CMStatistics 2018: Start Registration
View Submission - CMStatistics
Title: Exact inference for Cox-Ingersoll-Ross driven hidden Markov models Authors:  Matteo Ruggiero - University of Torino (Italy) [presenting]
Guillaume Kon Kam King - University of Torino (Italy)
Omiros Papaspiliopoulos - UPF (Spain)
Abstract: Inference is considered on the hidden signal and on the respective parameters for a hidden Markov model driven by a Cox-Ingersoll-Ross diffusion, with discretely collected observations from a marginally conjugate emission density. We show that a set of sufficient conditions related to dual processes that allow us to derive in closed forms all quantities of interest are in this case met, and provide recursive forms for the filtering and the smoothing distributions, as well as for the marginal likelihood of the observations. All these expressions are computable in that their evaluation entails only a finite computational effort. We investigate the implications of our results, which easily accommodate certain pruning techniques for speeding up inference, which are in turn tested against competitive alternative methods.