Longstaff-Schwartz Method
Longstaff-Schwartz Least-Squares Monte Carlo · Also known as: LSM, Least-Squares MC, Optimal Stopping
The Longstaff-Schwartz method (2001) is a Monte Carlo algorithm for pricing American options and Bermudan swaptions by approximating the optimal exercise boundary via least-squares regression. It has become the industry standard for pricing path-dependent derivatives where analytical solutions do not exist.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use Longstaff-Schwartz for pricing American and Bermudan options where other methods (tree, PDE) are infeasible or slow. Essential for long-dated Bermudan swaptions, interest rate exotics, and multi-dimensional derivatives. Convergence is slow (1/sqrt(N)); use variance reduction for speed.
Strengths & limitations
- Flexible: applies to any path-dependent payoff and any number of dimensions
- Practical: simple to implement; regress-backward approach is intuitive
- Convergent: provably converges to true value as path count increases
- Widely adopted: industry standard for Bermudan swaptions and complex exotics
- Computational cost: requires many paths (100K+) and regression at each time step
- Low convergence rate: 1/sqrt(N) means need large N for high precision
- Bias: can under-estimate continuation value if basis functions misfit; use high-order basis
- High dimensionality curse: regression becomes unstable in 5+ dimensions without dimensionality reduction
Frequently asked
How do I choose the basis functions?
Start with orthogonal polynomials (Hermite, Laguerre) or simple powers (1, S, S^2, ...). Use 2-4 terms for low-dim problems, 5-7 for moderate. Test basis convergence: if adding more terms changes price significantly, basis is too small. Use Chebyshev or Laguerre if spot is lognormal (more efficient).
How many paths do I need?
Start with 10K paths for 1% precision. LSM convergence is ~1/sqrt(N), so 100K paths give ~0.3% precision. For 0.1% precision, need 1M paths. Use low-discrepancy sequences (Sobol) to reduce paths by factor of 2-3.
Why backward regression?
American options are naturally solved backward: at maturity, payoff is known. At each earlier date, you use future payoffs to estimate continuation value. Forward simulation would require solving the option value at each state in advance, which is the standard PDE or tree approach (slow in high dimensions).
Can I use LSM for all exotics?
LSM works for any payoff that can be simulated forward. Limitations: (1) slow for very high dimensions (5+) without dimensionality reduction, (2) slow for very long-dated products (many time steps), (3) subtle when multiple exercise dates or path-dependent features interact.
Sources
- Longstaff, F. A., & Schwartz, E. S. (2001). Valuing American options by simulation: A simple least-squares approach. Review of Financial Studies, 14(1), 113-147. DOI: 10.1093/rfs/14.1.113 ↗
- Clements, D. J., & Minca, A. (2008). A simulation approach to estimating near-optimal valuation functions for Bermudan options. Journal of Computational Finance, 12(2), 73-96. link ↗
How to cite this page
ScholarGate. (2026, June 3). Longstaff-Schwartz Least-Squares Monte Carlo. ScholarGate. https://scholargate.app/en/quantitative-finance/longstaff-schwartz-method
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Bates ModelQuantitative Finance↔ compare
- Local Volatility (Dupire)Quantitative Finance↔ compare
- Risk-Neutral ValuationQuantitative Finance↔ compare
- SABR ModelQuantitative Finance↔ compare