Monte Carlo maximum likelihood estimation for discretely observed diffusion processes (Q1002156): Difference between revisions

The Monte Carlo method introduced in this paper gives unbiased and a.s. continuous estimates of the likelihood function of the discrete observations of a diffusion process. The likelihood contribution of each time step is estimated through independent copies of a random function. This function is based on a recently developed retrospective rejection sampling algorithm called the Exact Algorithm. There is no discretization error. Under regularity conditions, the Monte Carlo maximum likelihood estimator converges a.s. to the true maximum likelihood estimator of the unknown parameter. When the datasize \(n\) tends to infinity, the optimal number of Monte Carlo iterations should be tuned as \({\mathcal O}(n^{1/2})\) and the resulting Monte Carlo maximal likelihood estimator converges a.s. to the true parameter value. A numerical illustration of the method is given for the estimation of the three parameters of a logistic growth stochastic differential equation.