Persistence of Iterated Partial Sums

Amir Dembo,Jian Ding,Fuchang Gao

DOI: https://doi.org/10.1214/11-aihp452

2013-01-01

Annales de l Institut Henri Poincaré Probabilités et Statistiques

Abstract:Let S-n((2)) denote the iterated partial sums. That is, S-n((2)) = S-1 + S-2 + ... + S-n, where S-i = X-1 + X-2 + ... + X-i. Assuming X-1, X-2, ..., X-n are integrable, zero-mean, i.i.d. random variables, we show that the persistence probabilitiesp(n)((2)) := P(max(1 <= i <= n)S(i)((2)) < 0) <= c root E vertical bar Sn+1 vertical bar/(n + 1)E vertical bar X-1 vertical bar,with c <= 6 root 30) (and c = 2 whenever X-1 is symmetric). The converse inequality holds whenever the non-zero min(-X-1, 0) is bounded or when it has only finite third moment and in addition X-1 is squared integrable. Furthermore, p(n)((2)) asymptotic to n(-1/4) for any non-degenerate squared integrable, i.i.d., zero-mean X-i. In contrast, we show that for any 0 < gamma < 1/4 there exist integrable, zero-mean random variables for which the rate of decay of p(n)((2)) is n(-gamma).