37.6%、VXUS 的 13.0%。把其余部分按比例放大，就会在你判断集中度的那一刻凭空制造集中度。所以每个数字都是下限，并标明覆盖率。