International Conference on Multivariate Approximation
September 24-27, 2011

Latex rendering failed!

Compactly supported correlations: a statistical view

Emilio Porcu
Estimation of covariance parameters requires a considerable cost from the computational viewpoint, in particular when dealing with massive datates. Such a problem is even bigger in the case of vector-valued random fields. We propose criteria for the construction of mappings from ${R}^d$ to $p \times p$ matrices with the requirement of being positive definite. We propose some correlation matrix-valued functions whose members have different compact support and different level of smoothness. We illustrate the features of such models through simulation as well as data analysis