Quasi-Monte Carlo
Quasi-Monte Carlo (QMC) is a numerical integration and simulation technique that replaces the pseudo-random number sequences used in standard Monte Carlo methods with low-discrepancy sequences — such as Sobol, Halton, or Faure sequences — that fill the sampling space more uniformly, achieving substantially faster convergence and greater accuracy for high-dimensional financial problems such as derivatives pricing, portfolio risk estimation, and scenario generation. Unlike true Monte Carlo, whose error decreases at O(N^{-1/2}), QMC achieves convergence rates approaching O(N^{-1}) in well-behaved problems.
Key takeaways
- QMC replaces pseudo-random samples with deterministic low-discrepancy sequences that fill the unit hypercube more uniformly, reducing clustering and gaps.
- For smooth integrands in moderate dimensions (up to ~50), QMC typically outperforms Monte Carlo by one to two orders of magnitude in accuracy for the same number of function evaluations.
- The convergence advantage of QMC diminishes in very high dimensions (hundreds or thousands) due to the 'curse of dimensionality,' though modern scrambled QMC methods partially mitigate this.
- Sobol sequences are the most widely used QMC sequences in finance, particularly for pricing path-dependent options, CDO tranches, and calculating VaR.
- QMC is deterministic, so standard error estimates (standard deviation of sample mean) cannot be computed directly; scrambled QMC variants restore randomization to enable error bounds while retaining much of the convergence advantage.
Explanation
Standard Monte Carlo simulation in finance generates random scenarios by drawing from pseudo-random number generators (PRNGs). While PRNGs produce sequences that pass statistical tests for randomness, they can cluster in localized regions of the sampling space, particularly in moderate dimensions (e.g., 10–100 state variables), leading to simulation error that converges slowly at O(N^{-1/2}). For a simulation requiring, say, 100,000 paths, standard MC achieves an error reduction factor of 1/√100,000 ≈ 0.003 — but doubling the sample size to 200,000 only reduces error by a further factor of 1/√2 ≈ 0.7.
Quasi-Monte Carlo addresses this through the theory of equidistribution. A low-discrepancy sequence is designed so that the proportion of sequence points in any subinterval of [0,1]^d closely matches the interval's volume. Sobol sequences achieve this through bit-reversal constructions that distribute points more evenly than random sampling, ensuring that each new point fills a region of the space not yet covered by previous points. The practical effect is that for smooth integrands — a common situation in Black-Scholes derivatives pricing — the integration error converges at roughly O((log N)^d / N) rather than O(N^{-1/2}), providing a substantial accuracy advantage.
In options pricing, QMC is particularly valuable for path-dependent instruments such as Asian options, barrier options, and mortgage prepayment models, where the payoff depends on the entire simulated path of the underlying asset. Each time step in the simulation corresponds to one dimension in the QMC integration problem, so a daily-frequency simulation over one year requires 252 dimensions. Sobol sequences handle this gracefully up to a few hundred dimensions when combined with effective dimension reduction techniques such as the Brownian Bridge and principal component analysis (PCA) construction of correlated Wiener processes.
In risk management, QMC improves the accuracy of VaR and CVaR calculations, particularly for portfolios with complex, path-dependent exposures. Banks and asset managers use QMC-based scenario engines to generate realistic stress scenarios for regulatory capital models, where accuracy in the tails of the distribution is paramount. The key practical challenge is extending QMC to very high-dimensional problems — CDO pricing with hundreds of credits, for example — where the low-discrepancy property degrades and careful construction of the simulation is required to maintain the convergence advantage.
Formula
QMC error ≈ O((log N)^d / N) vs. MC error ≈ O(N^{-1/2})
Example
A derivatives desk prices a Himalaya option — a path-dependent multi-asset option that pays off based on the best-performing asset in each period across a basket of 10 stocks over 12 monthly periods, effectively a 120-dimensional integration problem. Using standard Monte Carlo with 100,000 paths, the pricing error (estimated by standard deviation of the mean) is ±$0.85 per unit of notional. Switching to a Sobol sequence QMC with the same 100,000 paths reduces the pricing error to ±$0.12 — a 7-fold improvement. Achieving equivalent accuracy with standard Monte Carlo would require approximately 4.9 million paths (since error ∝ N^{-1/2}), a 49x increase in computation. The QMC approach thus enables real-time pricing on the trading desk without sacrificing accuracy.
Related terms
Autoregressive Model Convergence Cross Sectional Momentum Monte Carlo Simulation Option Overfitting Principal Component Analysis Random Forest Reversal Standard Deviation Time Series Momentum