Replication strategies get pitched with the capacity question waved away, usually with some version of the argument that the managers being cloned are enormous, so anything you do is small by comparison. That argument is wrong in a specific and important way. The managers being cloned had a whole quarter and complete discretion over their execution schedule. You have a public filing date, a book that has to be built quickly because the information is already stale, and a crowd doing the same thing on the same morning. The constraint is not their size. It is your window.
The number that comes out of doing this properly is usually smaller than people expect, and it is worth having before a mandate is written rather than after the first quarter of implementation shortfall shows up in attribution.
The lag sets the trading schedule, and the schedule sets everything else
Start from the timeline, since every other quantity depends on it. Positions are reported as of a quarter end. The filing arrives up to forty-five days later. You act on the filing, so entry begins after the filing date. You then hold until the next quarter's filing tells you the position changed, which is roughly three months later.
Two consequences follow. First, the holding period is fixed at about a quarter by the structure of the disclosure regime, whatever your view on the names. You do not get to hold a good position longer, because you do not learn anything new until the next filing. Second, the entry window is short by necessity. A signal that is already forty-five days old does not improve while you work an order patiently over six weeks, so the practical build window is days, not months.
That is what makes cloning capacity-constrained in a way the underlying managers are not. A manager accumulating over a quarter can take five percent of daily volume for sixty days. A cloner needs the same relative position in five or ten days. The same target position therefore requires several times the participation rate, and impact does not scale kindly with participation.
The participation constraint, name by name
The per-name limit is arithmetic and should be computed before any impact modelling, because it often binds first.
Set a participation cap, the fraction of average daily volume you are willing to be, and a build window in trading days. The maximum notional you can establish in a name is the cap multiplied by dollar average daily volume multiplied by the number of days. A ten percent cap over ten days lets you buy about one day of dollar volume. That is the whole budget for that name.
Now invert it. If a name is meant to be three percent of your book, and your budget in that name is one day of dollar volume, then the largest book that can hold the target weight is that dollar volume divided by three percent, or roughly thirty-three times one day of volume in the least liquid name at its full target weight.
Run that calculation across every line and take the minimum. The result is almost never driven by the large positions in large companies. It is driven by a mid-sized weight in a name whose daily volume is an order of magnitude below the rest of the book. Cloning portfolios inherit exactly this shape, because the managers worth cloning are usually the ones taking real positions in names that are not mega-caps, which is the same property that makes their book hard to replicate.
This is the first honest place to stop and reconsider. If the binding name caps you well below the mandate size, you have three options and all of them change the strategy. Cap position weights, which changes the portfolio you are supposedly replicating. Drop illiquid names, which drops the part of the book most likely to carry the manager's actual edge. Or extend the build window, which increases the staleness of an already stale signal.
Impact against an edge that is already decaying
The participation limit tells you what is possible. Impact tells you what is affordable.
Use whatever impact model your desk has calibrated. The standard shape has a spread-crossing term plus a term that grows with the square root of the ratio of order size to daily volume, scaled by volatility. The exact coefficient matters less than the structure, which is that cost per share rises as you push more of a day's volume through, so doubling the book more than doubles the cost of building it.
Then build the cost side of the ledger properly, which means counting both directions and the frequency. Each quarter you trade whatever fraction of the book changed, and you trade it twice, out of the old names and into the new. Four quarters a year. So annualised implementation cost is roughly four times the quarterly turnover times the round-trip cost per dollar traded. A replication book that inherits fifty percent quarterly turnover from the manager it clones is trading its whole book twice a year in short, crowded windows.
Capacity is then the point where that annualised cost consumes whatever gross edge you were willing to underwrite. Note that you have to supply the edge assumption yourself, and that the honest version of this exercise is run across a range of edge assumptions rather than a single flattering one. Present the capacity number as a curve against assumed gross edge, because that is the form in which it is actually decision-useful and the form in which it cannot be quietly optimised.
The crowd in the same window
One adjustment separates a capacity estimate that survives contact with reality from one that does not.
Average daily volume measured over a trailing period is not the volume available to you in the days after a filing deadline. Every other replication book, every screener output and a good deal of retail attention hits the same names in the same window. Some of that volume is other people trying to buy what you are trying to buy, which means it is not liquidity you can consume, it is competition for the same liquidity.
The practical adjustment is to haircut available volume in the post-deadline window and to check realised volume in those specific days for the names you actually trade. If your fills in the three days after a filing are systematically worse than your model, the crowd is the explanation and your capacity number needs to come down accordingly. This is measurable from your own execution data after a single quarter, and measuring it is cheaper than assuming it away.
Reporting the number so it survives a review
Capacity is a range with assumptions attached, and it should be documented as one. The four inputs that determine it are the participation cap, the build window in days, the impact coefficient, and the assumed gross edge. State all four next to the number. A single capacity figure with no visible assumptions is not a risk statement, it is a marketing claim, and it will be treated as one the moment the strategy underperforms.
Two ongoing checks keep the estimate honest after launch. Track realised implementation shortfall per quarter against the modelled cost and treat a persistent gap as evidence that the model is wrong rather than that the quarter was unusual. And re-run the binding-name calculation every quarter, because the manager you are cloning will change the book and the constraint moves with them.
The strategic point underneath all of this is that a cloning book's capacity is set by a disclosure schedule you do not control. Inside Insider Alpha the institutional holdings sit alongside the Form 4 feed as cross-reference material, and the difference in timing between the two is the whole argument. A quarterly filing arrives on a public calendar and concentrates everyone's trading into the same days. A Form 4 arrives on its own schedule with a transaction date attached. Signals that arrive on a common clock are structurally more crowded, and crowding is the mechanism by which a capacity estimate turns out to have been optimistic.