Tail probabilities of St. Petersburg sums, trimmed sums, and their limit (Q2412518): Difference between revisions

In a single St. Petersburg game, the gain \(X\) is distributed according to \(P(X=2^k)=2^{-k}\) for \(k\in\mathbb N\). The total winning in \(n\) games is \(S_n=\sum_{k=1}^nX_k\), where the \((X_k)_{k\in\mathbb N}\) are i.i.d. as \(X\). The paper additionally considers the \(r\)-trimmed sums \(S_{n,r}\) with the \(r\) largest winnings subtracted and gives a substantial contribution to and new insights into the limiting and tail behavior of St. Petersburg sums. The authors determine the exact tail behavior \(P(S_{n,r}>x)\) as \(x\to\infty\) for fixed \(r\geq0\), which in the untrimmed case \(r=0\) shows that the distribution of \(X\) (known to be \(O\)-subexponential) is almost subexponential in a given precise sense. A corresponding result for generalized St. Petersburg games \(P(X=q^{-k/\alpha})=q^{k-1}(1-q)\) for some \(q\in(0,1)\) and \(\alpha>0\) is also presented without proof. The authors further present all subsequential limit distributions of \(n^{-1}S_{n,1}-\log_2n\) by a merging theorem in the spirit of \textit{S. Csörgő} [Acta Sci. Math. 68, No. 3--4, 815--847 (2002; Zbl 1046.60017)] which covers distributional convergence along certain subsequences and a series representation of the limiting distribution functions presented by \textit{A. Gut} and \textit{A. Martin-Löf} [J. Theor. Probab. 29, No. 1, 277--291 (2016; Zbl 1342.60030)]. A further series representation of the \(r\)-trimmed limiting random variables by means of i.i.d. exponentially distributed random variables is shown and enables the authors to give the precise tail behavior of the limiting distributions for \(r\)-trimmed St. Petersburg sums. To complete the picture, uniform tail bounds for the appropriately normalized and centered trimmed sums are given.