The Uncorrelated Book: What Diversification Actually
Buys a Multi-Manager Platform

In one line: Combining ten trading desks cut total risk by 57 percent, which says nothing about profit.

Abstract. I decompose the risk of the museum’s Trading Floor, a disclosed paper book constructed as a multi-manager platform, into standalone desk risk and the risk that survives their combination. Ten desks (three systematic: equity beta, low-volatility/BAB, and trend/TSMOM; and seven within-sector long/short) are combined by inverse-volatility risk parity to a 10% annualized volatility target. The weighted-average standalone desk volatility is 18.0%, while the combined, unlevered book realizes 7.7%. That is a diversification ratio of 2.35x and a reduction in risk of roughly 57% from diversification alone. Average pairwise absolute correlation across the ten desks is 0.29, with a built-in hedge (low-volatility/BAB at −0.66 to equity beta) offsetting a momentum-aligned cluster of sector desks. The universe is current S&P constituents, so the return behind the book is an upper bound; the exhibit is the risk discipline, not the raw return.

1. Introduction

A multi-manager platform earns its structure not from any single desk but from the imperfect correlation among many of them. The fundamental law of active management makes the mechanism explicit: value added rises with the breadth of independent bets, not with the conviction behind one of them (Grinold and Kahn, 2000). This note measures what that breadth is worth on the museum’s Trading Floor, a disclosed paper book assembled as ten desks and combined into one portfolio. The object of study is the gap between the risk each desk carries alone and the risk the combined book actually realizes, and the correlation structure that produces the gap.

2. Construction of the book

The book comprises ten desks: three systematic sleeves (equity beta, a low-volatility/betting-against-beta desk, and a trend/time-series-momentum desk) and seven within-sector long/short desks. Desks are combined by inverse-volatility risk parity and the whole is scaled to a 10% annualized volatility target. Because the seven sector desks share a single momentum factor and therefore co-move, the allocator budgets them as one group rather than ten independent positions, so the cluster cannot swamp the book. The construction is deliberately disclosed and unlevered; the point of the exhibit is the arithmetic of combination, not a proprietary allocation.

3. Results

3.1 Risk decomposition

Following Choueifaty and Coignard (2008), the diversification ratio is the weighted average of standalone desk volatilities divided by the volatility of the combined portfolio. The weighted-average standalone desk volatility is 18.0%. Combined and unlevered, the book realizes 7.7%. The ratio is therefore 2.35x, equivalent to a reduction in risk of roughly 57% attributable to diversification alone, before any leverage is applied to reach the target.

3.2 Correlation structure

Average pairwise absolute correlation across the ten desks is 0.29, low enough that the sleeves offset rather than reinforce one another. Two features carry most of the benefit. The low-volatility/BAB desk is −0.66 correlated to equity beta, a built-in hedge that fires precisely when beta hurts. The trend/TSMOM desk is +0.60 to beta, adding directional exposure in persistent regimes. The seven sector desks cluster at roughly +0.3 to +0.5 among themselves because they share one momentum factor, which is the reason the allocator treats them as a single risk budget.

4. Discussion

The reading is straightforward. Uncorrelated sleeves, not a single large bet, are the platform’s engine, and the 2.35x ratio is a direct measure of what that diversification buys. The negative correlation of the low-volatility desk to equity beta and the modest average pairwise correlation are what let ten desks of 18.0% standalone volatility combine to 7.7%. Grouping the momentum-aligned sector desks into one budget prevents a correlated cluster from quietly rebuilding the concentration that risk parity is meant to remove. The exhibit demonstrates risk discipline in construction; it is not a claim about how much the book earns.

5. Limitations

The book trades a survivorship-biased universe of current S&P constituents, so any return or Sharpe ratio behind the combined book is an upper bound and is not the subject of this note. The measured object is risk: standalone volatilities, the combined volatility, the diversification ratio, and the correlation structure. Correlations are historical and regime-varying, and the −0.66 hedge and the sector cluster can both migrate in stress; the budgeting of the sector desks as one group is a guard against exactly that migration rather than a promise it will not occur.