Monte Carlo Simulation for Quant Finance — Overview — Wiki | SOPHIE

Overview

Monte Carlo Simulation (MCS) estimates the outcomes of uncertain events by modeling their full probability distribution through repeated random sampling, rather than deriving a single deterministic outcome from average inputs. Originating from Stanislaw Ulam and John Von Neumann's work on the Manhattan Project, it is now a cornerstone of quantitative finance — but it cannot capture genuine alpha, since it implicitly assumes market efficiency.

Key Concepts

The MCS Workflow — (1) model and calibrate the SDE from historical data, (2) parameterize input distributions, (3) generate random samples via RNGs, (4) run the simulation across N paths and M time steps, (5) perform statistical inference on the resulting empirical distribution.
MC vs. Machine Learning — complementary, not competing: ML discovers potential alpha structures from historical data; MC verifies a strategy's stability against randomness and estimates its risk profile across synthetic scenarios.
Weak vs. Strong Convergence — financial applications typically only need weak convergence (accuracy of the expected value), which allows faster, coarser simulation than strong convergence (path-level accuracy) would require.

Critical Limitations

Crisis Underestimation — standard MC models assuming normal returns systematically underestimate the probability and severity of Black Swan events; switching to a Student's t distribution helps capture fat tails.
The Alpha Dilemma — MC's core limitation: it implicitly assumes perfectly efficient markets and random-walk price movements, modeling only passive Beta returns and structurally excluding any informational or systematic edge a quant strategy relies on.
Variance Reduction Techniques (VRTs) — Control Variates (reference a similar option with known analytical price), Antithetic Variates (simulate Z and -Z path pairs to induce negative correlation), Common Random Numbers (share seeds across related calculations, e.g. Greeks estimation, to reduce variance in the difference).

Technical Methodologies

Stochastic Differential Equations (SDEs) — Geometric Brownian Motion is the standard model (dX_t = a(X_t)dt + σ(X_t)dB_t), discretized via the Euler-Maruyama scheme for simulation.
Bootstrapping — Simple historical bootstrapping (I.I.D. resampling with replacement), Block Bootstrapping (preserves temporal structure like volatility clustering), and Filtered Historical Simulation (GARCH-extracted residuals bootstrapped and recombined with forecasted volatility) trade off simplicity against realism.
Path Perturbation for Robustness — Trade Order Shuffling (tests path-dependency sensitivity), Parameter Jittering (tests sensitivity to latency/slippage assumptions), and MACHR block randomization (tests robustness to radically different historical regime sequences).

Applications in Quant Finance

VaR/CVaR — computed under the Physical (ℙ) Measure using historical drift; Cholesky Decomposition generates correlated random returns from the historical covariance matrix. CVaR (Expected Shortfall) is superior to VaR since it averages losses beyond the VaR threshold rather than just marking it.
Strategy Validation — MC exposes "lucky backtests" (overfitting to one historical path) by producing a full distribution of performance metrics (e.g. Sharpe, Max Drawdown) instead of a single point estimate; strategies should be judged on worst-case simulated performance (e.g. 5th percentile Sharpe), not the single realized backtest number.
Path-Dependent Derivatives — the only practical method for pricing exotics (Asian, Barrier, Lookback) whose payoffs depend on the entire price path; priced under the Risk-Neutral (ℚ) Measure by averaging discounted payoffs across simulated paths.

Key Takeaways