Bayesian semiparametric quantile regression modeling for estimating earthquake fatality risk

Abstract

This paper develops a Bayesian semiparametric quantile regression model for count data. The count responses are converted to continuous responses through the “jittered” method and a transform function. A Bayesian semiparametric quantile regression modeling approach is then developed. The error distribution in the quantile regression model is assumed to be a mixture of asymmetric Laplace distributions constructed with Dirichlet process. Historical death tolls of China caused by earthquakes from 1969 to 2006 are used for fitting, and a parametric model is employed for model comparison. The results of model comparison show that the proposed semiparametric quantile regression model outperforms the parametric model. The empirical analysis illustrates that the impact of earthquake magnitude on death tolls is significant. Moreover, the impact of the magnitude is more pronounced on higher percentiles of death tolls.