Monte Carlo Simulation in Risk Assessment

Monte Carlo simulation is the workhorse of modern risk analysis. By generating thousands of possible future paths for a portfolio under an assumed return process, it produces a full distribution of outcomes that closed-form formulas cannot — particularly when positions contain options, path-dependent payoffs, or nonlinear relationships with risk factors.

The core idea

Given a model for how risk factors evolve, simulate many independent paths forward, reprice the portfolio along each path, and read statistics off the resulting distribution. The law of large numbers guarantees that empirical statistics converge to the model's true values as the number of paths grows.

For a portfolio with value $V_t$ depending on risk factors $X_t$, a single path produces a terminal value $V_T = f(X_T)$ where $X_T$ is simulated from the assumed process. Repeating $N$ times yields a sample from which any statistic — VaR, expected shortfall, probability of ruin — can be estimated.

Choosing the return process

• Geometric Brownian motion: $dS = \mu S,dt + \sigma S,dW$. Simple, but assumes log-normal returns and constant volatility — both empirically false.
• Stochastic volatility models (Heston, GARCH): Let volatility itself follow a random process, capturing volatility clustering.
• Jump-diffusion models: Add discontinuous jumps to Brownian motion to capture crash risk.
• Historical bootstrap: Resample actual historical returns, preserving the empirical distribution without assuming a parametric form.

The model is the assumption. A Monte Carlo result is only as honest as the process generating it.

Applications for traders

• Path-dependent payoffs: Options with barriers, lookbacks, Asians, and structured products have no closed-form pricing under realistic processes. Monte Carlo prices them directly.
• VaR and Expected Shortfall for portfolios with options: Reprice each path through the option valuation, capturing gamma and vega that parametric VaR ignores.
• Strategy robustness: Simulate a strategy across thousands of randomly generated price paths to estimate the distribution of terminal wealth, drawdown depth, and recovery time.

Estimating accuracy

Monte Carlo estimates carry simulation error. The standard error of an estimated quantile $\hat{q}_\alpha$ is approximately:

$$\text{SE}(\hat{q}\alpha) \approx \frac{1}{f(q\alpha)}\sqrt{\frac{\alpha(1-\alpha)}{N}}$$

where $f(q_\alpha)$ is the density at the quantile. Estimating tail quantiles accurately requires many more paths than estimating the mean — often tens of thousands for a stable 99% VaR.

Pitfalls

• Model risk dominates simulation error. The biggest uncertainty is the process, not the number of paths. Report results across multiple model assumptions.
• Correlation structure matters enormously. Simulating factors independently destroys the clustering that drives portfolio crises. Use a full covariance matrix or a copula.

Run Monte Carlo with at least three different return processes and act only if all three agree the portfolio is survivable. Monte Carlo makes uncertainty visible, forcing the trader to confront the full distribution of outcomes.