The VaR parametric method — also called the variance-covariance approach or delta-normal method — is one of the most widely used techniques for calculating Value at Risk. It estimates potential portfolio losses using just two statistical parameters: the mean return and standard deviation. This guide explains how the parametric method works, walks through the formula and a worked example, compares it with other VaR methods, and covers common implementation pitfalls.

What Is the VaR Parametric Method?

The parametric method calculates Value at Risk by assuming that portfolio returns follow a known probability distribution — typically the normal (Gaussian) distribution. Rather than simulating scenarios or replaying historical data, it derives VaR directly from the statistical parameters of the return distribution.

Key Concept

The parametric method is a closed-form, analytical approach to VaR. Given an estimate of the portfolio’s mean return (μ) and standard deviation (σ), it uses the z-score for your chosen confidence level to calculate the maximum expected loss. It is also called the variance-covariance method because multi-asset portfolios require the full covariance matrix to compute portfolio volatility.

The term “parametric” means the method relies on parameters of an assumed distribution — specifically μ and σ for the normal distribution. This stands in contrast to the historical method, which uses raw past returns with no distributional assumption, and the Monte Carlo method, which simulates thousands of random scenarios. Because the parametric method uses a formula rather than simulation, it is fast and computationally efficient — but it depends on the normality assumption holding.

Video: VaR Parametric Method Explained

How the Variance-Covariance Approach Works

The parametric method follows a straightforward process. For a single asset or a portfolio with linear exposures, the steps are:

  1. Collect historical return data — gather daily (or weekly/monthly) returns for the asset or portfolio over a representative period
  2. Estimate the mean return (μ) and standard deviation (σ) — these are the two parameters that define the normal distribution
  3. Choose a confidence level — typically 95% or 99%, depending on regulatory or internal risk requirements
  4. Look up the corresponding z-score — the one-tailed z-score for your confidence level (1.645 for 95%, 2.326 for 99%)
  5. Apply the parametric VaR formula — combine z, σ, μ, and the portfolio value to get the VaR estimate
Key Assumptions

The parametric method relies on four core assumptions: (1) returns are normally distributed, (2) portfolio payoffs are linear in risk factors (delta-normal), (3) the covariance matrix is stable over the estimation window, and (4) returns are independent and identically distributed (i.i.d.) for time-scaling purposes.

Pro Tip

Always match your data frequency to your VaR horizon. If you want daily VaR, estimate μ and σ from daily returns. If you use monthly data to calculate a daily VaR, you’ll need to scale the parameters — and scaling introduces additional approximation error.

The Parametric VaR Formula

The parametric VaR formula expresses the maximum expected loss at a given confidence level. Using the loss-positive convention (VaR is reported as a positive number representing a potential loss):

Parametric VaR Formula
VaRα = (zα × σ – μ) × V
The z-score times volatility minus the mean return, scaled by portfolio value

For short time horizons (daily VaR), the mean return μ is typically very small relative to the volatility term and can be approximated as zero. This gives the simplified form used in most practical applications:

Simplified Daily VaR (μ ≈ 0)
VaRα ≈ zα × σ × V
For daily VaR, dropping the mean return is a common and reasonable simplification

Where:

  • zα — the one-tailed z-score for confidence level α (1.645 for 95%, 2.326 for 99%)
  • σ — standard deviation of portfolio returns (matching the VaR time horizon)
  • μ — mean portfolio return over the same period
  • V — current portfolio value

To convert daily VaR to a multi-day horizon, the square root of time rule is commonly used. This approximation assumes returns are i.i.d. and works best when the mean return is negligible:

VaR Time Scaling (Square Root of Time Rule)
VaRT-day ≈ VaR1-day × √T
Scale 1-day VaR to T-day VaR — an approximation assuming i.i.d. returns and μ ≈ 0

Parametric Method Example

$500K Portfolio — 99% 1-Day VaR

Given:

  • Portfolio value (V): $500,000
  • Daily mean return (μ): 0.04% (0.0004)
  • Daily standard deviation (σ): 1.2% (0.012)
  • Confidence level: 99% → zα = 2.326

Calculation:

VaR = (zα × σ – μ) × V

VaR = (2.326 × 0.012 – 0.0004) × $500,000

VaR = (0.027912 – 0.0004) × $500,000

VaR = 0.027512 × $500,000

VaR = $13,756

Interpretation: There is a 99% probability that this portfolio will not lose more than $13,756 in a single trading day. Equivalently, on 1 out of every 100 trading days, losses are expected to exceed this amount.

10-day VaR (using the square root rule): $13,756 × √10 = $13,756 × 3.162 ≈ $43,496

Parametric vs Monte Carlo Method

The parametric and Monte Carlo methods take fundamentally different approaches to estimating VaR. Here is how they compare:

Parametric (Variance-Covariance)

  • Speed: Fast, closed-form calculation
  • Distribution: Assumes normality
  • Complexity: Simple for linear portfolios
  • Tail risk: May underestimate extreme losses
  • Best for: Liquid assets with approximately normal returns

Monte Carlo Simulation

  • Speed: Computationally intensive (thousands of simulations)
  • Distribution: Can model any distribution
  • Complexity: Handles non-linear instruments (options, derivatives)
  • Tail risk: Can capture fat tails and skewness if modeled
  • Best for: Complex portfolios with derivatives or non-normal exposures

For portfolios of stocks and bonds with roughly normal returns, the parametric method is often the practical choice due to its speed and simplicity. For portfolios containing options or structured products, the Monte Carlo method is more appropriate because it can capture non-linear payoff profiles. For a comparison with the non-parametric approach, see our guide on the VaR historical method.

Portfolio VaR with Multiple Assets

For a portfolio with multiple assets, calculating parametric VaR requires the variance-covariance matrix — this is where the method gets its alternative name. The portfolio variance accounts for each asset’s individual risk and the correlations between them:

Portfolio Variance
σp2 = w′Σw
Portfolio variance equals the weight vector transposed times the covariance matrix times the weight vector

For a two-asset portfolio, this expands to:

σp2 = w12σ12 + w22σ22 + 2w1w2σ1σ2ρ12

Where w1 and w2 are the portfolio weights, σ1 and σ2 are the individual asset volatilities, and ρ12 is their correlation coefficient. The key insight: when ρ12 < 1, portfolio volatility is less than the weighted sum of individual volatilities. This diversification benefit directly reduces portfolio VaR.

Pro Tip

Diversification lowers portfolio VaR because imperfectly correlated assets partially offset each other’s losses. The parametric method captures this benefit through the covariance matrix — which is one of its key advantages over simply summing individual asset VaRs.

How to Implement the Parametric Method

You can implement the parametric VaR method in a spreadsheet in five steps:

  1. Download historical prices — get daily closing prices for your asset(s) from Yahoo Finance or another data provider
  2. Calculate daily returns — compute log returns: ln(Pt / Pt-1). Be consistent — if you estimate μ and σ from log returns, apply the formula to log returns
  3. Compute the mean (μ) and standard deviation (σ) — use Excel’s AVERAGE() and STDEV.S() functions on the return series
  4. Get the z-score — use =-NORM.S.INV(0.01) for 99% VaR (returns 2.326) or =-NORM.S.INV(0.05) for 95% VaR (returns 1.645). The negative sign converts the left-tail quantile to a positive z-score
  5. Apply the formula — multiply (z × σ – μ) by your portfolio value to get the dollar VaR

Use an estimation window of at least 250 trading days (one year) to capture a meaningful range of market conditions. Windows that are too short or drawn entirely from calm periods will understate risk. For model validation, compare VaR predictions against actual losses — if losses exceed VaR significantly more than expected (e.g., more than 1% of days for 99% VaR), the model needs recalibration.

For a comprehensive overview of all three VaR methods, see our Value at Risk guide. For a Python implementation of the parametric method, see Parametric Method: VaR in Python.

Video: Parametric Method — Value at Risk (VaR) in Excel

To learn the parametric method and all three VaR approaches through structured video lessons and Excel exercises, enroll in our Value at Risk course.

Common Mistakes with the Parametric Method

These are the most frequent errors practitioners make when applying the variance-covariance VaR method:

1. Assuming normality when returns have fat tails. Financial returns often exhibit leptokurtosis — extreme gains and losses occur more frequently than a normal distribution predicts. Applying the parametric method without acknowledging this systematically underestimates tail risk. Consider supplementing parametric VaR with Expected Shortfall (CVaR) for a more complete picture of tail losses.

2. Using the wrong z-score. VaR uses a one-tailed z-score because we only care about losses in one direction. The 99% one-tailed z-score is 2.326 (not 2.576, which is two-tailed). Mixing these up produces materially incorrect VaR estimates.

3. Scaling VaR incorrectly across time horizons. To convert daily VaR to a 10-day figure, multiply by √10 — not by 10. The square root of time rule comes from the property that variance scales linearly with time for i.i.d. returns, so standard deviation (and therefore VaR) scales with the square root.

4. Ignoring correlation changes during market stress. Correlations between assets tend to spike during crises — precisely when diversification benefits are needed most. Covariance matrices estimated from calm periods understate the risk of simultaneous losses across positions.

5. Mixing log returns and simple returns. If you estimate μ and σ from log returns, you must apply the parametric formula to log returns (and vice versa). Mixing the two introduces a systematic bias in the VaR estimate, especially for volatile assets or longer horizons.

Limitations of the Parametric Method

Important Limitation

The parametric method’s reliance on the normal distribution assumption is its biggest weakness. Financial returns consistently exhibit fatter tails and more negative skewness than the normal distribution predicts, meaning the parametric method tends to underestimate the probability and magnitude of extreme losses.

1. Normality assumption rarely holds. Empirical studies consistently show that financial returns have excess kurtosis and negative skewness. The parametric method cannot capture these features without modification (e.g., using a Student’s t-distribution instead).

2. Underestimates tail risk. Because the normal distribution assigns very low probabilities to extreme events, parametric VaR underestimates both the frequency and severity of large losses. During the 2008 financial crisis, several major banks experienced daily losses that far exceeded their parametric VaR estimates.

3. Correlations are unstable. The covariance matrix used for multi-asset VaR is typically estimated from a historical window. During market stress, correlations shift — often increasing dramatically — making the historical covariance matrix unreliable exactly when accuracy matters most.

4. Poor for non-linear instruments. Options, convertible bonds, and structured products have non-linear payoff profiles that violate the delta-normal assumption. For portfolios with significant options exposure, the Monte Carlo method is more appropriate.

5. VaR is a single-point estimate. VaR tells you the threshold loss at a given confidence level, but it says nothing about how bad losses could be beyond that threshold. For a measure that captures average tail losses, see Expected Shortfall (CVaR).

Frequently Asked Questions

Because the method’s core inputs are the variance of asset returns (for single-asset VaR) and the covariance matrix (for multi-asset portfolio VaR). The variance measures individual asset risk, while the covariance matrix captures how assets move together — both of which are statistical parameters of the assumed normal distribution. The name reflects the fact that these variance and covariance estimates are what drive the entire calculation.

Use the square root of time rule: multiply daily VaR by √T, where T is the number of trading days. For example, 10-day VaR ≈ daily VaR × √10 ≈ daily VaR × 3.162. This approximation assumes returns are independent and identically distributed (i.i.d.) and works best when the daily mean return is negligible. For longer horizons where drift (μ) matters, the full scaling formula is VaRT = (zα × σ × √T – μ × T) × V.

The normality assumption breaks down during market crises, earnings surprises, geopolitical shocks, and other tail events. Financial returns consistently exhibit leptokurtosis (fat tails) and negative skewness, meaning extreme losses happen more frequently and are larger than a normal distribution predicts. For example, during the 2008 financial crisis, multiple trading days produced losses well beyond what normal-distribution-based VaR models considered plausible. When normality is a concern, consider using a Student’s t-distribution or switching to the Monte Carlo method.

Through the covariance matrix. When assets in a portfolio are not perfectly correlated (ρ < 1), the portfolio’s combined volatility is lower than the weighted sum of individual asset volatilities. The parametric method captures this diversification benefit directly in the portfolio variance calculation: σp2 = w′Σw. This means portfolio VaR is typically less than the sum of individual asset VaRs — a natural reflection of how diversification reduces risk.

Use the parametric method when you need a fast, analytically tractable estimate and your portfolio consists of liquid assets with approximately normal returns. Use the historical method when you want to avoid distributional assumptions entirely — it uses actual past returns and naturally captures fat tails, skewness, and non-linear effects present in the data. The trade-off is that the historical method requires a sufficiently long and representative data history, while the parametric method only needs μ and σ estimates. Many practitioners run both methods as a cross-check.

Disclaimer

This article is for educational and informational purposes only and does not constitute investment advice. VaR estimates are based on statistical assumptions and historical data that may not reflect future market conditions. Always conduct your own analysis and consult a qualified financial professional before making risk management decisions.