Hold or Sell? A More Realistic Way to See the Decision
The decision
A common question reaches an LP investment committee: a concentrated, under-performing sub-portfolio sits inside a much larger program. Hold it to its natural wind-up, or sell it on the secondary market at a discount and redeploy the cash?
The decision is genuinely hard. The standard way of approaching it tends to obscure why it is hard — and a more transparent approach makes the real source of the difficulty visible.
Where the usual approach loses transparency
The instinct is to forecast each path — project the cash flows of holding, project the proceeds of selling and redeploying, take the expected outcome of each, and compare.
The trouble is that a point forecast of a private-market cash flow is not identifiable from the available data. Two careful analysts, working the same position, will reasonably choose different inputs and arrive at materially different "expected" answers — and the data do not adjudicate between them. When a single number is reported, that genuine uncertainty disappears from view: the answer looks precise, but its precision is an artifact of the modeling choices, not of the evidence.
A more transparent frame
The alternative is to make the uncertainty explicit rather than hide it. Instead of a single forecast, the framework works with the full range of outcomes an LP could reasonably believe in — and that range is built from three ingredients, each one visible and open to inspection:
- a peer anchor — how comparable funds, matched on vintage and sector, have actually performed;
- a stated adjustment — the committee's explicit, reasoned view of why this sub-portfolio differs from the generic peer (its specific concentrations); and
- an honest uncertainty envelope — an open acknowledgment that no one knows these numbers to the decimal point.
The recommendation is then evaluated across that whole range, not at a single guess. Nothing is buried in an estimator; every input is on the table.
What this makes visible
When the recommendation is the same across every reasonable belief, that is a genuinely informative result: the choice does not hinge on any one forecast, and the committee can see that clearly.
When the range straddles the decision boundary, the framework says so directly: the decision is close, and which way it tips depends on the institution's own risk preferences — not on an assumption quietly embedded in a model. Surfacing that fact is the contribution. A frame that shows a committee when a decision is genuinely close is more useful, and more honest, than one that manufactures false confidence.
Why this is realistic, not a workaround
A fair question: if the data are too thin to forecast, how can they support anything at all?
Because estimating a precise number and assessing a broad range are different tasks. The data are indeed too sparse to pin a sub-portfolio's loss probability to within a point or two — and a realistic method does not pretend otherwise. But they are perfectly adequate to inform a range: peer cohorts, matched on vintage and sector, show how outcomes of this kind have actually distributed. Working with that range — rather than collapsing it into one number the evidence cannot support — is simply a more faithful representation of what is known and what is not.
This note is a starting point. The fuller practitioner companion works the decision through a complete example for an investment committee; the underlying research develops the framework and its statistical foundations in full.