Strong laws for the maximal gain over increasing runs (Q1591170)

This paper contains two results and a brief survey of numerous attempts to answer questions like the following one: how much can a player win in successive repetitions of the game with increasing gain? The authors consider a general setting which includes a number of special cases, not investigated earlier. Let \((X_i, Y_i)_{i=1,2,\dots}\) be a sequence of i.i.d. random vectors and \[ M_n(j)=\max_{0\leq k\leq n-j}(X_{k+1}+\dots+X_{k+j}) I\{Y_{k+1}\leq \dots \leq Y_{k+j}\}, \] where \(I\{\cdot\}\) denotes the indicator function of the event in brackets. Just in this case, \(M_n(j)\) can be interpreted as the maximal gain of the player over increasing runs in an accompanying sequence \(\{Y_i\}.\) The main concern is to derive the following two results for \(M_n\) \((L_n)\) in the case of continuous \(Y_1\) \((P(Y_1=y)=0\) for all \(y\)), where \[ L_n=\max\{j\in\{1, \dots, n\}: \exists k\in \{0, \dots, n-j\} \text{ such that }Y_{k+1} \leq \dots \leq Y_{k+j}\}. \] Theorem 1. Suppose that \(E|X_1|^r< \infty\) for some \(r>3.\) Then \[ \lim_{n\to \infty}\frac {M_n\;(L_n)}{\log n/\log \log n}=\max\{0, EX\}\quad \text{ a.s.} \] Theorem 2. Suppose that \(EX_1=0, EX_1^2=1,\) and \(n^pP(X>n)\to 0\) for some \(p>6\), \(EX_1^2I(|X_1|>n)=o(1/\log n)\) \( (n\to \infty).\) Then \[ 1\leq \limsup_{n\to \infty}\frac {M_n\;(L_n)}{\sqrt{2\log n}}\leq \sqrt{2}\quad \text{ a.s.} \] They may be called the strong law of large numbers and the law of the iterated logarithm type results for \(M_n\;(L_n)\) because \textit{B. G. Pittel} [Ann. Probab. 9, 119--129 (1981; Zbl 0453.60017)] has proved that \[ \lim_{n\to \infty}\frac {L_n}{\log n/\log \log n}=1\quad \text{ a.s}. \]