Asymptotic e-processes

TL;DR

This paper introduces asymptotic e-processes, a new theoretical framework for approximate sequential hypothesis testing, providing bounds and properties relevant for practical applications with model errors.

Contribution

It develops the concept of asymptotic e-processes, derives an asymptotic Ville's inequality, and explores their properties and construction methods.

Findings

Derived an asymptotic version of Ville's inequality.

Established connections between asymptotic e-processes and supermartingales.

Provided examples of constructing asymptotic e-processes from e-variables.

Abstract

We introduce the concept of an asymptotic e-process, which is a doubly indexed stochastic process $(E_{m,n})_{m,n\in\mathbb{N}}$ that approximates an e-process with monitoring time $n$ in terms of a suitable limiting behavior for an approximation parameter $m\to \infty$. This theory is motivated by practical applications in sequential hypothesis testing, in which e-variables can only be constructed approximately from observations due to model misspecification or estimation errors. We derive an asymptotic version of Ville's inequality, which bounds excursion probabilities of $(E_{m,n})_{m,n\in\mathbb{N}}$ over some threshold uniformly over $n$ up to a time horizon $r_m$ that is determined by the quality of process approximation over $m$. We investigate properties of asymptotic e-processes, their connections to asymptotic supermartingales, and provide examples of how they can be constructed from asymptotic e-variables.