Maximum Diversification Ratio

Maximum diversification chooses weights that maximize the weighted average of standalone asset volatilities relative to portfolio volatility. The diversification ratio rewards holdings that have meaningful individual risk but do not move together:

DR(w) = (w'σ) / sqrt(w'Σw)

where σ is the vector of asset volatilities. It is a risk-only allocation rule, like risk parity, but it favors assets whose correlations diversify the existing basket rather than forcing equal component risks.

Interpretation

The numerator is average constituent volatility; the denominator is realized aggregate volatility. A portfolio of perfectly correlated assets has little ratio benefit. A portfolio containing imperfectly correlated risk sources can have a high ratio even when some individual assets are volatile.

Portfolio	Average standalone risk	Portfolio risk	Diversification ratio
concentrated equity	high	similarly high	near 1
correlated equity sleeves	moderate	moderate	modest
balanced diversifiers	moderate	lower	higher

Under a budget constraint and no additional restrictions, maximizing DR is equivalent in direction to:

w* ∝ Σ⁻¹ σ

then normalizing weights. This inverse covariance expression makes the chief risk obvious: noisy correlations produce unstable allocations. It is not a substitute for a robust risk model.

A constrained implementation

The ratio is homogeneous: scaling weights does not change it. One convenient convex form fixes the numerator and minimizes variance:

min_w  w'Σw
s.t.   w'σ = 1,  w ∈ C

import cvxpy as cp
import numpy as np

def maximum_diversification(cov, upper=0.25):
    n = cov.shape[0]
    vol = np.sqrt(np.diag(cov))
    w = cp.Variable(n)
    problem = cp.Problem(cp.Minimize(cp.quad_form(w, cov)),
                         [vol @ w == 1, w >= 0, w <= upper])
    problem.solve()
    raw = w.value
    return raw / raw.sum()

The cap changes the ideal ratio solution and is usually desirable. Add asset-class, factor, liquidity, and turnover constraints. Target portfolio volatility afterward only if borrowing and derivatives are economically available.

Relation to minimum variance and ERC

Minimum variance minimizes the denominator without considering whether the selected assets have low or high standalone volatility; it can concentrate in low-volatility, highly correlated bonds. Maximum diversification normalizes by w'σ and generally spreads across differentiated risk. Equal risk contribution controls component contributions directly. None of the three uses expected return, so none can distinguish cheap diversification from a structurally negative-premium asset.

Method	Primary objective	Typical concentration
minimum variance	lowest total volatility	lowest-vol assets
maximum diversification	highest risk reduction	low-correlation assets
ERC	equal component risk	balanced asset risks

Estimation and regime risk

The ratio can look excellent because historical cross-asset correlations were low. During an inflation shock or liquidity event, those correlations may rise together and the projected ratio evaporates. Use covariance shrinkage, regime-specific matrices, and correlation stress tests. Consider a floor on stressed covariance or optimize an average of normal and stressed risks.

Backtest with the actual covariance lookback, rebalance calendar, and bid-ask costs. The strategy's benefit often comes from periodic rebalancing, so turnover must be reported. Use cost-aware portfolio optimization when small covariance changes create frequent weight changes.

Use diversification ratio as a diagnostic, too. A falling realized ratio can indicate that correlations changed faster than the covariance estimator; it should prompt exposure review rather than mechanically forcing the historical optimum back into place.

Asset selection precedes optimization. Adding many near-duplicate funds can inflate the apparent opportunity set and make the ratio reward wrapper-level distinctions rather than genuinely independent sources of economic risk.

Review holdings at both security and factor levels.

That discipline improves interpretation when correlations converge under market stress.

Key takeaways

• The diversification ratio compares weighted standalone volatility with total portfolio volatility.
• Its unconstrained solution depends on an inverse covariance and therefore inherits estimation error.
• Constraints and stressed, shrunk covariances are central—not afterthoughts.
• Compare it with minimum variance and ERC on identical investable assets and after trading costs.