Methodology — Manager vs Manager

Transparency is a feature. This page explains what the compare surface measures and why — in plain terms. The precise formulas and numeric conventions are documented in our internal methodology; the summary here is deliberate, not an omission.

Two 13F filers, side by side, answering two separate questions: what do they hold, and how did their reported books move together.

The caveat that governs the whole page. A 13F filing is a quarterly list of long US positions. It has no shorts, no cash, no options and no non-US listings, and it arrives about 45 days late. Everything here is built from those lists. The return curves are hypothetical replications of the reported books, not fund returns — so the correlation describes how two filed books moved, not how two funds did.

Holdings

Both books are read at the latest quarter both filers reported, so you are always comparing one vintage against itself. If that shared quarter is not a filer's own most recent one, the page says so rather than letting you assume otherwise.

Positions are grouped by company, not by ticker line. Two share classes of the same issuer (for example the A and B shares of one company) count once, and the class-ticker spellings that differ between filings and market data are reconciled first. Any group we combined is named on the page, because a combined weight is larger than either class you could look up on its own.

A position's weight is its filed value over that filer's entire filed book — including lines whose identifier we cannot resolve to a company. A book we can only partly resolve therefore reads as partly covered, rather than being quietly rescaled to the part we matched.

The two overlap numbers

Overlap weight answers: how much of one book would you already own by holding the other? It is a dollar statement, and in each shared name it is limited by whichever filer holds less of it.

Cosine answers a different question: how similar are the two books in shape? It ignores size, so a book that is a faithful smaller copy of another scores near the top on cosine while scoring around half on overlap weight. The two together are the story; either alone is half of it.

Neither number is rescaled to just the shared names — doing that would make a sliver of a match read like a whole-book match. When a book is empty or carries no value, the result is reported as undefined rather than as zero, because zero would mean "measured, and they share nothing".

Returns

Both curves come from the same engine the manager pages use: hold each reported book from its quarter-end to the next, snap to the newly reported book, and value it daily. Both sides run under the same convention for when a filed book starts counting — either as filed, or on the real date the filing became public — so the two are never measured on different rules.

Every figure is measured over the overlap of the two filers' histories: the span where both have a curve. Each filer's return and risk figures on this page are computed over that same span, never over one side's longer history, which would put two different windows under one label.

Three views of co-movement are reported: day to day, month to month, and each curve's daily move net of the market's. A rolling correlation over a trailing year shows how the relationship drifted; the earliest stretch, before a full year has accumulated, is left blank rather than filled with a partial figure that would mean something else.

Short overlaps are refused, not shrunk. Under twelve overlapping months, no correlation is shown and the page states why — a few months of overlap produces a number that looks precise and carries no information. The curves and each filer's own statistics are still shown. The same rule applies when one curve is perfectly flat over the window: a flat book and a moving one are not uncorrelated, they are unmeasurable, so the page says so instead of printing a zero.

Gaps are named, never smoothed. A filer that has stopped filing keeps producing a curve from its last reported book, and the page says how far behind that book is. A quarter whose stored book carries fewer positions than were actually filed is flagged with both counts.