Semiparametric Estimation for Multivariate Skew-Elliptical Distributions

• Activity Number: 204
• Type: Contributed
• Date/Time: Monday, August 1, 2016 : 11:35 AM to 12:20 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Jing Huang*
• Companies: European School of Management and Technology
• Keywords: Skewed-elliptical ; Semi-Parametric

Abstract

We investigate a class of skew-elliptical distributions which generalize the skewed distributions proposed by Azzalini and Dalla Valle (1996). These distributions are generated by elliptical contours plus an unknown distribution function which assigns weights or densities to these contours. Our model can decompose such a skewed distribution into those two parts above and allow us to estimate them separately, with both parametric and Non-parametric method. Our model is analyzed explicitly with simulated data estimation in one and two dimensional case. In a parametric framework and under the assumption that the density assigning function has a closed form, our approach leads to a faster estimation process than traditional log likelihood estimation (MLE) through direct optimization or through expectation maximization (EM). At end we also show the results of a fitting a real data with our model which better captures the empirical tail distribution.