An endowment has no terminal date and one job: support an institution's operating budget at a level that does not decline in real terms, while the corpus does not decline in real terms either. Those two conditions together are what intergenerational equity means, and stated arithmetically they are brutal.
What return does an endowment need to earn?
Suppose an endowment distributes a share s of its market value each year and inflation runs at i. For the corpus to hold its purchasing power, the portfolio's nominal total return net of every investment cost must be at least s + i, and rather more once the interaction terms are taken properly. At a 5% spending rate and 2.4% inflation that is a real return requirement of 5% and a nominal one above 7.4%, sustained across every market environment, forever. No asset class offers that expected real return at low volatility, and the endowment's whole allocation problem follows from that one line.
Advertisement
Why it is a floor rather than a target deserves stating precisely. A pension that misses its return assumption for a decade can be fixed by the sponsor writing a cheque, and pension accounting makes that obligation visible. An endowment that misses for a decade has transferred wealth from students not yet born to students currently enrolled, and no accounting entry records the transfer. It is invisible, it is irreversible, and it is the failure the spending rule exists to prevent.
Endowments are also frequently the operating budget's largest single line, which turns portfolio volatility into a staffing decision. When a university's endowment funds a third of its operations and the distribution falls 22% in one year, the university does not smoothly reduce consumption. It freezes hiring, defers maintenance and cuts what is easiest to cut, which is rarely what should be cut. Every rule below tries to put a shock absorber between a volatile portfolio and an inflexible budget.
Endowments, part two: the four spending rules and where the volatility lands
Four rule families cover almost everything in use, and they differ in one dimension: how much of last year's market move reaches this year's budget.
The simple rule distributes a fixed percentage of the prior year-end market value. It is transparent, impossible to game, and it transmits the entire market move into the budget with a one-year lag.
A smoothed rule distributes a fixed percentage of a moving average of the last three, five or seven year-end values, so the budget moves at a fraction of the market's speed. The cost is that it keeps spending at the old average after a crash, distributing a larger share of a shrunken corpus at the worst moment.
A hybrid rule distributes a weighted blend of last year's distribution grown by inflation and the target rate applied to current market value, typically weighting the first term at 70% or 80%. The two terms pull against each other by design: the inflation term is what the budget wants, the market-value term is what the corpus needs, and the weight is where a board states its preference.
A banded inflation rule grows last year's distribution by inflation and then clips the result so the effective rate stays inside a stated band, say 4.25% to 5.75% of market value. The band stops an inflation-linked rule from drifting into a spend-down after a long bear market or an unaffordable underspend after a boom.
Worked example: four spending rules over one ten-year path
Run all four over the same ten-year path and the differences stop being rhetorical. On a $1,200.0 million corpus at a 5.00% target rate, with a return path whose geometric mean is 6.051379% and whose worst year falls 18.4%, and cumulative inflation of 33.535403% over the decade:
| Measure | Simple | Smoothed (3y) | Hybrid (70/30) | Banded |
|---|---|---|---|---|
| Total distributed, $m | 612.1477 | 609.5223 | 637.4481 | 636.6810 |
| Mean distribution, $m | 61.2148 | 60.9522 | 63.7448 | 63.6681 |
| Standard deviation of the distribution, $m | 4.2987 | 1.6429 | 1.5578 | 2.7463 |
| Worst one-year fall in the budget | -22.4800% | -3.9837% | -2.3549% | -5.5826% |
| Standard deviation of the year-on-year change | 11.6737% | 3.7432% | 2.4616% | 2.8368% |
| Terminal corpus, nominal $m | 1292.9464 | 1292.6907 | 1254.6959 | 1259.9716 |
| Terminal corpus in year-zero dollars, $m | 968.2424 | 968.0509 | 939.5980 | 943.5487 |
| Real change in the corpus | -19.3131% | -19.3291% | -21.7002% | -21.3709% |
Read the first two columns together and the standard textbook claim needs qualifying. Smoothing does spend more than the simple rule after a crash, by $9.5294 million in the year following the fall, which is 18.30% more out of a corpus that has just shrunk. That intergenerational transfer is real. Over the full decade, though, the smoothed rule distributes $2.6254 million less than the simple rule and ends with a corpus $0.2556 million lower, a rounding error on $1.29 billion. Smoothing lags on the way up as well as down, and over a path with more up years than down those two lags very nearly cancel.
The rules that genuinely cost the corpus are the two anchored to inflation rather than to market value. The hybrid and the banded rules distribute about $25 million more over the decade and end with $38 million and $33 million less nominal corpus, because an inflation-linked floor keeps paying when the portfolio cannot afford it. Their mean effective spending rates come out at 5.2841% and 5.2444% against a stated policy of 5.00%. A board that adopted a hybrid rule believing it had adopted a 5% policy has been spending at 5.28% for a decade without a vote.
How should a board choose between the rules?
All four columns are defensible and the choice among them is a governance decision rather than a technical one. The hybrid protects the operating budget best, at a cost to the corpus visible only over a decade. The simple rule protects the spending discipline perfectly and hands the university a 22% budget cut in a single year. Nobody can have both, and the useful contribution is to put the trade in front of the board in those two columns rather than in a sentence.
Two honest limits. This is one return path, so the numbers illustrate a mechanism rather than estimate any rule's expected behaviour: the ranking of budget volatility holds on any path, the ranking of terminal corpus does not. And no fees are charged beyond what is already inside the return series, which flatters every column equally.