|
Activity Number:
|
492
- Recent Advances in Modeling Complex Dependent Data
|
|
Type:
|
Invited
|
|
Date/Time:
|
Wednesday, August 1, 2018 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Business and Economic Statistics Section
|
|
Abstract #326721
|
Presentation
|
|
Title:
|
Spatio-Temporal Modeling of Heavy-Tailed Data via Non-Gaussian Latent Processes
|
|
Author(s):
|
Gabriel Huerta* and Kellin Rumsey
|
|
Companies:
|
University of New Mexico and University of New Mexico
|
|
Keywords:
|
Non-Gaussian process;
latent variables;
Generalized Pareto distribution;
MCMC;
Gamma/Poisson process;
Environmental data
|
|
Abstract:
|
In this work we introduce a new spatio-temporal model with Generalized Pareto marginal distributions (GPD) where time and space dependence is incorporated thru the use of latent variables in a hierarchical fashion. Furthermore the model relies on a conjugate, non-Gaussian structure that considers a Gamma/Poisson process. Our GPD process is motivated by the "peaks over threshold" approach that is well known from Extreme Value theory (EVT) and where the GPD is seen as a scale mixture of an Exponential and Gamma distributions. We study some of the properties of the proposed process and in particular, if the process is asymptotically independent or dependent. We follow a Bayesian approach via customized MCMC methods to estimate the model and produce inference on posterior quantiles or predictions. Our approach is illustrated with data simulations and a data set of pollution concentrations over a large metropolitan area.
|
Authors who are presenting talks have a * after their name.
