Advanced search
Start date
Betweenand

Time-clustering and forecasting performance in semi-parametric INAR(1) models

Grant number: 17/22914-5
Support type:Scholarships abroad - Research Internship - Doctorate
Effective date (Start): May 01, 2018
Effective date (End): April 30, 2019
Field of knowledge:Physical Sciences and Mathematics - Probability and Statistics
Principal Investigator:Hedibert Freitas Lopes
Grantee:Helton Graziadei de Carvalho
Supervisor abroad: Igor Pruenster
Home Institution: Instituto de Matemática e Estatística (IME). Universidade de São Paulo (USP). São Paulo , SP, Brazil
Local de pesquisa : Università Commerciale Luigi Bocconi, Italy  
Associated to the scholarship:17/10096-6 - Bayesian semiparametric analysis of autoregressive models, BP.DR

Abstract

In this project, we propose extensions of the INAR(1) model by using nonparametric priors on the time-varying innovations distributions. We first investigate clustering aspects and the predictive performance of the Dirichlet Process Mixture (DPM) model. In order to overcome limitations of the DPMs, we introduce more general non-parametric priors based on Pitman-Yor Processes and Gibbs-Type Measures. These generalizations facilitate the estimation of the posterior distribution of clusters and may improve predictive performance in a wide range of applications. Moreover, we explore the predictive performance of the proposed models in overdispersed datasets. (AU)