Tail-corrected semiparametric inference for regime-switching jump diffusions

Abstract

Regime-switching jump diffusions describe continuous-time dynamical systems that exhibit both abrupt jumps and changes in regime, with applications in economics, ecology, and physics. Statistical inference for such processes is complicated by the interaction between jumps and regime switching. We study an ergodic jump diffusion with exogenous finite-state regime switching. From discrete observations of both the state and regime processes, the estimation targets are the drift and diffusion parameters and the unknown regime-wise L\'evy densities, in settings with either finite-variation jumps or locally stable infinite-variation jumps. We first estimate the tail means and coefficient parameters. For finite-variation jumps, coefficient estimation is based on a Gaussian quasi-likelihood; for locally stable infinite-variation jumps with common \(1<\beta<8/5\), diffusion estimation uses a two-step debiasing of truncated realized variation. We then use drift-corrected detected residuals to estimate each L\'evy density away from zero. We establish consistency and asymptotic normality for the coefficient estimators, together with an \(L^2(B)\)-convergence rate for the density estimators. Tail-mean estimation contributes an explicit L\'evy-measure term to the drift covariance, whereas the diffusion block retains the Gaussian quasi-score covariance. Simulations illustrate the finite-sample performance of the estimators.