What's the real value of Harvard's endowment?

For years, financial journalists have been hearing about the extraordinary talents of the clever chaps who run Harvard's endowment.  In some way that is not quite clear to me, the vast and shadowy operations of the richest and most elitist university in the United States--quite possibly in the world--was supposed by many liberals of my acquaintance to be a rebuke to Wall Street, and capitalism more generally.

The admission, several weeks ago, that the endowment had lost a quarter of its assets, took down the icon a peg.  Harvard, it turned out, had not discovered some magic way to make a ton of money without risk after all.

Now Edward Jay Epstein suggests that it's not as bad as all that . . . it's worse:

Harvard University's admission that it lost $8 billion from its $36 billion endowment fund, as staggering as it sounds, may grossly underestimate the true magnitude of the loss between from July 1 through Oct. 31 2008. According to a source close the Harvard Management Corporation (HMC), which runs the fund for Harvard, the loss is closer to $18 billion if the losses on the fund's illiquid investment are realistically appraised.

To be fair, the fund managers may believe, justly, that since the Harvard Endowment (unlike almost any other fund) can guarantee that it won't need to sell any of its illiquid assets any time soon, their valuation is closer to "true" value than a massive write-down.  On the other hand, that's how we value funds--by what they could get now, not by what they might get at some unspecified point in the future.  And it seems to indicate that Harvard was getting its high returns just like all the other financial firms were:  by investing in risky, illiquid assets (including shares in hedge funds who did same).  Their ability to massively diversify--again, much more than most funds who have shorter time horizons and narrower mandates--made that almost always a winning strategy.  But as Brad DeLong has noted, in a financial crisis all correlations go to one.