Value at Risk &
Expected Shortfall
Move beyond per-trade stops and measure the loss distribution of an entire portfolio—especially the outcomes hiding in its left tail.
VaR
Threshold
ES / CVaR
Tail average
Horizon
Time window
Confidence
Coverage
VaR marks the door.
ES looks through it.
VaR identifies a percentile boundary. It does not describe how severe losses become beyond that boundary. Expected Shortfall averages the observations inside the tail.
Value at Risk asks
“At this confidence level and horizon, what loss threshold should the portfolio cross only rarely?” A 95% one-day VaR still anticipates a breach on roughly one day in twenty.
Expected Shortfall asks
“When the threshold is breached, how large is the average loss?” That makes the severity of tail events visible instead of stopping at a single percentile.
THREE ESTIMATION LENSES
The method shapes the answer
Parametric
Uses volatility and a chosen distribution to estimate a percentile. Efficient, but highly dependent on its assumptions.
Useful for
Stable portfolios and rapid monitoring
Watch for
Normality can understate fat-tail risk
Historical simulation
Replays actual historical returns against today's portfolio and ranks the resulting gains and losses.
Useful for
Capturing nonlinear moves seen in the sample
Watch for
The future may not resemble the lookback window
Monte Carlo
Generates many hypothetical paths from a chosen process, then measures the simulated loss distribution.
Useful for
Complex portfolios and scenario exploration
Watch for
A sophisticated engine can still have poor assumptions
Parametric tail-risk estimator
Estimate loss thresholds from portfolio value and daily volatility under a normal-distribution assumption. This simplified model excludes drift, fat tails, changing correlations, and liquidity costs.
Model output
$2,468
Losses should exceed this threshold about 5% of periods under the model.
$3,093
Estimated average loss when the VaR boundary is breached.
A risk number needs a challenge process
No single statistic is a safety guarantee. Pair the model with scenario analysis, exposure limits, and a record of actual breaches.
Lesson complete
Use thresholds and tails together
VaR supports limits. Expected Shortfall exposes the damage beyond them. Stress tests ask what neither model has seen.