Likelihood estimation for stochastic differential equations with mixed effects

TL;DR

This paper introduces efficient likelihood estimation methods for stochastic differential equations with mixed effects, enabling analysis of complex models with random parameters, demonstrated through simulations and neuronal data application.

Contribution

It presents Gibbs samplers and stochastic EM-algorithms based on diffusion bridge simulation for mixed effects SDEs, with extensions to measurement errors and simplified forms for exponential family models.

Findings

Methods perform well in simulations with Ornstein-Uhlenbeck and t-diffusions.

Algorithms are computationally efficient at low sampling frequencies.

Application to neuronal data demonstrates practical utility.

Abstract

Stochastic differential equations provide a powerful tool for modelling dynamic phenomena affected by random noise. In case of repeated observations of time series for several experimental units, it is often the case that some of the parameters vary between the individual experimental units, which has motivated a considerable interest in stochastic differential equations with mixed effects, where a subset of the parameters are random. These models enable simultaneous representation of randomness in the dynamics and variability between experimental units. When the data are observations at discrete time points, the likelihood function is only rarely explicitly available, so for likelihood-based inference to be feasible, numerical methods are needed. We present Gibbs samplers and stochastic EM-algorithms based on augmented data obtained by the simple method for simulation of diffusion bridges in Bladt and S{\o}rensen (2014). This method is easy to implement and has no tuning parameters. The method is, moreover, computationally efficient at low sampling frequencies because the computing time increases linearly with the time between observations. The algorithms can be extended to models with measurement errors. The Gibbs sampler as well as the EM-algorithm are shown to simplify considerably for exponential families of diffusion processes, including many models used in practice. In a simulation study, the estimation methods are shown to work well for Ornstein-Uhlenbeck processes and t-diffusions with mixed effects. Finally, the Gibbs sampler is applied to neuronal data.