Home Statistical Dictionary Optional Stopping Theorem

Optional Stopping Theorem

The result that, under suitable regularity conditions, the expected value of a martingale evaluated at a random stopping time equals its initial expected value, you cannot gain an edge in a fair game by choosing a clever moment to stop.

Stochastic ProcessesAdvancedFree

In Plain English

Imagine playing a genuinely fair game (a martingale) and using some strategy to decide when to walk away, maybe you stop as soon as you're up \$10, or after a fixed number of rounds. The optional stopping theorem says that, as long as your stopping strategy satisfies some technical conditions (it can't rely on seeing into the future, and can't let you wait forever hoping for an unbounded windfall), your expected winnings when you stop are exactly the same as when you started, zero net gain, on average, no matter how cleverly the stopping rule is designed. It's the rigorous version of 'you can't beat a fair game by choosing when to quit.'

Definition

Let ${M_{n}}$ be a martingale with respect to filtration ${F_{n}}$ , and let $τ$ be a stopping time (a random time whose occurrence, by time $n$ , is determined by $F_{n}$ alone). The optional stopping theorem states that $E [M_{τ}] = E [M_{0}]$ provided at least one of several sufficient conditions holds: (a) $τ$ is bounded ( $τ \leq N$ for some fixed $N$ ); (b) $τ$ has finite expectation and the martingale's increments are uniformly bounded; or (c) the martingale is uniformly integrable and $τ$ is finite almost surely. Without one of these conditions, the conclusion can fail dramatically, the classic counterexample is a symmetric simple random walk started at 0, stopped at $τ = in f {n : M_{n} = 1}$ (the first time it hits $+ 1$ ): $τ$ is finite almost surely (since the walk is recurrent), but $E [M_{τ}] = 1 \neq = 0 = E [M_{0}]$ , because $τ$ has infinite expectation, violating the needed regularity.

Formula

E [M_{τ}] = E [M_{0}]

Notation

M_{n}

—A martingale process

τ

—A stopping time, a random time depending only on information available up to that time

Properties

The theorem's regularity conditions are not bureaucratic fine print, dropping them entirely changes the answer, the random-walk hitting-time counterexample is a standard warning that 'the game is fair, and I stop eventually' is not, by itself, enough to conclude the expected payoff is unchanged.
In finance, the optional stopping theorem underlies the 'no arbitrage from timing alone' principle for martingale price processes, if a discounted asset price is a martingale under the risk-neutral measure, no trading or exit strategy based on that price alone can produce positive expected profit.
The theorem is a core tool for deriving classical results outside its home field, including Wald's identity for random sums and elegant proofs of gambler's ruin probabilities, by cleverly choosing an auxiliary martingale and a well-chosen stopping time.

At a Glance

Conclusion

E [M_{τ}] = E [M_{0}]

: no expected advantage from clever stopping

Requires one ofbounded stopping time; bounded increments + finite expected stopping time; or uniform integrability

Classic failure casehitting time of

+ 1

for a symmetric random walk:

τ

has infinite expectation

Interpretationa rigorous 'no free lunch' guarantee for fair-game strategies

Last updated September 8, 2026← Back to the Dictionary