An ownership aggregate is a number people say out loud. A manager owns four percent of the company. Institutional ownership went from sixty-one to sixty-eight percent over the quarter. Someone sizes a position off that, and later someone else has to defend it. Almost always the number is a sum over 13F holding rows, and the sum only means anything if every row in it refers to a distinct slice of the same economic pie.
A meaningful share of rows do not. The same issuer arrives under several identifiers, and unless you have collapsed them deliberately, they all get added. The distortion is not uniform across names, which is the part that hurts. It is largest exactly where the capital structure is complicated, so a screen ranked on institutional ownership ends up ranked partly on share-class structure. That is a hidden factor bet nobody signed off on.
Three rows, one company, one wrong number
Take the plainest case first. Alphabet has two listed lines that appear in filings, the A shares and the C shares. They are separate securities with separate identifiers, and a filer that holds both files two rows. Neither row is wrong. What is wrong is summing the share counts, because the denominator you are about to divide by belongs to one class and the numerator now spans two. Berkshire is the same shape with an extra wrinkle, since the A shares convert into B shares at a ratio fixed in the charter, so the two lines are not merely similar, they are the same claim measured in different units.
Now add a foreign issuer held through a depositary receipt. The receipt is a distinct security with its own identifier, and it represents some number of ordinary shares set by the depositary agreement, which is very often not one. A filer holding the receipt has exposure to the ordinary shares, but the share count in the filing is denominated in receipts. Sum it against ordinary-share rows and you have added apples to a multiple of apples.
Then add the wrappers. A filer holding a sector fund or a broad index product has economic exposure to hundreds of issuers, and none of that appears as a row against those issuers. A filer holding a single-name product has exposure that does appear elsewhere in the fund's own reporting. Count both and you double. Count neither and your aggregate is short by whatever the wrapper channel carries, which for some issuers is now a large number.
The mapping table and the four keys it needs
The fix is a mapping table, maintained deliberately, not a set of string heuristics applied at query time. Ticker matching will not do this. Names will not do this. The table needs four things per security identifier.
- An issuer key that is stable across the identifiers of every class, receipt and successor entity of one economic issuer. Do not use the ticker and do not use the identifier of the largest class, because both change.
- A class label, so you can still see the A and C lines separately when you want to. Normalisation should collapse on demand, not destroy detail.
- A conversion factor into a single reference unit, expressed as reference shares per held share. For an ordinary line it is one. For a depositary receipt it is the ratio from the depositary terms. For a convertible class it is the charter ratio.
- An instrument type, so wrappers, convertible debt and option lines route to different logic instead of into the share sum.
Version the table with effective dates. Ratios change. Depositary programs are restated, classes are collapsed, issuers reorganise. If your table has no time dimension, a ratio change silently rewrites your history the next time you rebuild, and the ownership series you showed the committee last quarter will no longer reproduce. That is the failure that costs you credibility rather than money.
Rolling up to economic units rather than share counts
With the table in place the rollup is mechanical. Convert each row's share count into reference units by its factor, group by issuer key, and sum. Divide by a denominator that is the reference-unit total across all classes, not the count of one class. Publish the aggregate and the class breakdown together, because the breakdown is what lets a reviewer check you.
Two rules make the result defensible. First, never let an unmapped identifier fall through into the sum. Route it to an exception queue and let the aggregate be explicitly incomplete rather than quietly wrong. An aggregate that says it covers ninety-four percent of reported value is more useful than one that claims completeness it does not have. Second, be consistent about whether the aggregate answers economic exposure or votes. Dual-class structures exist precisely because those two answers differ, so a single number cannot serve both. Pick one, name it in the field label, and keep the other as a separate series.
The wrappers that are not the issuer
Fund and trust lines need a policy decision, and the decision should be written down before anyone asks for the number. There are only three coherent choices. Exclude wrappers entirely and say so, which makes the aggregate a direct-holdings measure. Include them at face value as holdings of the wrapper, which makes the aggregate a measure of filer positioning and not of issuer ownership. Or look through them to constituents using published baskets, which is the most informative and by far the most expensive, because you need basket history at the same effective dates as the filings.
What you cannot do is mix. A universe where some wrappers are looked through and others are not produces an aggregate that is not comparable across issuers, which defeats the point of aggregating. The same discipline applies to option lines, which carry a put or call designation and a share-equivalent count. Delta-adjusting them is reasonable. Adding them to the share sum at face value is not, and puts in particular will add exposure that points the other way.
The reconciliations that catch a bad rollup
Three checks catch most of it. Run total reference units held by all filers against reported shares outstanding by issuer and flag anything above one hundred percent. Real overages happen, through securities lending chains and through timing, but a name at three hundred percent is a mapping error and not a market structure story. Run the ratio of reference units to raw share count by issuer and look at the tail, since a large ratio means a conversion factor is doing heavy lifting and deserves a manual look. Run the exception queue by reported value, not by row count, because one unmapped line in a mega-cap outweighs a hundred unmapped micro-caps.
It is worth being blunt that this work does not generate an idea. It generates the right to trust an idea that came from somewhere else. In Insider Alpha the institutional side sits alongside the Form 4 feed precisely as cross-reference material, and the value of a cross-reference is entirely a function of whether the two sides count the same thing. When an ownership number is challenged in a review, the answer that ends the conversation is the mapping table, with effective dates, and the exception rate for that name on that date.