Showing 1–1 of 1 results for author: Woźny, J

Gaussian dependence structure pairwise goodness-of-fit testing based on conditional covariance and the 20/60/20 rule

Authors: Jakub Woźny, Piotr Jaworski, Damian Jelito, Marcin Pitera, Agnieszka Wyłomańska

Abstract: We present a novel data-oriented statistical framework that assesses the presumed Gaussian dependence structure in a pairwise setting. This refers to both multivariate normality and normal copula goodness-of-fit testing. The proposed test clusters the data according to the 20/60/20 rule and confronts conditional covariance (or correlation) estimates on the obtained subsets. The corresponding test… ▽ More We present a novel data-oriented statistical framework that assesses the presumed Gaussian dependence structure in a pairwise setting. This refers to both multivariate normality and normal copula goodness-of-fit testing. The proposed test clusters the data according to the 20/60/20 rule and confronts conditional covariance (or correlation) estimates on the obtained subsets. The corresponding test statistic has a natural practical interpretation, desirable statistical properties, and asymptotic pivotal distribution under the multivariate normality assumption. We illustrate the usefulness of the introduced framework using extensive power simulation studies and show that our approach outperforms popular benchmark alternatives. Also, we apply the proposed methodology to commodities market data. △ Less

Submitted 22 April, 2024; v1 submitted 19 April, 2024; originally announced April 2024.