Optional Stopping Theorem

The optional stopping theorem says that, under conditions that rule out unbounded waiting, stopping a fair game at a cleverly chosen random time cannot change its expected value.

Definition

The optional stopping theorem asserts that for a martingale and a stopping time satisfying suitable integrability or boundedness conditions, the expected value of the martingale at the stopping time equals its initial expected value, so a stopped martingale is still a martingale.

Scope

This topic covers stopping times and the stopped process, the statement of the optional stopping theorem and its hypotheses such as bounded stopping times, bounded martingales, or uniform integrability, counterexamples like the doubling strategy that show why hypotheses are needed, and applications to gambler's ruin, hitting probabilities of random walks, and Wald's identities.

Core questions

• What is a stopping time and what does it mean to stop a process at one?
• Under what conditions does optional stopping preserve the expectation?
• Why do some stopping strategies appear to beat a fair game, and what hypothesis fails?
• How does the theorem yield hitting probabilities and expected hitting times?

Key theories

Optional stopping under sufficient conditions

If a stopping time is bounded, or the martingale is bounded, or the family of stopped values is uniformly integrable, then the expectation of the martingale at the stopping time equals its starting value, preserving the fair-game property.

Wald's identities and ruin problems

Applying optional stopping to the random-walk martingale yields Wald's first and second identities relating the stopped sum to the stopping time and gives explicit gambler's-ruin probabilities and expected durations.

Clinical relevance

Optional stopping is the rigorous reason no betting system can beat a fair game, it gives clean derivations of ruin and hitting probabilities for random walks, and in sequential statistics it controls the error of tests that stop adaptively as data arrive.

History

Doob formulated optional sampling in the 1940s and 1950s, generalising Wald's sequential-analysis identities of the 1940s, and the theorem with its careful hypotheses, illustrated by the failure of the doubling strategy, became a cornerstone of applied martingale theory and mathematical finance.

Key figures

• Joseph Doob
• Abraham Wald
• David Williams

Seminal works

• doob1953

Frequently asked questions

Can a clever stopping rule beat a fair game?

No, provided the conditions of the optional stopping theorem hold; strategies that seem to win, like doubling bets, require unbounded capital or infinite expected time, which violate the theorem's hypotheses.

What is a stopping time?

It is a random time whose occurrence can be decided using only information available up to that moment, without looking into the future, such as the first time a process hits a given level.