Strong limit theorems of empirical functionals for large exceedances of partial sums of i.i.d. variables (Q1180582)

Let \(\{ (X_ k,U_ k),\;k\geq 1\}\) be i.i.d. (with possible dependence between \(X_ k\) and \(U_ k\)) such that \(X_ 1\) and \(U_ 1\) have bounded range, \(E\;X_ 1<0\) and \(P(X_ 1>0)>0\). Put \(S_ m = \sum_{k=1}^ m X_ k,\;S_ 0=0\), and define the stopping times \[ K_ 0=0,\quad K_ \nu = \min \{k:\;k>K_{\nu -1},\;S_ k - S_{K_{\nu -1}}\leq 0,\;\nu = 1,2,\dots \} \] and the first exit time of \(S_ k - S_{K_{\nu -1}}\) from the range \((0,y),\;y>0\), i.e. \[ T_ \nu (y) = \inf \{ k:\;k>K_{\nu-1}\text{ such that either }S_ k - S_{K_{\nu -1}}\leq 0\text{ or }S_ k - S_{K_{\nu-1}}\geq y\} . \] Define \(I_ \nu (y)= 1\) if \(T_ \nu (y)<K_ \nu\) and \(I_ \nu (y) = 0\) otherwise. Put \(L_ \nu (y) = T_ \nu (y) - K_{\nu -1}\) and \(W_ \nu (y) = \sum_{i=K_{\nu -1}+1}^{T_ \nu (y)} U_ i.\) Theorem 1: Under the assumptions above, let \(T_ \nu (y)\) be the first time in the above succession where \(I_ \nu (y) = 1\). Then \(L_ \nu (y)/y \to 1/E\;Xe^{\theta^* X} (>0)\) a.s. as \(y\to \infty\), where \(\theta^*\) is the unique position root of the equation \(E\;e^{\theta X} = 1\). Theorem 2: Excluding a set of measure zero with the index \(\nu\) determined as in Theorem 1, then \(W_ \nu (y)/L_ \nu(y) \to E\;Ue^{\theta^* X}\) a.s. as \(y\to \infty\). Applications of these two theorems concern molecular sequence comparisons, queueing theory, insurance risk models and traffic flow. The authors are mainly interested in the model when \(X_ 1,X_ 2,\dots \) are i.i.d. based on observations from a finite alphabet \(\{ a_ i,\;i=1,\dots ,s\}\) where \(P\{ X=s_ i\} = p_ i,\;i=1,\dots ,r,\;p_ i>0,\;\sum p_ i =1\) to be interpreted as that sampling the letter \(a_ i\) yields a score \(s_ i\). Defining \(U=U(X) = 1\) if \(X=s_ 1\) and 0, otherwise, the quantity \(W(y)/L(y)\) counts the proportion of occurrences of \(a_ 1\) during an excursion confined strictly to positive values over a time segment achieving a level beyond \(y,\;y\to \infty\). Corollary 1: The empirical frequency distribution \(\mu(y)\) of the letters \(\{ a_ 1,\dots ,a_ r\}\) observed during the excursion epoch, defined in Theorems 1 and 2, converges as \(y\to \infty\) with probability 1 to the frequency measure \(\mu^*\) which takes the value \(a_ i\) with probability \(p_ i e^{\theta^* s_ i},\;i=1,\dots ,r\).