Risk Budgeting and Marginal Risk Contribution

A capital weight says little about what drives portfolio losses. Risk budgeting assigns each asset, sleeve, or factor a target share of portfolio volatility (or another coherent risk measure). Its basic diagnostic is marginal risk contribution (MRC): how much portfolio volatility changes when an exposure is increased infinitesimally.

From covariance to component risk

For weights \(w\) and covariance matrix \(Σ\), portfolio volatility is \(σ_p=\sqrt{w'Σw}\). Differentiating gives:

MRC_i = (Σw)_i / σ_p
CRC_i = w_i · MRC_i

CRCi is component risk contribution, and the Euler identity guarantees that the CRCs sum to portfolio volatility. A risk-budgeted portfolio targets CRCi / σp = bi, where budgets b sum to one. Equal risk contribution is the special case with equal budgets; it is not a claim that every asset deserves equal capital.

Portfolio	Capital weight	Volatility	Correlation	Risk share
Equity	50%	20%	0.2 to bonds	Often dominant
Bonds	50%	7%	0.2 to equity	Often modest

This is why a 60/40 allocation is usually an equity-risk allocation. It also explains why naively inverse-volatility weights are insufficient: correlations determine MRC.

import numpy as np

def risk_contributions(w, cov):
    vol = np.sqrt(w @ cov @ w)
    mrc = cov @ w / vol
    crc = w * mrc
    return vol, mrc, crc / vol

Solving for a budget

With long-only weights, solve a constrained nonlinear problem such as minimizing the squared gap between achieved and target risk shares. Normalize weights after every step and constrain position, leverage, turnover, and group exposures. Exact equalization is rarely worth violating these real-world constraints. A robust optimizer reports the residual budget error so the investment committee knows which budget is aspirational.

Risk budgets can be hierarchical: allocate 40% to equity factors, 30% to rates, 20% to credit, and 10% to diversifiers; then allocate inside each sleeve. This avoids an asset universe with twenty correlated equities receiving twenty separate votes. Covariance shrinkage is essential; unstable correlations produce unstable MRCs and forced turnover.

Beyond volatility

Volatility is symmetric and can understate a short-option or credit allocation's bad-state risk. The same budgeting logic works with expected shortfall, but the marginal contribution must be estimated from tail scenarios. For scenario loss L, component ES is approximately the conditional average position loss on days where portfolio loss exceeds its VaR threshold. It is noisier than volatility contribution, so use longer windows, filtered scenarios, and explicit stress tests alongside it.

Regime changes matter. During a flight to quality, correlation assumptions can converge toward one and bond hedges can fail. Maintain normal and stressed covariance/risk budgets, then ask which allocation is feasible when liquidity and margins worsen simultaneously.

Implementation discipline

The model is only one layer of the trade. Store input timestamps, vendor identifiers, contract specifications, FX conversion conventions, and the exact version of each risk model alongside every signal. A result that cannot be reproduced from raw inputs is not a research result; it is an anecdote. Use point-in-time constituent data, distinguish an indicative quote from an executable quote, and make the decision clock explicit. For an execution-sensitive strategy, a close price or composite midpoint is usually an unattainable benchmark rather than a fill.

Evaluate the full distribution, not just a headline Sharpe: drawdown, tail loss, turnover, capacity, exposure concentration, and degradation during the stressed periods that motivated the idea. Split design, parameter choice, and final evaluation across separate samples. If alternatives were explored, record them and account for the search using the Deflated Sharpe Ratio. Finally, run a paper or shadow phase with the same order, reconciliation, and limit logic planned for production. A backtest should establish a conditional expectation, not a promise of a tradable return.

Key takeaways

• MRC is the derivative of portfolio volatility; CRC allocates total volatility to positions.
• Equal capital and equal risk are different, especially across equity and bond sleeves.
• Optimize to risk budgets with practical constraints and report residual errors.
• Use factor or hierarchical budgets to avoid counting correlated holdings repeatedly.
• Tail and stressed contributions complement volatility-based budgeting.