Efficient Rare-event Simulation for Heavy-tailed Compound Sums (with C. Li)

Summary:

Many algorithms (Asmussen-Kroese, Dupuis-Leder-Wang, Juneja-Shahabuddin, etc.) have been proposed to efficiently estimate via simulation the tail of the steady-state waiting time of a heavy-tailed M/G/1 queue. This has served as a model problem in order to develop intuition and ideas for more general heavy-tailed models. The existing literature develops algorithms that satisfy either weak or strong optimality under special tail assumptions (such as regularly varying, Weibull-that-are-not-too-light, etc.). This paper develops the first strongly efficient estimator under basically minimal assumptions (it just requires subexponential input).

Bibtex:

@Article{BlaLiCS10,

author = { J. Blanchet and C. Li},

title = {Efficient rare-event simulation for heavy-tailed compound sums},

journal = {ACM Transactions on Modeling and Computer Simulation - TOMACS},

year = {2011},

volume = {21},

pages = {1-10}

}