Monte Carlo VaR
Monte Carlo Value at Risk (Monte Carlo VaR) estimates the maximum potential loss of a portfolio over a specified horizon at a given confidence level by simulating thousands of stochastic scenarios and identifying the loss threshold at the specified tail probability. It is the most flexible of the three primary VaR methodologies, capable of capturing nonlinear payoffs, fat-tailed distributions, and complex cross-asset correlations.
Key takeaways
- Monte Carlo VaR is the most flexible VaR methodology, capable of incorporating non-normal distributions, complex option payoffs, and fat-tailed correlation structures that parametric methods cannot model.
- The method requires explicit stochastic model specification for each risk factor, making model risk—particularly correlation breakdown and distributional misspecification—a primary concern.
- Stressed Monte Carlo VaR uses volatility and correlation parameters calibrated to historical crisis periods (2008 financial crisis, 2020 COVID shock) rather than recent calm conditions.
- Expected Shortfall (CVaR) derived from the same Monte Carlo output provides a more complete picture of tail risk than the single-percentile VaR statistic.
- Under Basel III's Fundamental Review of the Trading Book (FRTB), banks must compute stressed Expected Shortfall at 97.5% confidence over a 10-day horizon using Monte Carlo or historical simulation approaches.
Explanation
Monte Carlo VaR represents the most comprehensive approach to portfolio risk quantification available to institutional investors, superseding the simpler parametric and historical simulation approaches for complex portfolios. By generating a full distribution of simulated portfolio returns, it captures nonlinear payoffs, path dependence, and fat-tailed distributions that simpler methods fundamentally misrepresent.
The methodological advantage over historical simulation VaR is that Monte Carlo is not constrained to scenarios that have actually occurred in the historical data window. Markets can generate loss environments that have no precedent in available data—novel crisis configurations, new asset classes, unprecedented policy actions—and Monte Carlo can incorporate these via model-based extrapolation. Compared to parametric (variance-covariance) VaR, Monte Carlo avoids the restrictive normality assumption that systematically underestimates tail risk for portfolios with options, credit exposures, or naturally fat-tailed return distributions.
Implementing Monte Carlo VaR involves three key design choices. First, the stochastic model must be specified for each risk factor: geometric Brownian motion with constant volatility is the simplest but most restrictive; GARCH processes capture volatility clustering; jump-diffusion models add discrete price jumps; and stochastic volatility models (Heston, SABR) allow the volatility smile to evolve dynamically. Second, the dependency structure between risk factors must be modeled—typically via a covariance matrix calibrated to historical data, or more flexibly via copula functions that capture non-linear tail dependence. Third, option revaluation at each simulated scenario requires full mark-to-market repricing using an option pricing model, not merely linear delta approximation.
For regulatory capital under Basel III's FRTB, banks compute Expected Shortfall (not VaR) at 97.5% confidence over a 10-day liquidity horizon. Critically, FRTB requires 'stressed ES'—parameters calibrated to a 12-month stress window identified as the worst period for the specific portfolio over a long historical lookback. This prevents institutions from reporting artificially low risk estimates during calm market periods and ensures capital adequacy through the cycle.
The principal limitation of Monte Carlo VaR is model risk: the quality of risk estimates is entirely contingent on the accuracy of the stochastic model used for simulation. Correlation matrices estimated from normal periods collapse during crises—correlations that are 0.2 in calm markets may spike to 0.9 during a selloff. This correlation instability is the primary reason Monte Carlo VaR systematically underestimated risk during the 2008 financial crisis despite its theoretical flexibility. Modern implementations address this with regime-switching models, correlation stress testing, and t-copulas that build in greater tail dependence by construction.
Formula
Monte Carlo VaR = -Quantile(simulated P&L distribution, α); CVaR = -E[P&L | P&L < VaR_α]
Example
A multi-strategy hedge fund runs a Monte Carlo VaR model with 500,000 daily simulations for its $2 billion portfolio combining equity long/short positions, fixed income relative value trades, and equity index options. The model uses GARCH(1,1) volatility processes with Student-t marginal distributions linked by a t-copula with 8 degrees of freedom to capture tail dependence. The 1-day 99% Monte Carlo VaR is calculated at $28 million (1.4% of NAV). The same portfolio evaluated using parametric normal VaR yields only $19 million—a 32% underestimate reflecting the parametric model's failure to capture fat tails and nonlinear option payoffs. The CVaR at 99% from Monte Carlo output is $41 million, indicating that on the worst 1% of days, the fund expects to lose on average $41 million rather than the $28 million VaR threshold. The stressed VaR, recalibrated to 2008 parameters, rises to $67 million—the figure the risk committee uses for capital allocation decisions.
Related terms
Basel Iii Brownian Motion Concentration Risk Copula Correlation Counterparty Risk Covariance Covariance Matrix Default Delta Equity Equity Index