Parametric Inference for Discretely Observed Subordinate Diffusions

Abstract

Subordinate diffusions are constructed by time changing diffusion processes with an independent L\'{e}vy subordinator. This is a rich family of Markovian jump processes which exhibit a variety of jump behavior and have found many applications. This paper studies parametric inference of discretely observed ergodic subordinate diffusions. We solve the identifiability problem for these processes using spectral theory and propose a two-step estimation procedure based on estimating functions. In the first step, we use an estimating function that only involves diffusion parameters. In the second step, a martingale estimating function based on eigenvalues and eigenfunctions of the subordinate diffusion is used to estimate the parameters of the L\'{e}vy subordinator and the problem of how to choose the weighting matrix is solved. When the eigenpairs do not have analytical expressions, we apply the constant perturbation method with high order corrections to calculate them numerically and the martingale estimating function can be computed efficiently. Consistency and asymptotic normality of our estimator are established considering the effect of numerical approximation. Through numerical examples, we show that our method is both computationally and statistically efficient. A subordinate diffusion model for VIX (CBOE volatility index) is developed which provides good fit to the data.