B1314
Title: Bayesian variable selection in double generalized linear Tweedie spatial process models
Authors: Aritra Halder - Drexel University (United States) [presenting]
Shariq Mohammed - Boston University (United States)
Dipak Dey - UCONN (United States)
Abstract: Double-generalized linear models provide a flexible framework for modeling data by allowing the mean and the dispersion to vary across observations. Common members of the exponential dispersion family, including the Gaussian, Poisson, compound Poisson-gamma (CP-g), Gamma and inverse-Gaussian, are known to admit such models. The lack of their use can be attributed to ambiguities in model specification under many covariates and complications that arise when data display complex spatial dependence. The hierarchical specification for the CP-g model with a spatial random effect is considered. The spatial effect is targeted at performing uncertainty quantification by modeling dependence within the data arising from location-based indexing of the response while focusing on a Gaussian process specification for the spatial effect. Simultaneously, the problem of model specification is tackled using Bayesian variable selection, effected through a continuous spike and slab prior to the model parameters, specifically the fixed effects. The novelty of the contribution lies in the Bayesian frameworks developed for such models. Various synthetic experiments are performed to showcase the accuracy of the frameworks, which are then applied to analyze automobile insurance premiums in Connecticut for 2008.
