Measuring dependence with local Gaussian correlation

Wednesday 18 February 2015, 1.00PM to 2:00pm

Speaker(s): Dag Bjarne Tjøstheim, University of Bergen

Abstract: The Pearson correlation is the most used dependence measure in statistics. It has many weaknesses and really only works very well for Gaussian variables. In this talk I introduce a local Gaussian correlation by approximating a bivariate density locally by a bivariate Gaussian density. The correlation coefficient of the approximating Gaussian is taken as the local correlation. I will give some theoretical properties of this dependence measure and present a number of applications to independence testing, copula description and recognition, high dimensional density estimation, financial contagion and measuring asymmetry patterns of financial returns.

References: Tjøstheim and Hufthammer, J. Econometrics (2013), Berentsen and Tjøstheim, Statistics and Computing (2014), Berentsen, Støve, Tjøstheim and Nordbøe, Insurance: Mathematics and Economics (2014), Berentsen, Kleppe and Tjøstheim, J. Statistical Software (2014), Støve and Tjøstheim, J. Empirical Finance (2014), Støve and Tjøstheim, in Nonlinear Time Series Econometrics, Oxford (2014).

Location: ARRC Auditorium (A/RC014)

Admission: All welcome.