Universality of the asymptotics of the one-sided exit problem for integrated processes (Q1943327): Difference between revisions

The classical one-sided exit problem is to find a persistence exponent \(\theta>0\) such that \[ \operatorname P\Big\{\sup_{t\in[0,T]}A_t\leq1\Big\}=T^{-\theta+o(1)}\quad\text{ as }T\to\infty, \] where \((A_t)_{t\geq0}\) or \((A_t)_{t\in\mathbb N_0}\) is a stochastic process in \(\mathbb R\) starting at the origin. The authors consider integrated processes of the form \[ A_t=I(X)_t=\int_0^tK(t-s)X_s\,ds, \] where \((X_t)_{t\geq0}\) is a Lévy martingale or a random walk \(X_t=S(\lfloor t\rfloor)\) with finite exponential moment and \(K(x)\leq k(x^{\alpha-1}+x^{\beta-1})\) for \(x>0\) and certain \(\alpha\geq\beta>0\) is a measurable function. In their main result, the authors link the asymptotic behavior of \(\operatorname P\{\sup_{t\in[0,T]}I(X)_t\leq1\}\) to the corresponding quantity for \(I(W)_t\), where \((W_t)_{t\geq0}\) is a suitable Brownian motion, and thus show that the persistence exponent (if it exists) is universal over all processes \((X_t)_{t\geq0}\) under consideration. Consequently, using classical results, the persistence exponent of the integrated Lévy processes and random walks (with \(K\equiv1\)) is \(\theta=1/4\). The authors further show the existence of the persistence exponent \(\theta(\alpha)\) for fractionally integrated Lévy processes with \(K(x)=\Gamma(\alpha)^{-1}x^{\alpha-1}\), where \(\alpha\mapsto\theta(\alpha)\in(0,1/2]\) is some non-increasing (unknown) function. It is further shown that the problem is robust to certain changes of the barrier \(F\equiv1\) in the classical one-sided exit problem, namely for barriers with \(1-F\) being absolutely continuous with density in \(L^2\). The questions have various applications and the authors show how their results improve the upper bound of a certain asymptotic rate for random polynomials having no zeros.