Martin Leibowitz's Failed Defense of the Endowment Model

Martin Leibowitz, a managing director at Morgan Stanley, is one of the most respected figures in the world of finance, having made significant contributions to our understanding of the fixed income and equity markets. His latest book is The Endowment Model of Investing, and is co-authored by his Morgan Stanley colleague Anthony Bova and P. Brett Hammond, the chief investment strategist for TIAA-CREF Asset Management.

My expectations for this book were high, given Leibowitz’ stature and the widespread interest in the endowment model among financial advisors.

Those expectations were wholly unfulfilled.

The book is a lengthy and repetitious exposition of a worthless, fraudulent pseudoscience. The book is written in obscurantist jargon, so I’ll begin this review by providing a brief summary of the book, using many of the authors’ own words but stating it in a little more accessible language. I’ll go into what they don’t say after this summary, but here is what they do say, in my translation to language approximating normal English:

The usual mean-variance model for optimizing allocations to asset classes must be “tortured” to produce acceptable results. Users often discover this only through an incremental process in which they run the model then find that if allocations are unconstrained, a succession of asset classes exceed acceptable bounds. Hence maximum allocations must be imposed on those asset classes. When such limits are imposed, allocations to those assets tend to go to their limits.

We have obtained inputs to this model – expected returns and covariance matrix – from an independent source, for the following asset categories: U.S. equities; U.S. bonds; cash; International Equity; Emerging Market Equity; Absolute Return; Equity Hedge Funds; Venture Capital; Private Equity; REITs; Real Estate; and Commodities. The model is used to create 179 of the 215 tables and graphs in the book using these inputs.

The return/covariance inputs imply positive alphas for all the alternative assets – that is, all except U.S. stocks, U.S. bonds, and cash, which we call “swing” assets. U.S. stocks by definition have an alpha of zero. Hence, the model, if unconstrained, will force the portfolio allocation to be too high to the highest alpha asset, and if that one is constrained, then too high to the next-highest alpha, and so on. Since such allocations are unacceptable for various reasons, constraints must be placed on all the alternative assets. When the model is run under those constraints, allocations to the alternatives go to their maximums.

These allocations are based on volatility risk. There may be other sources of risk, especially for alternative assets. We call these “dragon risks” because they are risks of the unknown, after the designation “there be dragons” on old maps. The model, however, only addresses volatility risk therefore allocations will be made based only on volatility risk. When only volatility risk is considered, all alternative assets have positive alphas and receive high allocations.

We assume as the base case that investments made in these alternative asset classes are “passive” – that is, only average for the asset class. Hence, these allocations capture only the “allocation alpha” for the asset class. Investing to capture an allocation alpha is not a zero-sum game because all investors in the class can, on average, achieve a positive alpha. We also note, however, that the typical investment in one of these alternative asset classes may not be available to all investors. We call this “channel risk”.