Semiparametric Multivariate and Multiple Change-Point Modeling

Abstract

We develop a general Bayesian semiparametric change-point model in which separate groups of structural parameters (for example, location and dispersion parameters) can each follow a separate multiple change-point process, driven by time-dependent transition matrices among the latent regimes. The distribution of the observations within regimes is unknown and given by a Dirichlet process mixture prior. The properties of the proposed model are studied theoretically through the analysis of inter-arrival times and of the number of change-points in a given time interval. The prior-posterior analysis by Markov chain Monte Carlo techniques is developed on a forward-backward algorithm for sampling the various regime indicators. Analysis with simulated data under various scenarios and an application to short-term interest rates are used to show the generality and usefulness of the proposed model.

A New Approach to Dating the Reference Cycle

“…The fact that they are ordered implies that τ P i < τ P i+1 , and if τ P i = τ P i+1 , then τ T i ≤ τ T i+1 . 6 Among other, recent contributions on change-point modelling are Peluso et al (2018), Chan and Koop (2014), Fearnhead (2006), Giordani and Kohn (2008), and Ko et al (2015).…”

“…The method provides exact matches for one peak and three troughs, differing by less than one quarter either way in the date of the remaining turning points. 23 On average, the method locates the peaks about one and a half months before and the troughs about one sixth of a month after the NBER dating. Remarkably, there are no instances in which an NBER turning point date is not matched by a similar date produced by the MSMM model, whose 95% credible intervals include the NBER-referenced dates in all the episodes.…”

“…The credible intervals, which range from 0.08 to 1.08 years for peaks and from 0.07 to 1.98 years for troughs, provide a precise estimate of the reference dates 23. The only exception is the 2001 trough, which is located only four months after the NBER dating.…”

A New Approach to Dating the Reference Cycle

“…The fact that they are ordered implies that τ P i < τ P i+1 , and if τ P i = τ P i+1 , then τ T i ≤ τ T i+1 . 6 Among other, recent contributions on change-point modelling are Peluso et al (2018), Chan and Koop (2014), Fearnhead (2006), Giordani and Kohn (2008), and Ko et al (2015).…”

Multipartition model for multiple change point identification

“…This is a long-standing limitation of the PPM. Recently, Peluso et al (2019) proposed a semiparametric model for detecting multiple change points occurring in several structural parameters. Extending the Chib (1998) model, they assume a specific latent Markovian discrete state variable for each structural parameter, allowing to identify which parameter experienced the change.…”