Attribution of What? Decision-Congruent Attribution for LP Portfolios
Performance attribution for private-markets investors is usually framed as a choice among breakdowns — by strategy, by vintage, by geography. Before asking how to attribute performance, though, it is worth asking what the attribution is of.
The question before the question
A limited partner's commitment to a fund is not a single act. In committing capital, the investor simultaneously chooses a strategy, a geography, a vintage (and so an implicit view on timing), a manager, a check size, and a position in a multi-year pacing plan. An attribution is interpretable only along the dimensions on which the investor's process actually made a decision. A breakdown along a dimension the process never decided on reports the contribution of a decision that was never taken — an artifact of the chosen grid, not a property of the program.
This is the idea behind decision-congruent attribution: derive the attribution axes from the investment policy itself — the levers the committee actually pulls — rather than imposing a generic factor grid from outside.
The trap: one bet, dressed as two
The harder problem is that an investor's decisions are rarely independent. If an LP's "emerging-markets" allocation is, in practice, always a growth-equity allocation, then geography and strategy are not two decisions — they are one decision wearing two labels. Classical attribution hides this entanglement inside an unexplained "interaction" residual, and the per-axis numbers then depend on the arbitrary order in which the effects are computed.
There is a clean way to see the entanglement. Treat each decision as a vector across the funds, and measure the angle between two such vectors. Decisions that always move together point the same way — a small angle — and the diagnostic flags them as effectively one bet. Decisions made independently sit at right angles. A portfolio can look diversified on every axis separately and still be a single underlying bet — a condition ordinary diversification statistics miss entirely, but the angle between the decisions makes immediate.
How much versus when: sizing and pacing
The two levers investors most often conflate are commitment sizing and pacing. In plain language a deployment plan mixes them: committing a lot in one year is both a magnitude statement and a timing statement. A decision-congruent attribution pulls them apart by what each is measured against:
- Sizing is how much — the magnitude tilt, measured with the deployment calendar held fixed. Did concentrating capital into certain funds help?
- Pacing is when — the timing of deployment, measured with magnitude held fixed. Did the calendar of commitments help, through its effect on the money-weighted return?
Separating them matters because an investor can be disciplined on one and not the other, and a single "allocation" number would blur the two into an uninformative average. Where the two genuinely overlap — an investor who concentrates capital often concentrates it in time, too — the method measures the overlap and splits the credit fairly, rather than awarding it arbitrarily to whichever lever is listed first.
What you get
The result is a bridge an investment committee can read line by line: starting from the return a policy-neutral, decision-free program would have produced, each policy lever adds or subtracts its contribution, and the running total arrives exactly at the return the program actually earned — with nothing left in an unexplained residual. Crucially, the method reports value destroyed by a lever as faithfully as value created, and localizes it to a specific, actionable decision rather than a catch-all "selection" bucket. An investor told that timing, not selection, cost the program return knows which part of its process to examine.
This note is a starting point. A practitioner companion develops the framework in plain language with a worked example; the underlying formal paper develops the construction, the residual-free guarantee, and the validation in full, and is available on request.