No label defined (Q4355646)

This book is the first monograph reporting on a fairly recent branch of the theory of optimal stopping, the theory of prophet inequalities. For a sequence \(X=(X_1, \dots, X_n)\) of real-valued random variables, the value \(V(X)\) is defined by \(V(X)= \sup\{EX_\tau: \tau \in T^n\}\), where \(T^n\) is the set of stopping times \(\tau\leq n\). The filtration defining \(T^n\) usually is the one generated by \(X\). Let \(M(X)= E(\sup_iX_i)\). \(V(X)\) is the optimal expected gain of a player observing the process, and \(M(X)\) is interpreted as the optimal expected gain of a ``prophet''. His expected reward is maximized if he stops at the maximum, and due to his perfect foresight, he knows when the maximum occurs. The basic question in this area is a comparison of \(V(X)\) and \(M(X)\). A well-known result asserts that \(M(X) \leq 2V(X)\) holds for independent nonnegative \(X_1, \dots, X_n\), see the reviewer and \textit{L. Sucheston} [Bull. Am. Math. Soc. 83, 745-747 (1977; Zbl 0336.60032)]. By now, more than 60 publications on related estimates for other processes have appeared, and this book deals with the diverse methods introduced by Hill, Kennedy, Samuel-Cahn, Rinott, Wittmann, Boshuizen, Gilat, and, last not least, N. Schmitz and his doctoral students Harten and Meyerthole. After a short introduction, the balayage technique of Hill and Kertz is explained, and it is applied both to general processes and to martingales. Then prophet inequalities, and ``prophet regions'' are discussed in the independent case. Extremal distributions are identified, and there are modifications like cost of observation and discounting. Chapter III treats the iid-case. In some cases, one can show that special types of stop rules (e.g. threshold stop rules) suffice to obtain strong estimates. Some of the material has not appeared yet except in the thesis of the first named authors. It is certainly useful to have a coherent rigorous introduction to this field now. However, the authors do not attempt complete coverage of the subject. It would have been nice if some of the more subtle estimates on dependent processes, Cesáro-averages and subadditive processes had at least been stated without proof. For these topics, the reader may wish to consult the excellent survey of \textit{T. P. Hill} and \textit{R. P. Kertz} [in: Strategies for sequential search and selection in real time. Contemp. Math. 125, 191-207 (1992; Zbl 0794.60040)] to which the authors refer in their introduction. Apparently, some more recent work like the surprising example of Wittmann showing that the prophet constant for Cesáro-averages of independent nonnegative random variables is strictly larger than 2, was written too late to be mentioned in the book. The presentation is self-contained and well readable. A basic course on measure theoretic probability will suffice as prerequisite. For a book presenting recent mathematical research, a publication in English might have helped to attract a wider readership.