QuarcHome
Model catalogueBack to the model index

Longstaff–Schwartz least-squares Monte Carlo (LSM)

Estimates continuation values from simulated paths to price early-exercise options.

How it works

LSM approximates the conditional value of waiting at each possible exercise date by regression on variables available at that date. Simulated future discounted payoffs serve as regression targets. The exercise policy compares fitted continuation against immediate payoff and works backward over dates.

Worked example

Suppose an exercised put pays 10 now. A fitted continuation value of 9.4 calls for exercise on that path. If another path has the same spot but different relevant rate or volatility state, the continuation estimate may differ. Use additional state variables if they are material to future exercise.

Validation and limitations

Evaluate the learned exercise policy on paths outside the training sample, test the regression basis and compare with a small tree where possible. More regressors can fit simulation noise. Use primal–dual bounds to measure the remaining price gap.

Further reading