Quickest Detection Problems for Ornstein–Uhlenbeck Processes

Abstract

Consider an Ornstein–Uhlenbeck process that initially reverts to zero at a known mean-reversion rate β0, and then after some random/unobservable time, this mean-reversion rate is changed to β1. Assuming that the process is observed in real time, the problem is to detect when exactly this change occurs as accurately as possible. We solve this problem in the most uncertain scenario when the random/unobservable time is (i) exponentially distributed and (ii) independent from the process prior to the change of its mean-reversion rate. The solution is expressed in terms of a stopping time that minimises the probability of a false early detection and the expected delay of a missed late detection. Allowing for both positive and negative values of β0 and β1 (including zero), the problem and its solution embed many intuitive and practically interesting cases. For example, the detection of a mean-reverting process changing to a simple Brownian motion ([Formula: see text] and [Formula: see text]) and vice versa ([Formula: see text] and [Formula: see text]) finds a natural application to pairs trading in finance. The formulation also allows for the detection of a transient process becoming recurrent ([Formula: see text] and [Formula: see text]) as well as a recurrent process becoming transient ([Formula: see text] and [Formula: see text]). The resulting optimal stopping problem is inherently two-dimensional (because of a state-dependent signal-to-noise ratio), and various properties of its solution are established. In particular, we find the somewhat surprising fact that the optimal stopping boundary is an increasing function of the modulus of the observed process for all values of β0 and β1.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 5 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.