Portfolio VaR & Risk Decomposition: Component and Marginal VaR

Published: March 30, 2026

Article by Ryan O'Connell, CFA, FRM

When managing a multi-asset portfolio, summing individual position risks dramatically overstates your true exposure. Correlations between assets create diversification effects that reduce aggregate risk. Portfolio VaR captures this reality, while risk decomposition tools — marginal VaR, component VaR, and incremental VaR — reveal which positions are driving risk and how to reduce it efficiently. This guide covers the complete framework for portfolio-level risk measurement and attribution.

What Is Portfolio VaR?

Portfolio Value at Risk (VaR) measures the loss threshold that a portfolio will not exceed over a specified time horizon at a given confidence level. For example, a 95% monthly VaR of $500,000 means there is a 5% probability that the portfolio will lose more than $500,000 in any given month.

The critical distinction is between diversified VaR (the true portfolio VaR) and undiversified VaR (the sum of standalone VaRs for each position). Undiversified VaR ignores correlations entirely, treating every position as if it moves independently. In practice, correlations — both positive and negative — determine how much risk is eliminated through diversification.

Portfolio VaR is used for internal risk limits, backtesting regulatory models, performance attribution, and strategic asset allocation. (Note: Basel market risk standards now use expected shortfall as the primary capital metric, though VaR remains important for backtesting and internal limits.) Understanding how VaR decomposes across positions is essential for identifying which VaR methodology fits your portfolio and for validating your VaR models through backtesting.

Portfolio VaR Formula (Variance-Covariance Method)

Under the parametric (variance-covariance) approach, portfolio VaR is calculated using the portfolio’s volatility, which incorporates all pairwise correlations between assets:

Where σ_p is derived from the covariance matrix:

For a two-asset portfolio, this expands to:

Where:

• α — confidence level multiplier (1.65 for 95%, 2.33 for 99%)
• w_i — weight of asset i in the portfolio
• σ_i — volatility (standard deviation) of asset i
• ρ_ij — correlation between assets i and j
• W — total portfolio value

The diversification benefit is the difference between undiversified and diversified VaR:

For details on constructing the covariance matrix, see our guides on correlation and covariance and the portfolio variance calculator.

Marginal VaR and Component VaR

Once you have portfolio VaR, the next question is: which positions are contributing to it? Two related measures answer this question from different angles.

Marginal VaR

Marginal VaR measures how portfolio VaR changes when you increase a position by a small amount. It is the partial derivative of portfolio VaR with respect to position size:

Where x_i is the dollar amount invested in asset i. Marginal VaR tells you: “If I add one more dollar to position i, how much does my portfolio VaR increase?”

Component VaR

Component VaR measures each position’s dollar contribution to total portfolio VaR:

Equivalently, component VaR can be expressed using the correlation between the position and the portfolio:

Incremental VaR

Incremental VaR measures the exact change in portfolio VaR when a new position is added or an existing position is changed by a specific amount:

While component VaR tells you each position’s current contribution to portfolio risk (an exact Euler decomposition), incremental VaR tells you the actual change in portfolio VaR from a specific proposed trade. Incremental VaR requires fully recalculating portfolio VaR with and without the trade. For small trades, the approximation Marginal VaR × trade size closely matches incremental VaR; for large trades, full revaluation is more accurate.

When to use each measure:

• Marginal VaR — For ranking positions by risk efficiency and identifying best/worst hedges (small changes)
• Component VaR — For risk attribution and understanding each position’s contribution to total VaR
• Incremental VaR — For pre-trade risk assessment of specific proposed trades (large changes)

Note that incremental VaR depends on how the trade is funded. Adding $1M to equities using new cash, selling $1M of bonds to fund it, or rebalancing the entire portfolio to maintain weights will each produce different incremental VaR values.

Component VaR vs Marginal VaR

These two measures are related but answer different questions. Understanding when to use each is essential for effective risk management.

A common point of confusion is the difference between marginal VaR and incremental VaR. Marginal VaR is the instantaneous rate of change (a derivative); incremental VaR is the actual change from a specific trade (a finite difference). For small trades, they are nearly identical. For large trades, incremental VaR is more accurate because it captures the curvature of the VaR function.

Portfolio VaR Example: 3-Asset Risk Decomposition

Let’s walk through a complete risk decomposition for a $10 million portfolio with three asset classes.

Common Mistakes in Portfolio VaR

Risk practitioners frequently encounter these errors when calculating and interpreting portfolio VaR:

1. Using Undiversified VaR as Portfolio VaR

Simply summing individual VaRs ignores correlations and dramatically overstates risk. In our example, undiversified VaR was $825,000 while true portfolio VaR was $693,000 — a 19% overstatement. When using the parametric method, calculate diversified portfolio VaR using the full covariance matrix; historical simulation and Monte Carlo methods capture diversification through the joint return distribution.

2. Assuming Correlations Are Stable

Correlations tend to increase during market stress — precisely when VaR matters most. The 0.70 correlation between US and international equities might spike to 0.90 during a crisis, significantly increasing portfolio VaR. Use stressed correlations for conservative risk limits.

3. Double-Counting Risk When Aggregating

Adding VaRs across trading desks or business units without accounting for inter-desk correlations overstates firm-wide risk. Build a consolidated covariance matrix or use consistent aggregation methods that capture diversification across the organization.

4. Using Marginal VaR Approximations for Large Position Changes

The approximation Marginal VaR × trade size works well for small trades but becomes inaccurate for large position adjustments. For trades representing more than 5-10% of portfolio value, use incremental VaR with full revaluation rather than the linear marginal VaR shortcut.

5. Treating VaR as Maximum Possible Loss

VaR is a threshold, not a ceiling. A 95% VaR of $693,000 means losses will exceed this amount 5% of the time — but says nothing about how large those exceedances might be. For tail risk, consider Expected Shortfall (CVaR), which measures the average loss when VaR is breached.

Limitations of Risk Decomposition

Linear Approximation — Marginal VaR measures the local slope of the VaR function, so using it to estimate trade impact (Marginal VaR × trade size) assumes linearity. This breaks down for portfolios with significant options exposure (where risk is non-linear) and for very large position changes. Component VaR itself is exact for the current portfolio, but applying the marginal VaR approximation to proposed trades can be misleading.

Normality Assumption — Parametric VaR assumes returns are normally distributed. In reality, asset returns exhibit fat tails (extreme events occur more often than normal distribution predicts) and skewness. This can cause VaR to understate tail risk.

Correlation Instability — The covariance matrix is estimated from historical data and assumed to be constant. Correlations change over time and tend to spike during crises, making historical estimates unreliable precisely when accuracy matters most.

Point-in-Time Measure — VaR decomposition reflects the current portfolio composition. As positions change through trading, rebalancing, or market movements, the decomposition becomes stale. Risk systems should recalculate decompositions at least daily.

No Tail Information — VaR tells you the loss threshold at a given confidence level but nothing about the severity of losses beyond that threshold. For a more complete tail risk picture, see our guide on Expected Shortfall (ES/CVaR).