Variance-constrained canonical least-squares Monte Carlo: An accurate method for pricing American options

• Variance-constraint for canonical least-squares Monte Carlo (vCLM) is innovative.
• The variance and the American put's price are obtained at once by iterative search.
• 16,249 American-style S&P 100 index puts are studied empirically.
• vCLM is truly accurate with an average absolute pricing error of only 5.94%.
• vCLM outperforms CLM and two other benchmarks.

The pricing accuracy of the canonical least-squares Monte Carlo (CLM) method can be improved significantly by incorporating innovatively a variance constraint in the derivation of the canonical risk-neutral distribution. This new approach is called the variance-constrained CLM (vCLM) in the paper. Operationally, the forward variance is set to be the square of the volatility implied under vCLM by the option's market price from a previous trading day. For 16,249 American-style S&P 100 index puts, vCLM produced an average absolute pricing error of 5.94%, easily outperforming CLM, a competing nonparametric approach, and a GARCH-based benchmark.