Graphical lassos for meta-elliptical distributions
نویسنده
چکیده
Gaussian graphical lasso is a tool for estimating sparse graphs using a Gaussian log-likelihood with an 1 penalty on the inverse covariance matrix. This paper proposes a generalization to meta-elliptical distributions. Conditional uncorrelatedness is characterized in meta-elliptical families. The proposed metaelliptical and re-weighted Kendall graphical lassos are computed from pseudo-observations which are functions of ranks of observations. They are invariant to strictly increasing transformations of the variables and do not assume the existence of moments. Simulations of receiver operating characteristic curves show noticeable improvements (in comparison with graphical lassos designed for meta-Gaussian distributions) for distributions which are not meta-Gaussian. These improvements are realized without ill effects when the distribution is meta-Gaussian. Deterministic and random contaminations of data are used to verify the robustness of the re-weighted Kendall graphical lasso. The Canadian Journal of Statistics 42: 185–203; 2014 © 2014 Statistical Society of Canada Résumé: Le lasso graphique gaussien est un estimateur de graphe épars basé sur la log vraisemblance comportant une pénalité 1 sur l’inverse de la matrice de covariance. L’auteur propose une généralisation aux distributions méta-elliptiques. La non-corrélation conditionnelle est caractérisée dans les familles métaelliptiques. Deux lassos graphiques sont proposés : le lasso méta-elliptique et le lasso de Kendall repondéré, tous deux calculés à partir de pseudo-observations basées sur les rangs. Ils sont invariants aux transformations strictement monotones croissantes et ne présupposent l’existence d’aucun moment. Dans le cadre de simulations, ils offrent une performance (en termes de courbe ROC) comparable aux lassos spécifiques aux distributions méta-gaussiennes lorsque les données suivent cette distribution. Une amélioration notable est cependant observée lorsque les données ne suivent pas la distribution méta-gaussienne. La robustesse du lasso de Kendall repondéré est aussi illustrée au moyen de données contaminées de manière aléatoire ou déterministe. La revue canadienne de statistique 42: 185–203; 2014 © 2014 Société statistique du Canada
منابع مشابه
Transelliptical Component Analysis
We propose a high dimensional semiparametric scale-invariant principle component analysis, named TCA, by utilize the natural connection between the elliptical distribution family and the principal component analysis. Elliptical distribution family includes many well-known multivariate distributions like multivariate Gaussian, t and logistic and it is extended to the meta-elliptical by Fang et.a...
متن کاملGeneralized Birnbaum-Saunders Distribution
The two-parameter Birnbaum–Saunders (BS) distribution was originally proposed as a failure time distribution for fatigue failure caused under cyclic loading. BS model is a positively skewed statistical distribution which has received great attention in recent decades. Several extensions of this distribution with various degrees of skewness, kurtosis and modality are considered. In particular, a...
متن کاملThe Tail Mean-Variance Model and Extended Efficient Frontier
In portfolio theory, it is well-known that the distributions of stock returns often have non-Gaussian characteristics. Therefore, we need non-symmetric distributions for modeling and accurate analysis of actuarial data. For this purpose and optimal portfolio selection, we use the Tail Mean-Variance (TMV) model, which focuses on the rare risks but high losses and usually happens in the tail of r...
متن کاملCombining Statistical Models
This paper develops a general framework to support the combination of information from independent but related experiments, by introducing a formal way of combining statistical models represented by families of distributions. A typical example is the combination of multivariate Gaussian families respecting conditional independence constraints, i.e. Gaussian graphical models. Combining informati...
متن کاملSmall witnesses, accepting lassos and winning strategies in omega-automata and games
Obtaining accepting lassos, witnesses and winning strategies in ω-automata and games with ωregular winning conditions is an integral part of many formal methods commonly found in practice today. Despite the fact that in most applications, the lassos, witnesses and strategies found should be as small as possible, little is known about the hardness of obtaining small such certificates. In this pa...
متن کامل