Key takeaways
- Yale's reported ten-year lead over a typical 70/30 stock and bond portfolio was 3.8 percentage points a year in its fiscal 2024 report and 2.2 points in its fiscal 2025 report.
- Harvard held 41% of its endowment in private equity and 14% in public equities on 30 June 2025, a shape no ordinary brokerage account can copy.
- Across the 657 institutions in the FY25 NACUBO study, private and alternative strategies took 54.5% of dollar-weighted assets, against 31.5% for public equities.
- Richard Ennis fitted an index blend to 41 large endowments over fiscal 2009 to 2023. The blend returned 8.3% a year; the endowments returned 7.4%.
- Buyout funds beat the S&P 500 by more than 3% a year in Burgiss data covering nearly 1400 US private equity funds, which is the counter-case at its strongest.
Yale's ten-year premium over a 70/30 fell by 1.6 points in a year
Can a household run money the way Yale runs its endowment, and what does the gap cost if it can't? Yale answers the second half of that question once a year, in a sentence most readers skim.
In its fiscal 2024 report, Yale said the endowment "returned 9.5% per annum" over the decade to 30 June 2024. It added that the result "exceeds that of a typical 70/30 stock and bond portfolio by 3.8% over both the past 10 and 20 years". Twelve months later the same disclosure read differently. Over the decade to 30 June 2025 Yale returned 9.4% a year. That beat the median college endowment by an estimated 1.4 points and a typical 70/30 portfolio by 2.2 points.
Nothing about the portfolio changed in between. The window changed. Drop fiscal 2015, add fiscal 2025, and the measured reward for holding illiquid assets falls from 3.8 percentage points a year to 2.2. That's 42% of the advantage, gone in one reporting cycle.
Yale doesn't publish the 70/30 series it compares itself to, so the level has to be inferred. Its own two statements imply that portfolio returned about 5.7% a year over the decade to June 2024 and about 7.2% over the decade to June 2025. Yale's fiscal 2025 return was 11.1% net of fees, taking the endowment to $44.1 billion. A strong year for listed markets lifted the comparison portfolio more than it lifted Yale.
What an endowment model portfolio actually holds
Harvard publishes its allocation, which makes the shape concrete. On 30 June 2025 the endowment held 41% in private equity, 31% in hedge funds, 5% in real estate, 4% in bonds and TIPS, 3% in cash, and 14% in public equities. Venture capital, inside that private equity figure, was 14% on its own. Harvard held as much in venture capital as in listed shares.
The sector is less extreme and points the same way. The FY25 NACUBO-Commonfund study covers 657 institutions and $944.3 billion. On a dollar-weighted basis, private and alternative strategies took 54.5% of assets against 31.5% for public equities. Private equity alone was the largest single allocation at 16.8%, ahead of US equities at 13.7%.
That shift took a little over three decades. Ennis's 2020 study puts alternatives at 4% of endowment assets in 1986 and 58% by 2019. An endowment model portfolio, as practised today, is mostly a private-assets portfolio with a listed remainder.
The rules remove the half of the portfolio you would need
The public-market approximation isn't a preference. It's what's left after the access rules take the rest.
In the US, the funds holding those assets are sold under exemptions open to accredited investors. The SEC's investor bulletin sets the test for an individual. It is income "that exceeded $200,000 (or $300,000 together with a spouse or spousal equivalent) in each of the prior two years", or "a net worth over $1 million" excluding the family home.
The UK route is narrower still. The FCA created the long-term asset fund for exactly this purpose. It expects an LTAF "to invest at least 50% of the scheme property of an LTAF in assets that are illiquid and need to be held over the longer term". Those funds also carry a promotion restriction. The FCA's guidance says they "are intended only for professional clients and for retail clients who are sophisticated investors, certified high net worth investors", plus a narrow further class. The same wall shows up elsewhere on this site, in the gap between what institutions report holding and what a household can buy, which is the pattern behind institutional crypto allocation.
More alternatives meant more return over 25 years, less over one
NACUBO reports allocation and outcome by endowment size. The two together are the closest thing to a controlled test the sector publishes. Larger endowments hold more alternatives, and here's how that has paid.
| Cohort | Alternatives, FY25 | One-year, FY24 | One-year, FY25 | Ten-year, to FY24 | 25-year, to FY24 |
|---|---|---|---|---|---|
| Under $50m | 12.5% | 13.0% | not reported | 6.5% | 4.5% |
| $251m to $500m | 32.3% | not reported | not reported | not reported | not reported |
| Over $5bn | 62.5% | 9.1% | 11.8% | 8.3% | 8.5% |
Read down the last column and the endowment model looks unanswerable. A margin of 4.0 percentage points a year for 25 years is enormous. Read the ten-year column and it's 1.8 points. Read the one-year columns and it flips with the direction of listed markets. In FY24 the smallest cohort returned 13.0% and the largest 9.1%. In FY25, the year listed markets ran, the largest cohort posted the highest return at 11.8%.
The honest referee call is that the 25-year number carries the argument. Those 25 years to June 2024 include the back half of the stretch Ennis calls the golden age of alternative investments, fiscal 1994 to 2008, when endowments beat their fitted benchmarks by 4.1 percentage points a year. Whether that era repeats is the whole question, and no table answers it.
An index blend fitted to 41 endowments beat the endowments
Richard Ennis ran the test directly. His April 2024 study takes "a sample of 41 large US endowments over the 15 fiscal years, 2009-2023". He then builds a benchmark by quadratic programming: the mix of Russell 3000 stocks, MSCI ACWI ex-US stocks and Bloomberg Barclays US Aggregate bonds that statistically best fits each return series. For the composite the weights came out at 61%, 22% and 17%.
The composite returned 7.4% a year. The blend returned 8.3%. The endowments trailed by 0.9 percentage points annualised, with "underperformance in 12 of 15 years" and a cumulative shortfall of 13%. Over the first 12 years the average miss was about 1.7 points a year.
The chart above sets those decade-scale numbers beside each other, and the spread across the whole endowment world is narrower than the folklore suggests. The individual results inside it are less comfortable than the average. Yale ranked 16th of 41 on excess return at -0.4 points a year. Harvard ranked 40th at -2.5.
The risk-adjusted table is starker. Yale returned 8.3% a year with a standard deviation of 13.7% and a Sharpe ratio of 0.560. The fitted index blend returned the same 8.3% at a standard deviation of 13.5%, for a Sharpe ratio of 0.569. Ennis concludes that "indexing with typical market exposures would have been a superior strategy for the great majority of the funds in terms of risk-adjusted return".
Two things are worth separating. Ennis fits his benchmark after the fact, to the returns he is measuring, which makes the fit flattering by construction. He also notes that "investment offices once widely considered to be elite" are "not in the top quartile over the past 15 years", while the top quartile holds Missouri, Michigan State and Rice.
The 2021 to 2023 swing is the smoothing, not the skill
The year-by-year pattern matters more than the average. Twelve years inside what Ennis calls a narrow range, averaging about -1.7 points, then +9.0 points of excess return in 2021, +8.2 in 2022, then -10.9 in 2023.
Ennis is direct about the cause. He writes that "the composite returns of 2022 and 2023 exhibit return smoothing caused by lags in reporting net asset values of private assets". Private holdings are marked by appraisal, not by the tape. In a falling year they look defensive because the marks haven't caught up. In the recovery they give it back. It's the same mechanism that separates listed property from the buildings themselves, covered in REITs vs direct property.
Both flagships described the lag in their own words last year. Yale's fiscal 2025 result "trailed public indices during a period of strong market performance as outperformance in several asset classes was tempered by underperformance in leveraged buyouts and, more meaningfully, real estate". Harvard's report says results "were dampened by having less public than private equity". Neither is an excuse. Both describe the same lag working in the unhelpful direction.
A portfolio reporting a smoothed return also reports a smoothed volatility, which flatters every ratio built on it. That's a measurement artefact, not a diversification benefit. It's a different thing from what actually held up in a crash, which is the subject of diversification in a crisis.
The best case for private assets is the buyout data
Harris, Jenkinson and Kaplan studied "nearly 1400 U.S. private equity (buyout and venture capital) funds" using cash-flow data from Burgiss, "sourced from over 200 institutional investors". Their finding on buyouts is blunt. "The outperformance versus the S&P 500 averages 20% to 27% over the life of the fund and more than 3% per year". That is a large margin built from cash flows rather than reported marks.
Three qualifications sit against it. Venture capital in the same study "outperformed public equities in the 1990s, but have underperformed public equities in the 2000s", so the two halves of private equity behave differently. The paper is an NBER working paper from February 2012, not a peer-reviewed statement about the 2020s. And a fund-level average isn't an investor-level outcome, because someone has to be in the funds that produced it.
Access is where the cost sits. Ennis reports that large endowments "use an average of 108 managers and invest only 6% of their assets passively". He estimates that "the cost of investing ranges from 1 to 2% of assets annually". If the buyout premium is 3 points a year gross and the portfolio-level fee load is 1 to 2 points, the survivable margin is thin. That is before you ask whether your manager is an average one. The link between what a fund charges and what it returns is measured on this site in whether fees predict fund returns.
What the public-market approximation costs, and what it can't copy
The listed side is cheap and getting cheaper. The iShares MSCI ACWI UCITS ETF factsheet dated July 2026 shows a total expense ratio of 0.20%. Against Ennis's institutional range, that's a fee gap of 0.8 to 1.8 percentage points a year. That gap is the same order of magnitude as the entire premium under discussion.
What the approximation can't copy is the liability. Harvard benchmarks itself to 8% a year, "which reflects the balance between annual distributions (approximately 5%) and inflation (roughly 3%)". NACUBO's participants ran an average effective spending rate of 4.9% in FY25. An endowment knows roughly what it has to pay out, and roughly forever. That's the structure the illiquidity is underwriting, rather than a return premium as such.
Nor can it copy the liquidity design. Harvard writes that its "allocation to uncorrelated hedge funds allows HMC to moderate risk and provides access to liquidity across market cycles". The private book is illiquid on purpose, and something else in the portfolio pays the bills while it is. A household's version of that is a cash and bond position sized to spending, which is a separate question from what the growth assets are. The same translation loss appears in the All Weather portfolio rebuilt in ETFs. If you're tracking how far a portfolio has drifted from its target weights, LedgerTouch reports that continuously.
What this data cannot tell you
All of it is US institutional data on fiscal years ending 30 June, which is neither a calendar year nor a UK tax year. NACUBO's returns are self-reported and net of fees. Its allocation figures are explicitly dollar-weighted, while the release never says how its return averages are weighted, so the two halves of the table above may not be measured on the same basis.
Yale's "typical 70/30 stock and bond portfolio" is not defined in the release that cites it, nor on Yale's own endowment page. The 5.7% and 7.2% figures implied by its own statements therefore carry whatever construction Yale chose. Ennis's benchmarks are fitted to the returns they measure, which is a real objection to the size of his excess-return numbers, though not to their sign. And the private marks underlying every endowment return here are appraisals with a reporting lag. None of these ten-year numbers is measured the way a listed portfolio's ten-year number is measured. A backtest of institutional returns is not a forecast, and the return series here cover one country over at most 25 years.
What would change the conclusion
The decade to 30 June 2026 drops fiscal 2016 and adds fiscal 2026. If listed markets stay strong, Yale's reported margin over a 70/30 narrows again from 2.2 points, and at some window it crosses zero. If private marks catch down to the exits instead, the margin widens, and the smoothing shows up as a delayed loss rather than a premium.
The number to watch isn't the flagship headline. It's the 25-year cohort spread of 8.5% against 4.5%, the strongest evidence the endowment model has, and the one most exposed to which quarter-century you measure. Once that window has rolled past the golden age of the late 1990s and 2000s, the gap between an endowment model portfolio and its public-market approximation gets measured on modern data for the first time. Cost is the other thing that could move it. If LTAF charges compress toward the 0.20% a tracker charges, rather than the 1% to 2% Ennis measures, the arithmetic changes at the fee line before it changes anywhere else.