Bivariate extension of Lomax and finite range distributions through characterization approach (Q1817487): Difference between revisions

For a stochastic description of the life of a decreasing failure rate (DFR) component or a system the Lomax distribution with survival function \(S(x)= (1+x/a)^{-q}\), \(a>0\), \(q>0\) has been widely used by practitioners. A multivariate generalization of the Lomax distribution has been proposed by \textit{D. V. Lindley} and \textit{N. D. Singpurwalla} [J. Appl. Probab. 23, 418-431 (1986; Zbl 0603.60084)]. It has a built-in structure of dependency. As a result, for no choice of the parameters can one obtain the situation where component lives are independently distributed. In fact, in this model, there is no scope of examining the hypothesis that the component lives are independent. This is a major limitation when used for stochastic modelling. \textit{B. C. Arnold} [J. Stat. Plann. Inference 24, No. 2, 249-258 (1990; Zbl 0731.62099)] has used multivariate geometric minima of univariate Pareto random variables to generate a standard multivariate Pareto distribution and proposed therefrom a flexible family of multivariate Pareto distributions. Arnold has also examined related properties and estimation issues of his class in two dimensional case. Our problem then arises in selecting a particular model from amongst the members of Arnold's flexible class so that the chosen model can retain interrelationships with a bivariate exponential and a bivariate FR distribution. To solve this problem we consider a bivariate generalization of a univariate characterizing property that links up Lomax distribution with exponential and FR distributions. We next try to identify the bivariate Lomax distribution for which this generalization is a unique property. The characterization result we present also provides answers to the problem of bivariate generalizations of the exponential and the FR distributions. The bivariate Lomax distribution thus obtained is a particular member of the Arnold class with suitable reparametrization. A bivariate exponential distribution due to \textit{E. J. Gumbel} [J. Am. Stat. Assoc. 55, 698-707 (1960; Zbl 0099.14501)] happens to be identical with what we have obtained here. We also examine the suitability of our proposed distributions in respect of other properties including constant memory.