A0470
Title: Approximate Bayesian computation for factor copula models
Authors: Clara Grazian - University of Sydney (Australia) [presenting]
Feng Chen - UNSW Syd (Australia)
Hanwen Xuan - The University of New South Wales (Australia)
Abstract: Analyzing and modelling high-dimensional data has attracted great interest in statistics, particularly in applications relevant to time-series analysis and econometrics. In recent years, people have attempted to combine the ideas from the literature on factor analysis and copulas theory together to build up the so-called factor copula models. The use of factor copula models has been growing due to their ability to explain the dependence structure of high dimensional variables in terms of a few latent factors. This feature saves a large amount of computational burden and provides a decent alternative for analyzing high-dimensional datasets. The focus is on the factor copula model proposed in another study, where people could incorporate the class of dynamic factor models proposed in the literature of time series analysis with arbitrary marginal distributions. Their proposed factor copula models are extended into a Bayesian framework by using approximate Bayesian computation (ABC) methods to replace the simulation-based estimation procedures. It enables to not only overcome the issues of lacking closed-form solution in the factor copulas but also capture the model parameter uncertainties and enhance the predictions. The performance of our Bayesian estimation method is examined in both a simulation study and a real time series dataset.
