Self-normalized lag increments of partial sums (Q1613101)

Let \(\{X,X_n\), \(n\geq 1\}\) be an i.i.d. sequence of random variables, and let \(\{a_n\), \(n\geq 1\}\) be a sequence of non-negative integers satisfying \(1\leq a_n\leq n\). For \(n=1,2,\dots\), \(k=1,2,\dots ,n\), set \[ S_0=0,\quad S_n=\sum_{i=1}^{n}X_i, \quad V^2_{n,k}=\sum_{i=n-k+1}^{n}X_i^2, \quad d(n,k)=\{2(\log(n/k)+\log\log k)\}^{1/2}, \] where \(\log x=\log_e(\max\{x,1\})\) and \(\log \log x=\log_e\log_e(\max\{x,e\})\). Using a suitable self-normalizing (or studentizing) factor for lag sums of i.i.d. random variables, a self-normalized lag increment theorem of partial sums is established. Theorem 1.1. Suppose that \({\mathbf E}X=0\) and \({\mathbf E}(X^2I\{|X|\leq x\})\) is slowly varying as \(x\to \infty\). Then we have \[ \limsup_{N\to \infty}\max_{a_n\leq k\leq N}\frac{S_N-S_{N-k}}{V_{N,k}d(N,k)}=1{\text{ a.s.,}} \] \[ \limsup_{N\to \infty}\max_{a_n\leq k\leq N} \max_{k\leq n\leq N} \max_{1\leq j\leq k} \frac{S_n-S_{n-j}}{V_{N,k}d(N,k)}=1{\text{ a.s.}} \] If we also have that \(a_N/\log N\to \infty\) as \(N\to \infty\) and for some \(\tau>0\), \(N/a^\tau_N\) is non-decreasing, and \(\lim_{N\to\infty} \frac{\log(N/a_N)}{\log\log N}= \infty\), then we also have \[ \lim_{N\to \infty} \max_{a_n\leq k\leq N}\max_{k\leq n\leq N}\frac{S_n-S_{n-k}}{V_{N,k}d(N,k)} =1 \text{ a.s.}, \] \[ \lim_{N\to \infty}\max_{a_n\leq k\leq N} \max_{k\leq n\leq N} \max_{1\leq j\leq k} \frac{S_n-S_{n-j}}{V_{N,k}d(N,k)} =1 \text{ a.s}. \]