-
Vine copulas as a mean for the construction of high dimensional probability distribution associated to a Markov Network
Authors:
Edith Kovacs,
Tamas Szantai
Abstract:
Building higher-dimensional copulas is generally recognized as a difficult problem. Regular-vines using bivariate copulas provide a flexible class of high-dimensional dependency models. In large dimensions, the drawback of the model is the exponentially increasing complexity. Recognizing some of the conditional independences is a possibility for reducing the number of levels of the pair-copula dec…
▽ More
Building higher-dimensional copulas is generally recognized as a difficult problem. Regular-vines using bivariate copulas provide a flexible class of high-dimensional dependency models. In large dimensions, the drawback of the model is the exponentially increasing complexity. Recognizing some of the conditional independences is a possibility for reducing the number of levels of the pair-copula decomposition, and hence to simplify its construction Aas et al (2009). The idea of using conditional independences was already performed under elliptical copula assumptions Hanea, Kurowicka and Cooke (2006), Kurowicka and Cooke (2002) and in the case of DAGs in a recent work Bauer, Czado and Klein (2011). We provide a method which uses some of the conditional independences encoded by the Markov network underlying the variables. We give a theorem which under some graph conditions makes possible to derive pair-copula decomposition of the probability density function associated to a Markov network. As the underlying Markov network is usually unknown, we first have to discover it from the sample data. Using our results published in Szantai and Kovacs (2008) and Kovacs and Szantai (2010a) we will show how to derive a multidimensional copula model exploiting the information on conditional independences hidden in the sample data.
△ Less
Submitted 6 June, 2012; v1 submitted 9 May, 2011;
originally announced May 2011.
-
Discovering a junction tree behind a Markov network by a greedy algorithm
Authors:
Tamas Szantai,
Edith Kovacs
Abstract:
In an earlier paper we introduced a special kind of k-width junction tree, called k-th order t-cherry junction tree in order to approximate a joint probability distribution. The approximation is the best if the Kullback-Leibler divergence between the true joint probability distribution and the approximating one is minimal. Finding the best approximating k-width junction tree is NP-complete if k>2.…
▽ More
In an earlier paper we introduced a special kind of k-width junction tree, called k-th order t-cherry junction tree in order to approximate a joint probability distribution. The approximation is the best if the Kullback-Leibler divergence between the true joint probability distribution and the approximating one is minimal. Finding the best approximating k-width junction tree is NP-complete if k>2. In our earlier paper we also proved that the best approximating k-width junction tree can be embedded into a k-th order t-cherry junction tree. We introduce a greedy algorithm resulting very good approximations in reasonable computing time.
In this paper we prove that if the Markov network underlying fullfills some requirements then our greedy algorithm is able to find the true probability distribution or its best approximation in the family of the k-th order t-cherry tree probability distributions. Our algorithm uses just the k-th order marginal probability distributions as input.
We compare the results of the greedy algorithm proposed in this paper with the greedy algorithm proposed by Malvestuto in 1991.
△ Less
Submitted 18 April, 2011; v1 submitted 14 April, 2011;
originally announced April 2011.
-
Multivariate Copula Expressed by Lower Dimensional Copulas
Authors:
Edith Kovacs,
Tamas Szantai
Abstract:
Modeling of high order multivariate probability distribution is a difficult problem which occurs in many fields. Copula approach is a good choice for this purpose, but the curse of dimensionality still remains a problem. In this paper we give a theorem which expresses a multivariate copula by using only some lower dimensional ones based on the conditional independences between the variables. In ge…
▽ More
Modeling of high order multivariate probability distribution is a difficult problem which occurs in many fields. Copula approach is a good choice for this purpose, but the curse of dimensionality still remains a problem. In this paper we give a theorem which expresses a multivariate copula by using only some lower dimensional ones based on the conditional independences between the variables. In general the construction of a multivariate copula using this theorem is quite difficult, due the consistency properties which have to be fulfilled. For this purpose we introduce the sample derivated copula, and prove that the dependence between the random variables involved depends just on this copula and on the partition. By using the sample derivated copula the theorem can be successfully applied, in order to to construct a multivariate discrete copula by using some of its marginals.
△ Less
Submitted 15 September, 2010;
originally announced September 2010.
