LAUSR

In this paper, we develop a regression-based approach for optimal stopping problems in a dual manner. The method is purely dual as it does not require given approximations to Snell envelope. This method produces stopping policies and upper and lower bounds for the stopping problem. Asymptotic properties and finite-sample errors for the bounds are analysed. We show that this method benefits from the variance property of martingale duality. We also apply this method to option pricing problems and compare it with state-of-the-art approaches. The results demonstrate its effectiveness in high-dimensional optimal stopping problems.