Elementary multivariate rearrangements and stochastic dominance on a Fréchet class (Q435907): Difference between revisions
From MaRDI portal
Added link to MaRDI item. |
Set profile property. |
||
Property / MaRDI profile type | |||
Property / MaRDI profile type: MaRDI publication profile / rank | |||
Normal rank |
Revision as of 00:15, 5 March 2024
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Elementary multivariate rearrangements and stochastic dominance on a Fréchet class |
scientific article |
Statements
Elementary multivariate rearrangements and stochastic dominance on a Fréchet class (English)
0 references
13 July 2012
0 references
The paper under review is to extend results on bivariate stochastic dominance and elementary rearrangements to ones on multivariate cases. A Fréchet class is the collection of multivariate joint distribution functions with the same marginals. The bivariate joint distribution function obtained by a bivariate rearrangement is stochastically dominated by initial one to the first order and vice verse. But this equivalence relationship no longer holds in a multivariate setting. The author first recalls the bivariate case in section 3, where section 2 introduces notations, and defines the bivariate function \(g\) is obtained from \(f\) by a 2-rearrangement (denoted by \(g=R^2(\underline{x}, \overline{x}, \delta) (f)\)). Hamada (1974) shows that marginals of a joint distribution function are more interdependent after such a rearrangement, and Epstein and Tanny (1980) show that on a bivariate Fréchet class, a finite sequence of positive (\(\delta > 0\)) 2-rearrangement is equivalent to stochastic dominance of the first order in terms of joint distribution functions as well as the survival functions. For multivariate functions, \(g = R^k(\underline{x}, \overline{x}, \delta) (f)\) (\(k> 2\)) is obtained from \(f\) by a \(k\)-rearrangement if there are two vectors \(\underline{x}\) and \(\overline{x}\) with \(\underline{x}<_k \overline{x}\) (it means componentwise) and \(g(x) = f(x) +\delta\) on \(2^{k-1}\) vertices in \(B^k(\underline{x}, \overline{x})= \prod_{j=1}^m [\underline{x}_j, \overline{x}_j]\) with an even number of entries \(x_j\) for which \(x_j = \underline{x}_j \neq \overline{x}_j\), \(g(x) = f(x) -\delta\) on \(2^{k-1}\) vertices in \(B^k(\underline{x}, \overline{x})= \prod_{j=1}^m [\underline{x}_j, \overline{x}_j]\) with an odd number of entries \(x_j\) for which \(x_j = \underline{x}_j \neq \overline{x}_j\), and \(g(x)=f(x)\) otherwise. With this definition (Definition 2 in the paper), the author proves that a finite sequence of alternating \(k\)-rearrangement is equivalent to the stochastic dominance of the first order in terms of joint distribution functions in Proposition 2, as well as the survival functions in Proposition 3. The author presents an example in Table 3 to illustrate a finite sequence of alternating rearrangements for the example in Muller and Scarsini (2000) in section 4. Section 5 concludes the results of this paper. It is interesting to apply the results of the paper into copulas and other type functions to decompose the information in the joint distribution function.
0 references
concordance order
0 references
Frechet class
0 references
multivariate rearrangement
0 references
multivariate stochastic dominance
0 references
Orthant dependence order
0 references
supermodular order
0 references