Binomial bias in pricing and early exercising American put options

The binomial algorithm (Cox and Rubinstein, 1985) is an accepted theoretically justifiable standard for measuring the accuracy of American put option pricing algorithms. An important question is whether it also generates accurate estimates of the early exercise boundary (Lamberton, 1993). I show that, in addition to the non-linearity and distribution errors recognised in the literature, the algorithm systematically misprices the early exercise boundary. While convergence to the true boundary ultimately occurs, the convergence rate is slow. An analytic integral equation (Carr et al., 1992) is solved sequentially for the true boundary and is a better benchmark for American put option pricing algorithms.