Reading a 13F without fooling yourself

A 13F is a quarterly report of long US equity positions filed by institutional managers over a size threshold. It is one of the few genuine windows into what large investors own, and it is also one of the most consistently over-read documents in public markets.

What it does not contain

• Short positions. None of them, ever. A manager can be net short a name that appears as a long holding.
• Non-US listings, most derivatives, cash, bonds and private positions.
• Anything that happened after the quarter end.
• Conviction. Position size relative to the book is inferable; why it is there is not.

The lag is longer than people think

Filings are due 45 days after quarter end. So on the day a 13F is published, the oldest position in it is four and a half months old and the newest is six weeks old. A manager who bought in January, published in mid-February and sold in March files nothing further until May. Treating a 13F as a current portfolio is the single most common error, and it is the one that costs money.

Three parsing mistakes that produce wrong numbers

The same security appears on several rows

A manager running multiple portfolios reports each separately. Berkshire's information table lists some names four times over. Read the first row and you report a fraction of the real position, and every quarter-over-quarter comparison built on it is noise with a confident label. Aggregate by CUSIP.

Not every row is shares

The share or principal amount field carries a type. SH is shares. PRN is a bond's principal amount. Summing the two is a category error that inflates a position out of nothing.

The staleness test belongs on the latest filing, not on every filing

If you filter out filings older than some cutoff, you discard the prior quarter, which is the document a change comparison needs. Every position then reads as unchanged. Age-filter the manager's most recent filing to decide whether to trust them at all, then use their history freely.

How to read one usefully

Count managers, not dollars. Berkshire's book dwarfs almost everyone else's, so weighting by value makes one filer the entire signal and everyone else decorative. A count of how many tracked managers added, held, trimmed or exited is a far more stable measure of what a set of investors thinks.

Use a deadband. Without one, corporate actions and rounding classify nearly every position as changed. Two per cent is a reasonable floor for calling a move material.