A short record is a small sample from a noisy distribution. Test it against the market's dispersion before reading the story.
Divide the manager's average outperformance by its standard error, measured with the dispersion of all funds rather than the manager's own few. Four funds that beat their vintage median by an average 0.35x of net TVPI look decisive, a t-statistic of 3.30 on their own spread, but against a 0.55x market dispersion the t-statistic is 1.27, and proving skill at the 2.0 level would take 10 funds, about 30 years. A deal-level record of 40 deals is no shortcut: correlated deals count as about 8.
Worked in full in The Private Markets Limited Partner by Julian R. Sterling, with every figure reproduced in a free workbook.See the book on Amazon →
Manager selection is the decision a limited partner is paid to make, and the track record is the evidence offered for it. Any record is a small sample drawn from a noisy distribution. Before reading the story, an LP should ask how likely the same numbers would be from a manager with no skill at all. The arithmetic is short, and it changes how much weight the deck's first page deserves. All figures are illustrative.
| Input | Value |
|---|---|
| Fund I, net TVPI above vintage median | +0.45x |
| Fund II | +0.20x |
| Fund III | +0.60x |
| Fund IV | +0.15x |
| Dispersion of fund TVPI around the vintage median, all managers | 0.55x |
| Realised deals across the four funds | 40 |
| Average deal gross multiple above benchmark; deal dispersion | 0.45x; 1.40x |
| Correlation between deals in the record | 0.10 |
Measuring each fund against its own vintage median removes the common market effect, which is the first adjustment any comparison needs. The 0.55x is the spread of outcomes across the whole market, the same scale used in how many private equity funds you need to diversify.
t = average excess ÷ (σ ÷ √n)
Funds needed for a target t: n = (t × σ ÷ average excess)²
Effective number of deals = deals ÷ (1 + (deals − 1) × ρ)
Excel: =AVERAGE(Excess)/(0.55/SQRT(COUNT(Excess)))
Step 1, the average. (0.45 + 0.20 + 0.60 + 0.15) ÷ 4 = 0.35x above median per fund.
Step 2, the wrong standard error. The four excesses have a sample standard deviation of 0.212, so the standard error is 0.106 and t is 3.30. That looks conclusive and is not: four observations cannot estimate their own dispersion, and these four happen to be tightly bunched.
Step 3, the right standard error. Use the dispersion of fund outcomes across the market, 0.55x. The standard error is 0.55 ÷ √4 = 0.275 and t = 0.35 ÷ 0.275 = 1.27. The one-sided probability of a record this good from a manager with no edge is 10.2 per cent.
Step 4, how long proof takes. n = (2.0 × 0.55 ÷ 0.35)² = 9.88, so 10 funds. At one fund every three years, that is 30 years of evidence. Even the weaker one-sided 95 per cent test, t of 1.645, needs 7 funds, or 21 years.
| Average excess per fund | t after 4 funds | Funds for t = 1.645 | Funds for t = 2.0 |
|---|---|---|---|
| 0.15x | 0.55 | 36.4 | 53.8 |
| 0.25x | 0.91 | 13.1 | 19.4 |
| 0.35x | 1.27 | 6.7 | 9.9 |
| 0.50x | 1.82 | 3.3 | 4.8 |
Only a very large edge, half a turn of TVPI on every fund, becomes visible within a normal fundraising history. A modest, real edge of 0.15x is statistically invisible for a manager's whole career.
Luck at scale. A manager with no skill has a 6.25 per cent chance of putting four of four funds above median. Among 200 managers in market, about 12.5 will show a perfect record by chance alone, and they are the ones whose decks an LP sees first.
Forty deals look like forty observations. Treated that way, an average excess of 0.45x with a 1.40x deal dispersion gives a standard error of 0.221 and a t of 2.03. But deals in one record share a team, a strategy, a sector focus and overlapping exit markets, so they are correlated. At a correlation of 0.10 the design effect is 4.9 and the record is worth 8.16 independent deals.
| Deal correlation | Effective deals | t-statistic |
|---|---|---|
| 0.00 | 40.0 | 2.03 |
| 0.05 | 13.6 | 1.18 |
| 0.10 | 8.2 | 0.92 |
| 0.20 | 4.5 | 0.69 |
Even a small correlation erases most of the apparent sample. The deal-level record is still worth reading, for the dispersion of outcomes, the loss ratio and the source of the gains, but not as a significance test.
Using the manager's own funds to measure the noise. A short, consistent record has a small sample spread, so the test rewards exactly the records that are least informative. The second mistake is to read the conclusion as "track records are useless". They are weak evidence, not no evidence, and the decision should weigh them as such: alongside the persistence odds priced in what a top-quartile track record is worth at re-up, the stability of the team that produced them and whether the strategy that earned them is still the one being sold.
The manager selection framework and the re-up arithmetic of the book are in the free workbook for this book, with forty questions that mark themselves.
Solve n = (t x dispersion / edge) squared. With an illustrative 0.55x dispersion of fund TVPI and an edge of 0.35x per fund, a t of 2.0 needs 9.88, so 10 funds, about 30 years at one fund every three. A smaller edge of 0.25x needs 19.4 funds.
Because four or five funds cannot estimate their own dispersion, and consistent records look artificially certain. In the worked case the four funds' own spread gives a standard error of 0.106 and a t of 3.30; the market's 0.55x dispersion gives 0.275 and a t of 1.27, with a 10.2 per cent chance of the record arising from luck.
Less than it seems, because deals in one record are correlated. Forty deals with an average excess of 0.45x and a 1.40x dispersion give a naive t of 2.03; at a deal correlation of 0.10 the record is worth 8.16 independent deals and the t falls to 0.92.
This article is one calculation from The Private Markets Limited Partner. The book takes the same case from first principles to the decision, chapter by chapter, and every figure it prints is a live formula in the free companion workbooks.
Get the book on Amazon →Free companion files
Also on Amazon UK · Amazon Germany · Amazon France · Amazon Canada
Reading guide: private equity and private markets → · All 1340 articles →
If this book helped, or didn’t, a few lines on Amazon are worth more than they look: they are what the next reader goes on. Write a review. The workbook stays free either way.