Skip to content
Learning path Gold Professional operator

Risk parity

A family of allocations that balances contributions to a selected risk measure; equal risk contribution is the special case with equal budgets.

Who this is for — Anyone building a portfolio in which no component should dominate a stated risk measure simply because it has higher volatility or greater covariance with the rest.

Risk parity describes a family of methods that distributes risk across portfolio components rather than assigning them equal capital. In its most precise and widely studied form, equal risk contribution (ERC), every component contributes equally to a selected total-risk measure, usually ex ante volatility.

The name does not identify one product. Some implementations balance asset classes, others factors or strategies; some calculate full covariance-based contributions, while others use inverse-volatility weights; some then add leverage to reach a risk target. Evaluating a risk-parity portfolio therefore requires the component definition, measure, estimator, constraints, frequency, costs and treatment of leverage.

Risk parity does not promise returns, protection from drawdowns or diversification in every scenario. It balances what its model measures. Tail, liquidity, funding, counterparty and concentration risk can remain uneven even when volatility contributions are equal.

Risk parity and volatility targeting solve different problems One distributes contributions across components; the other scales exposure through time Risk parity and volatility targeting solve different problems One distributes contributions across components; the other scales exposure through time RISK PARITY · CROSS SECTIONWeights and dependencies seek targetcontributions under a chosen riskmeasure.VOLATILITY TARGETING · TIME AXISA multiplier compares target andestimated volatility, then meetsconstraints and costs. Inverse volatility It matches the broadersolution only underparticular assumptions. Estimates Volatilities and covariancesare sample estimates and canchange. Leverage and caps A target may require leverageor be unreachable underconstraints. Trading Turnover, costs and lagsseparate a theoretical rulefrom outcomes. Cyclepedia · source-checked conceptual map
Balancing relative contributions and selecting total scale are separate operations.

Definition through contributions

Let R(w) be a differentiable risk measure that is homogeneous of degree one. Component i has Euler contribution:

RCᵢ(w) = wᵢ × ∂R(w)/∂wᵢ

An ERC portfolio with n components requires:

RCᵢ(w) / R(w) = 1 / n   for every i

More generally, risk budgeting assigns shares bᵢ that need not be equal and seeks RCᵢ/R = bᵢ. ERC is therefore a special case of risk budgeting.

For linear portfolio volatility:

σₚ = √(wᵀ × Σ × w);   RCᵢ = wᵢ × (Σw)ᵢ / σₚ

The solution depends on the full covariance matrix Σ, not only on the diagonal volatilities. With long-only constraints and a suitable matrix, the problem has useful properties established in the literature. Short positions, zero budgets, singular matrices and additional constraints require more care.


Inverse-volatility weights: a shortcut

The rule:

wᵢ ∝ 1 / σᵢ

equalizes standalone volatility exposures wᵢσᵢ. It coincides with ERC under restrictive dependence structures, such as all-zero or common pairwise correlations. With a general matrix, it does not guarantee equal contributions because each component has a different relationship with the rest.

For example, three assets have volatilities of 10%, 15% and 20%. Inverse-volatility weights normalize to about 46.2%, 30.8% and 23.1%. Suppose correlations are 0.20 between assets 1 and 2, 0.80 between 1 and 3, and 0.10 between 2 and 3. Although the exposures wᵢσᵢ are equal, the sums of correlations across rows differ. Contribution shares are therefore about 38.5%, 25.0% and 36.5%, not one third each.

This calculation does not make inverse-volatility weighting useless. It shows that it is a distinct, simpler method requiring fewer inputs. A technical description should use its own name rather than automatically assigning it the properties of an ERC solution.


Verifiable construction

  1. Define the components. Assets, factors and strategies produce different decompositions; look-through prevents duplication.
  2. Choose the measure. Volatility, Expected Shortfall or another measure needs a stated formula, horizon and estimation method.
  3. Prepare the inputs. Aligned returns, currency, frequency, corporate actions and stale prices all affect the matrix.
  4. Impose real constraints. Long/short rules, maximum weights, leverage, turnover, liquidity, collateral and lots may preclude exact ERC.
  5. Solve the problem. The numerical algorithm should report convergence, contributions, residuals and sensitivity to inputs.
  6. Stress the structure. Alternative matrices and rate, inflation, credit and liquidity shocks test what historical volatility does not contain.
  7. Implement and monitor. Costs and market moves create deviations; bands and frequency are mandate-specific.

The result should be compared with capital weights, concentrations, gross exposures and scenario losses. “Equal contributions” is a property conditional on a date, model and scope.


Scale, leverage and volatility targets

An ERC portfolio often assigns more capital to lower-volatility components. Its total risk may consequently be below that of a traditional portfolio dominated by volatile assets. Some strategies apply leverage to reach a return or volatility objective.

Leverage is not a necessary part of the ERC definition. It is a separate decision that adds financing costs, margin variation, liquidation risk, collateral needs and operational risk. Volatility targeting can scale the full weight vector, but it cannot fix a misspecified covariance matrix or guarantee that realized volatility will equal the target.

Comparing a risk-parity portfolio with a 60/40 allocation requires equal risk scale, the same costs and coherent treatment of leverage. Comparing raw returns from portfolios with very different volatility attributes to the method what may merely be an effect of scale.


Dependence, crises and unmeasured risks

Correlations may change in turbulent periods, but there is no law under which all correlations converge to +1. Some rise, others fall or change sign, and average linear correlation need not describe tail dependence. Correlation observed in a high-volatility period can also be distorted by the change in variances.

An ERC portfolio based on a historical window may therefore lose its balance. Bonds and equities can share inflation or rate factors; different commodities can depend on common funding; apparently different strategies may take the same position during a shock. Scenarios and factor contributions complement an asset-class view.


What the method does not decide

Risk parity does not determine:

  • which assets or strategies have positive expected return;
  • which risk measure suits the mandate;
  • which window or estimation frequency is correct;
  • how much leverage is sustainable;
  • which concentration, liquidity or loss limits should apply;
  • when to rebalance regardless of costs and capacity.

Those are governance choices. A mathematical solution can be exact relative to its model and unsuitable for the real portfolio.

Common mistake — Presenting “less weight in the volatile asset” as the complete definition of risk parity. The object being equalized is contribution to a selected measure, which includes dependencies and may require a joint solution.


Sources