Title: “Bayesian Copula Modelling in the Presence of Covariates”
Abstract:
Copula models separate the dependence structure in a multivariate distribution from its
univariate marginals, so overcoming many of the issues associated with commonly used
statistical modelling methods by allowing, for example, different complex asymmetric
dependencies and tail behaviours to be modelled. We discuss the modelling of bivariate
data using copulas, of which there are now a rich choice. The parameter or parameters of
the copula density are modelled as a function of a covariate using a natural cubic spline.
Working in the Bayesian framework, we perform inference on the natural cubic spline
and an associated smoothing parameter. We also discuss the choice of the copula density
itself. We illustrate our approach using data from child health and finance. We mention
the extension of our methodology to more than one covariate and to multivariate data.
Please note that this talk will last considerably less than one hour. There may be the
possibility also to discuss brie
y a fun application of survival analysis to historical data
that yields interesting insights about how life expectancies may have changed since 1400,
but this is not guaranteed.
The Royal Statistical Society SouthWest Local Group Annual General Meeting
will take place before the talk.