COMPSTAT 2022: Start Registration
View Submission - COMPSTAT2022
A0647
Title: Bayesian semiparametric copula estimation and model selection: A comparison study Authors:  Jichan Park - Korea University (Korea, South) [presenting]
Taeryon Choi - Korea University (Korea, South)
Abstract: There have been developed a lot of different models using Copulas to model a multivariate probability distribution and develop an efficient inference. We propose an extended semiparametric bayesian bivariate copula model based on existing bayesian copula models. First of all, the dependence structure of bivariate random variables is modeled with various bivariate copulas, and mixtures of B-spline densities are used for marginal distributions. The proposed copulas proposed Student $t$ copula, Clayton copula and Gumbel copula. They have the characteristic of maintaining the degree of dependence present in tails between random variables, so unlike Gaussian copula in the existed model, models with them can be appropriately used to explain data with different dependence structures. Also, we perform a bayesian model selection through a Reversible Jump Markov Chain Monte Carlo algorithm. In simulation studies of various settings, the performance of each model and the model selection were confirmed, and finally, a comparative analysis with real data sets was conducted using a stock price index and a crude oil data set.