Maximum likelihood estimation of a latent variable time-series model

Recently, Fridman and Harris proposed a method which allows one to approximate the likelihood of the basic stochastic volatility model. They also propose to estimate the parameters of such a model maximising the approximate likelihood by an algorithm which makes use of numerical derivatives. In this paper we propose an extension of their method which enables the computation of the first and second analytical derivatives of the approximate likelihood. As will be shown, these derivatives may be used to maximize the approximate likelihood through the Newton-Raphson algorithm, with a saving in the computational time. Moreover, these derivatives approximate the corresponding derivatives of the exact likelihood. In particular, the second derivative may be used to compute the standard error of the estimator and confidence intervals for the parameters. The paper presents also the results of a simulation study which allows one to compare our approach with other existing approaches.