Public pension private equity performance — the statistics

Pensions: all tracked plans · search every PE fund · CalPERS PE table · CalSTRS PE table · where they diverge

US public pension plans publish, fund by fund, how every private-equity fund they hold has performed — and pooled, those disclosures make a dataset no vendor sells. This page is the numbers in one place: 2,065 fund disclosures from 5 plans (CalPERS, CalSTRS, Washington WSIB, Oregon OPERF, Minnesota SBI), covering vintages from 1981 to 2026 and over $426bn of committed capital. 302 of those funds are reported by more than one plan, and where two plans report the same fund the median gap between their net IRRs is just 0.39 points — but the widest is 8.09 points. Every figure below comes from the plans' own published reports; each carries its as-of date.

The dataset at a glance

Each plan disclosure lists the fund, the money committed, the net IRR (the yearly return after fees) and the multiple (dollars back per dollar in). Pooled across the five plans:

1,694 of the 2,065 disclosures carry a reported net IRR; the rest are young funds where the plan itself withholds the figure — early-life IRRs are not meaningful (the J-curve: fees are charged before profits arrive), and the plans say so in their own footnotes. The whole pooled table is searchable on the PE fund search page.

Where plans hold the same fund

The overlap is the interesting part — the same fund, reported independently by several investors:

One of the nine funds held by all five plans shows what the overlap buys you. Advent International GPE IX (vintage 2019) is reported five times, and no two numbers agree:

That is a 4.65-point spread between five honest reports of one fund — different vehicles, different fee terms, different valuation dates. The gotcha for any reader of a single plan's report: an LP-reported IRR is that investor's return, not the fund's return, and no single disclosure tells you which end of the spread you are looking at.

The divergence statistics — same fund, different numbers

For the two California plans we ran the comparison formally, with conservative name-matching (sleeves and parallel vehicles are not treated as the same fund; a name match whose vintages disagree is rejected). Full findings: the divergence page.

The three widest, as each plan reports them:

Even a 2004-vintage fund that has long since returned its capital — Permira Europe III — still prints 26.6% at one plan and 24.15% at the other, a 2.45-point gap two decades on. The plans report to different dates (CalPERS September 30, 2025; CalSTRS June 30, 2025), may hold different vehicles with different fee terms, and each computes its IRR from its own cash-flow dates — most of a gap is timing and terms, not a disagreement about what the asset is worth.

What the two California tables show

The two fund-by-fund tables carry the full range of private-equity outcomes, from compounding winners to capital-destroying energy vintages:

Both plans' figures run about two quarters behind — private funds report slowly, and the plans publish what they have. Browse the full tables: CalPERS, 462 funds · CalSTRS, 477 funds.

The plan universe behind the PE tables

The PE disclosures sit inside a wider tracked universe of 320+ pension plans (the pension hub), including 253 US state and local plans from the Public Plans Database — 247 of the 253 have reported fiscal year 2024, the newest audited year (audited reports reach the database 12–18 months after year end). The two plans whose PE tables anchor this page:

For scale, the next-largest US plans tracked: NY State & Local ERS ($226.0bn, 11.6% one-year return), Texas Teachers ($210.5bn, 9.0%) and Florida RS ($198.7bn, 10.5%) — all FY2024, all from their own filings. The gotcha: a funded ratio (the share of promised benefits covered by assets on hand) is a GASB accounting figure on a lagged audited year — it is not a market-value snapshot of the plan today.

How fresh the numbers actually are

Three clocks run at different speeds, and conflating them is the classic reading error:

The same institution can honestly show $506.6bn (audited FY2024) and $634.1bn (monthly snapshot, May 2026) at once — a $127bn difference that is entirely reporting lag, not performance. The monthly figure is unaudited and not on the database's normalised basis, so the two are never merged. Gotcha: whenever a pension-PE number is quoted anywhere, the first question is which of these three clocks it came from.

Method and coverage gaps

• Five plans is not the market. The pool is the five large US plans that publish machine-readable fund-level PE performance. Other large disclosers exist but are not yet in the pool (see To verify).
• Committed totals are floors, summed from each plan's own rows and rounded down; the plans report to different dates, so the pooled totals are not a single point-in-time snapshot.
• CalSTRS multiples are derived from its published contributions and distributions — its report carries no multiple column.
• WSIB vintages are derived from each fund's inception date rather than stated by the plan.
• Oregon flags funds sold in the secondary market as not comparable with funds held to liquidation; those rows are marked, not dropped, in the search table.
• Cross-plan matching is conservative — exact name matches after style normalisation plus a hand-checked alias list; sleeves and parallel vehicles stay separate; unmatched stays unmatched.