# Publications and Preprints

Ordering convolutions of gamma random variables
by
Baha-Eldin Khaledi and Subhash C. Kochar
Let $a_{(i)}$ and $b_{(i)}$ be the $i$th smallest components of ${\bf a}=(a_1,\ldots,a_n)$ and ${\bf b}=(b_1,\ldots,b_n)$ respectively, where ${\bf a}, {\bf b}$ are in $R$. The vector ${\bf a}$ is said to be $p$-larger than the vector ${\bf b}$ (denoted by ${\bf a} \stackrel {p} \succeq {\bf b}$ ) if $\prod_{i=1}^{k}a_{(i)} \le \prod_{i=1}^{k} b_{(i)},\mbox{ for } k=1,\ldots,n$. Let $X_{\lambda_1},\ldots,X_{\lambda_n}$ be independent random variables such that $X_{\lambda_i}$ has gamma distribution with shape parameter $a \ge 1$ and scale parameter $\lambda_i$, $i= 1, \ldots, n$. It is shown that if $\mbox {\boldmath$ \lambda $} \stackrel {p} \succeq \mbox {\boldmath$ \lambda ^*$}$, then $\sum_{i=1}^n X_{\lambda_i}$ is greater than $\sum_{i=1}^n X_{\lambda_i^*}$ according to dispersive as well as hazard rate orderings. This strengthens the results of Kochar and Ma [Statistics \& Probability Letters 43 (1999), 321-324] and Korwar [J. Multivariate Analysis 80 (2002), 344-357] from usual majorization to $p$-larger ordering and leads to better bounds on various quantities of interest.

isid/ms/2004/02 [fulltext]