Generalized Rank Dirichlet distributions

From MaRDI portal



Abstract: We introduce a new parametric family of distributions on the ordered simplex yinmathbbRd:y1geqdotsgeqydgeq0,sumk=1dyk=1, which we call Generalized Rank Dirichlet (GRD) distributions. Their density is proportionate to prodk=1dykak1 for a parameter ainmathbbRd satisfying ak+ak+1+dots+ad>0 for k=2,dots,d. Random variables of this type have been used to model ranked order statistics for positive weights that sum to one. We establish a change of measure formula that relates GRD distributions with different parameters to each other. Leveraging connections to independent exponential random variables we are able to obtain explicit expressions for moments of order MinmathbbN for the weights Yk's and moments of all orders for the log gaps Zk=logYk1logYk when a1+dots+ad=M for any dimension d. Additionally, we propose an algorithm to exactly simulate random variates in this case. In the general case when a1+dots+adinmathbbR we obtain series representations for the same quantities and provide an approximate simulation algorithm.











This page was built for publication: Generalized Rank Dirichlet distributions

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6192366)