Maximum Diversification Portfolio: Maximizing the DR Ratio

Key Takeaways

• The maximum diversification portfolio maximizes DR = (w'σ) / sqrt(w'Σw), giving meaningful weight to a high-volatility asset that is uncorrelated with the rest, unlike minimum variance which would de-weight it.
• Under a single-factor model, the most diversified portfolio also maximizes the Sharpe ratio when expected returns are proportional to volatilities, giving the optimization a theoretical return-based anchor.
• The unique property of the MDP is that every asset in the final portfolio has the same correlation with the portfolio itself; no single holding dominates the correlation structure.
• Without sector caps, the optimization concentrates in low-volatility sectors like Utilities and Staples, similar to minimum variance, institutional implementations routinely add sector limits to prevent this.

What It Is

Choueifaty and Coignard defined the diversification ratio of a portfolio as:

DR(w) = ( w' * sigma ) / sqrt( w' * Sigma * w )

The numerator is the weighted average of single-asset volatilities. The denominator is the portfolio volatility. If the assets were perfectly correlated, the two would be equal and DR = 1. As correlations and weights spread out, DR rises. The most diversified portfolio is the long-only w that maximizes this ratio.

The 2008 paper showed that under a single-factor assumption, the most diversified portfolio is also the one that maximizes Sharpe ratio when expected returns are proportional to volatilities. That gives the optimization a theoretical anchor and explains why it has historically delivered Sharpe ratios competitive with minimum variance and risk parity in long backtests.

The Intuition

Diversification has two components: how many assets you hold, and how independently they move. The diversification ratio captures both. A 50-stock portfolio of bank stocks has lots of names but very high correlations, so its DR is close to 1. A 10-asset portfolio spanning equities, government bonds, gold, and cash has fewer names but much lower correlations, so its DR can exceed 2.

The most diversified portfolio chooses weights so the portfolio is as far as possible from being a single concentrated bet, in the precise sense that its volatility is much smaller than the volatility you would expect from the weighted average of its parts. Like minimum variance, it ignores expected returns, which makes it useful when you do not trust your forecasts.

How It Works

The optimization problem is:

maximize    DR(w) = ( w' * sigma ) / sqrt( w' * Sigma * w )
subject to  sum(w) = 1,  w >= 0
            sector, factor, or position constraints as needed

A useful equivalent form: the most diversified portfolio is the long-only portfolio whose weights are proportional to the inverse correlation matrix times the inverse-volatility vector, before applying constraints. In closed form for the unconstrained case:

w_MDP  proportional to  Sigma^{-1} * sigma

where sigma is the vector of single-asset volatilities. Normalizing to sum to one gives the unconstrained MDP. With long-only and other constraints, the problem becomes a small quadratic program over a transformed objective.

Choueifaty, Froidure, and Reynier (2013) showed that under mild assumptions, the most diversified portfolio is the unique long-only portfolio whose correlation with every asset in the universe is the same. That property gives it a clean economic interpretation: every asset contributes equally to the portfolio's "spanning" in correlation space.

Worked Example

Consider a four-asset portfolio with volatilities 20 percent, 15 percent, 12 percent, and 8 percent. Suppose the correlation matrix has all off-diagonal entries equal to 0.30.

Step 1: weighted average volatility under equal weighting is (20 + 15 + 12 + 8) / 4 = 13.75%. Portfolio volatility under equal weighting comes out around 11.7 percent. So DR(equal) = 13.75 / 11.7 = 1.18.

Step 2: solve for w_MDP. The closed-form solution proportional to Sigma^{-1} * sigma produces approximate weights:

asset 1 (vol 20%):  18%
asset 2 (vol 15%):  22%
asset 3 (vol 12%):  26%
asset 4 (vol  8%):  34%

Portfolio volatility is around 8.7 percent. Weighted average volatility at these weights is around 12.0 percent. Diversification ratio: DR(w_MDP) = 12.0 / 8.7 = 1.38.

The MDP weights are tilted toward lower-volatility assets but, unlike minimum variance, also reward assets whose marginal correlation with the portfolio is lower. Adding a fifth asset that is uncorrelated with the others but has 25 percent volatility would still receive meaningful weight, because it raises numerator and lowers denominator at once.

Common Mistakes

1. Confusing it with minimum variance. Both ignore expected returns and depend only on Sigma, but they optimize different things. Minimum variance pushes toward the lowest absolute volatility. The MDP pushes toward the largest gap between weighted-average volatility and portfolio volatility, which often gives a more diversified holding pattern across the volatility spectrum.

2. Ignoring covariance estimation error. Like every covariance-based optimizer, the MDP relies on Sigma. Raw sample covariance with high asset counts produces unstable solutions. Pair the optimization with a shrinkage estimator or a factor-model covariance.

3. Treating the long-only constraint as optional. The unconstrained MDP can hold short positions in highly correlated assets to create artificial low-correlation residuals. The published methodology and most index implementations are long-only. Removing that constraint changes the portfolio's risk profile substantially.

4. Assuming the MDP is sector-balanced. Without sector caps, the optimization can stack into low-volatility sectors much like minimum variance. Most institutional implementations add sector or country caps to prevent concentration in defensives.

5. Reading historical Sharpe outperformance as guaranteed. TOBAM's published backtests look strong over long samples, but the strategy has its own underperformance regimes, particularly when the highest-volatility assets dominate returns. The MDP is not arbitrage; it is a structural alternative to cap weighting with its own cycle.

Frequently Asked Questions

Q: What is the maximum diversification portfolio in simple terms? It is the long-only allocation that maximizes a metric called the diversification ratio, the weighted average of individual asset volatilities divided by the portfolio volatility. The larger the gap between the two, the more the portfolio benefits from assets not moving together.

Q: How does the maximum diversification portfolio affect investment decisions? It explicitly rewards including uncorrelated assets, even volatile ones, because they raise the numerator of the diversification ratio while having a small impact on the denominator. This makes the approach more accommodating of unconventional diversifiers like commodities or gold than minimum variance.

Q: What is a real-world example of the maximum diversification portfolio? TOBAM's Anti-Benchmark strategy, launched commercially from the 2008 paper, applies maximum diversification to large equity universes. TOBAM's published backtests over 30-year samples show Sharpe ratios competitive with minimum variance and risk parity, with distinct performance regimes depending on how asset correlations cluster.

Q: How can investors implement the maximum diversification portfolio? Compute the diversification ratio for a range of long-only weight vectors and find the maximum numerically. The unconstrained closed-form solution is proportional to Σ⁻¹σ, but with long-only and sector constraints the problem becomes a small quadratic program. Use a good covariance estimate (shrinkage or factor model) as the input.

Q: How is the maximum diversification portfolio different from minimum variance? Minimum variance minimizes total portfolio volatility in absolute terms, heavily favoring stable low-volatility assets. The maximum diversification portfolio maximizes the relative gap between average-asset volatility and portfolio volatility, giving meaningful weight to higher-volatility assets when they are genuinely uncorrelated. Both use only the covariance matrix but answer different questions.

Sources

1. Choueifaty, Y. and Coignard, Y. (2008). "Toward Maximum Diversification." Journal of Portfolio Management, 35(1). https://www.tobam.fr/wp-content/uploads/2014/12/TOBAM-JoPM-Maximum-Div-2008.pdf
2. Choueifaty, Y., Froidure, T., and Reynier, J. (2013). "Properties of the Most Diversified Portfolio." Journal of Investment Strategies, 2(2). https://www.tobam.fr/wp-content/uploads/2014/12/TOBAM-JoIS-Most-Diversified-Portfolio.pdf
3. TOBAM. "Anti-Benchmark and Maximum Diversification Methodology." https://www.tobam.fr/anti-benchmark/
4. Qian, E. (2016). Risk Parity Fundamentals. CFA Institute Research Foundation. https://rpc.cfainstitute.org/research/foundation/2016/risk-parity-fundamentals

Disclaimer

This article is educational content only and is not financial advice. Nothing here is a recommendation to buy, sell, or hold any security. Consult a licensed advisor before making investment decisions.