Parametric VaR: The Variance-Covariance Method Explained
Parametric VaR: The Variance-Covariance Method Explained
Parametric Value-at-Risk, also called the variance-covariance method, is the fastest and most elegant approach to calculating portfolio loss estimates. Rather than looking backward at historical price changes or running thousands of simulations, parametric VaR assumes your portfolio returns follow a normal distribution and uses only two inputs: volatility and correlation. Calculate the portfolio's overall volatility, apply a confidence factor from the normal distribution table, and you have your VaR in seconds.
This speed and simplicity made parametric VaR the industry standard for decades. Most major investment banks and trading firms use it as their primary risk monitoring tool. Yet its elegant assumptions hide a critical flaw: real markets don't always behave normally, especially during crashes. This article explains how parametric VaR works, when it's reliable, and where it fails.
Quick definition: Parametric VaR uses standard deviation and the normal distribution to calculate the maximum loss at a confidence level, without requiring historical simulation or Monte Carlo sampling.
Key takeaways
- Parametric VaR relies on volatility and assumes returns follow a normal bell curve
- It's fast to calculate and works well in stable markets with predictable volatility
- The confidence factor comes from the standard normal distribution (Z-score)
- Correlation between assets affects portfolio-level VaR through the covariance matrix
- The method dramatically underestimates losses during market regime changes and crashes
The Normal Distribution Assumption
Parametric VaR's foundation is the assumption that stock returns (and portfolio returns) follow a normal distribution. In a normal distribution, roughly 68% of outcomes fall within one standard deviation of the mean, 95% fall within two standard deviations, and 99.7% fall within three standard deviations.
This shape is symmetric: large gains are as likely as large losses. This assumption is convenient for mathematics—it's easy to calculate probabilities from a normal distribution. But it's a severe simplification of reality. In actual markets, extreme moves happen more often than the normal distribution predicts. Crashes cluster together. A sudden news event can move a market 5% in a day when the normal distribution says such moves should occur only once per year.
However, for routine risk measurement on liquid portfolios in calm markets, the normal distribution approximation works reasonably well most of the time. This is why parametric VaR remains useful despite its limitations.
The Mathematics: From Volatility to Loss
The core formula for parametric VaR is straightforward. For a single-asset portfolio:
VaR = Portfolio Value × Z-score × Daily Volatility
Let's break each component:
Portfolio Value is your current investment amount. If you hold a $100,000 stock portfolio, that's your starting number.
Volatility measures how much daily returns swing around the average. It's expressed as a percentage—for example, 2% daily volatility for a stock that typically moves 2% per day. Volatility is calculated as the standard deviation of historical daily returns, usually over the past 252 trading days (one year).
Z-score comes from the normal distribution table. For a 95% confidence level, the Z-score is 1.645. For 99% confidence, it's 2.326. These numbers say: if returns are normally distributed, about 1.645 standard deviations capture 95% of outcomes, and 2.326 standard deviations capture 99%.
Suppose you hold $100,000 in a single stock with 2% daily volatility. Your 95%, 1-day VaR is:
$100,000 × 1.645 × 0.02 = $3,290
This says there's a 95% chance you won't lose more than $3,290 tomorrow. The 99% VaR would be:
$100,000 × 2.326 × 0.02 = $4,652
Much larger, because you're protecting against a worse-case scenario. The relationship is linear: double the volatility, and VaR doubles. Double the confidence level's Z-score, and VaR doubles.
Multi-Asset Portfolios: The Covariance Matrix
Most portfolios hold multiple assets. The power of parametric VaR emerges here: it accounts for correlation without simulation. Assets that move together amplify portfolio risk; assets that move oppositely reduce it.
The parametric method uses a covariance matrix—a table that captures how each asset's returns move with every other asset's returns. For a two-asset portfolio, the formula becomes:
Portfolio Volatility = sqrt(w1^2 * vol1^2 + w2^2 * vol2^2 +
2 * w1 * w2 * correlation * vol1 * vol2)
Where w1 and w2 are portfolio weights, vol1 and vol2 are individual volatilities, and correlation ranges from -1 (perfect negative correlation) to +1 (perfect positive correlation).
Example: A portfolio with $60,000 in Stock A (1.5% daily volatility) and $40,000 in Stock B (2.0% daily volatility), with 0.6 correlation:
Weights: w1 = 0.6, w2 = 0.4
Portfolio Vol = sqrt((0.6)^2 × (0.015)^2 + (0.4)^2 × (0.020)^2 +
2 × 0.6 × 0.4 × 0.6 × 0.015 × 0.020)
= sqrt(0.000081 + 0.000064 + 0.000086)
= sqrt(0.000231)
= 0.0152 = 1.52% daily volatility
Portfolio value is $100,000. Your 95%, 1-day VaR:
$100,000 × 1.645 × 0.0152 = $2,499
Notice that the portfolio's volatility (1.52%) is lower than a simple average of the two stocks' volatilities (1.75%). This is diversification at work. The 0.6 correlation means the stocks don't move in perfect lockstep, so they partially hedge each other.
If the stocks were perfectly correlated (1.0), portfolio volatility would be the weighted average: 1.5% × 0.6 + 2.0% × 0.4 = 1.7%. If they were uncorrelated (0.0), portfolio volatility would be:
sqrt((0.6)^2 × (0.015)^2 + (0.4)^2 × (0.020)^2) = sqrt(0.000145) = 1.20%
Much lower. This illustrates the fundamental insight: negative or low correlations create risk reduction through diversification.
Time-Scaling Volatility
One-day volatility is the most common input. But parametric VaR scales easily to longer time horizons using the square root of time rule:
10-day Volatility = 1-day Volatility × sqrt(10)
If your portfolio has 1.5% daily volatility, the 10-day volatility is approximately:
1.5% × sqrt(10) = 1.5% × 3.162 = 4.74%
So a $100,000 portfolio's 95%, 10-day VaR is:
$100,000 × 1.645 × 0.0474 = $7,793
The square root of time assumes volatility is constant and returns are independent day to day (no momentum or mean reversion). In practice, both assumptions break down over longer periods. Still, the square root of time rule is a quick, industry-standard approach that works reasonably well for 1-10 day horizons.
Advantages of Parametric VaR
Speed: A parametric VaR calculation for a 100-asset portfolio takes microseconds. You can update risk metrics in real-time as positions change.
Transparency: The calculation is fully transparent. You can explain exactly why portfolio VaR increased (higher volatility, higher correlation, larger positions) without a "black box" simulation.
Scalability: The math extends from single positions to complex portfolios with thousands of assets, as long as you have a covariance matrix.
Stability: Unlike historical VaR, which changes abruptly when one bad historical day rolls off the lookback window, parametric VaR updates smoothly as volatility estimates adjust.
The Critical Flaw: Non-Normal Returns
The method's greatest weakness is its assumption of normality. Real stock returns have "fat tails"—extreme moves occur far more frequently than a normal distribution predicts.
In 2011, the S&P 500 fell 5.5% on August 5th. Under a normal distribution assumption with historical volatility estimates, a 5.5% daily loss should occur once every 50+ years. Yet market crashes and sudden policy shocks produce such moves every few years.
This means parametric VaR systematically underestimates risk in three scenarios:
Market crashes: When volatility spikes suddenly, yesterday's volatility estimates are obsolete. Parametric VaR responds slowly because volatility must be recalculated, and recent calm days are still averaged into the estimate.
Tail correlations: During crashes, correlations spike toward 1.0. Stock pairs that usually move independently suddenly move together. Parametric VaR, using average historical correlation, underestimates the risk that all your "diversified" holdings will fall at once.
Regime changes: Volatility is not constant. Normal market periods alternate with high-volatility crisis periods. A volatility estimate from calm days becomes dangerously optimistic when a regime change occurs.
The 2008 financial crisis exposed this flaw. Many banks' parametric VaR models said the risk was manageable in early 2008. Actual losses far exceeded VaR estimates across the industry because the market behavior violated the normal distribution assumption so badly.
When Parametric VaR Works Well
Despite its limitations, parametric VaR works reliably in several contexts:
Large, liquid, domestic stock portfolios: U.S. large-cap stocks are relatively normally distributed on 1-3 day horizons. A parametric VaR for a 100-stock S&P 500 portfolio will usually be reasonably accurate.
Bond portfolios with stable duration: Bond returns are more normally distributed than stocks, especially over short horizons. Parametric VaR for bonds tends to be more reliable.
Routine monitoring: Even if parametric VaR occasionally misses tail risks, it's excellent for day-to-day risk tracking. It flags rising volatility quickly and helps you notice when markets are becoming more dangerous.
Real-time risk updates: Because it's so fast, parametric VaR is the natural choice for intraday risk dashboards and live position monitoring.
Real-World Example: A Tech-Heavy Portfolio
Suppose you hold a $200,000 portfolio: 60% in a tech ETF (QQQ), 30% in a broad market ETF (SPY), and 10% cash.
Calculate historical volatilities:
- QQQ: 2.8% daily volatility
- SPY: 1.6% daily volatility
- Cash: 0% volatility
Correlation QQQ-SPY: 0.75 Correlation QQQ-Cash: 0.0 Correlation SPY-Cash: 0.0
Weights: w_QQQ = 0.6, w_SPY = 0.3, w_Cash = 0.1
Portfolio volatility calculation (simplified, two correlated assets plus cash):
Portfolio Vol = sqrt((0.6)^2 × (0.028)^2 + (0.3)^2 × (0.016)^2 +
2 × 0.6 × 0.3 × 0.75 × 0.028 × 0.016)
= sqrt(0.000284 + 0.000023 + 0.000048)
= sqrt(0.000355)
= 0.0188 = 1.88% daily volatility
Your 95%, 1-day VaR:
$200,000 × 1.645 × 0.0188 = $6,179
Your 99%, 1-day VaR:
$200,000 × 2.326 × 0.0188 = $8,747
This says: under normal conditions, there's a 95% chance the portfolio won't lose more than $6,179 tomorrow, and a 99% chance it won't lose more than $8,747. During the 2020 COVID crash, the actual daily loss exceeded both estimates many times over, showing the limitation of parametric assumptions during market dislocations.
Combining Parametric VaR with Stress Testing
The solution is not to replace parametric VaR but to supplement it. Professionals use parametric VaR for routine monitoring, then stress-test against hypothetical scenarios: What if the tech sector falls 10%? What if the Fed raises rates 75 basis points? What if volatility doubles? These stress tests catch tail risks that parametric VaR's normal distribution assumption might miss.
Common Mistakes with Parametric VaR
Using stale volatility: If you calculate volatility once per month, you miss regime changes. Volatility should be updated daily or use an exponentially weighted average that emphasizes recent observations.
Ignoring correlation changes: Correlation is not constant. During crashes, correlations spike. Use rolling correlations or stress-test against tail correlation scenarios.
Applying parametric VaR to options portfolios: Options have highly non-normal payoffs. Parametric VaR severely underestimates tail risk for short option positions. Use Monte Carlo or historical simulation instead.
Trusting parametric VaR alone: Never rely solely on parametric VaR for risk governance. Combine it with historical VaR, stress testing, and backtesting.
FAQ
How do I update volatility estimates?
The standard is a 252-day lookback (one trading year), recalculated daily. For faster response to regime changes, use exponential weighting that gives recent days more weight. Or, if markets are clearly volatile, switch to a 63-day lookback.
What if my portfolio has negative correlations?
Negative correlations reduce portfolio volatility, which is good for diversification but bad for risk measurement accuracy. Parametric VaR will show low risk even though diversification only helps if the negatively correlated asset actually moves in the opposite direction when needed. Stress-test against correlation changes.
Can I use parametric VaR for intraday trading?
Yes, but use intraday volatility (5-minute or 15-minute returns), not daily. Intraday volatility is typically lower and more stable than daily volatility.
How does parametric VaR handle gaps at market open?
It doesn't explicitly. The volatility estimate includes overnight gaps, but if a major news event occurs after hours, parametric VaR won't adjust until the next day's calculation. This is a limitation; use a pre-market stress test if gap risk is material.
Should I use parametric VaR for emerging markets?
With caution. Emerging market returns have fatter tails than developed markets. Parametric VaR will underestimate tail risk. Historical simulation or Monte Carlo with fat-tailed distributions is safer.
How often do actual losses exceed parametric VaR?
In normal markets, a 95% VaR should be exceeded about 5% of the time. If you're seeing more exceedances, either your volatility estimate is stale or your market regime has shifted. Check if volatility has increased and if correlations have changed.
What's a typical Z-score for different confidence levels?
90% confidence: 1.282 95% confidence: 1.645 99% confidence: 2.326 99.9% confidence: 3.090
Use 95% for routine monitoring and 99% for stress scenarios.
Related concepts
- What Is Value-at-Risk?
- Historical VaR: Using Past Returns
- Monte Carlo VaR: Simulating the Future
- Understanding Correlation
Summary
Parametric VaR is the fastest, most transparent way to calculate portfolio loss estimates. By assuming normal distributions and using volatility and correlation, it delivers a single number in milliseconds. This speed made it the industry standard for decades. However, the normal distribution assumption is a severe simplification; real markets have fat tails, regime changes, and tail correlations that violate the model. Parametric VaR works well for routine monitoring of stable portfolios but systematically underestimates risk during market crashes and regime shifts. The solution is to use parametric VaR as your primary metric while supplementing it with historical simulation, Monte Carlo analysis, and stress-testing to catch tail risks.