The Multiprocess Dynamic Linear Model with Biased Perturbations: A Real Time Model for Growth Hormone Level

Abstract

This paper introduces biased perturbation distributions into the multiprocess dynamic linear model in order to represent growth hormone levels in animals, which are characteristically noisy pulsatile time series. The use of biased perturbation distributions allows a pulse to be detected on the first observation immediately after it occurs, thus allowing the process to be modelled in real time. This approach is found to be quite effective at detecting the occurrence of a pulse in growth hormone level, and revising the parameter estimates and predictions accordingly.