Quant GT
Browse all lessons
Section 13 · Lesson 13.5

Martingales

Fair-game processes and the optional stopping theorem.

A martingale is a stochastic process XtX_t where the conditional expectation of the future given the past equals the present:

E[Xt+sXu,ut]=XtE[X_{t+s} \mid X_u, u \le t] = X_t

Translation: a martingale is a fair game. No matter what's happened so far, your expected future value is exactly where you are now.

Brownian motion is a martingale. So is a fair-coin gambling balance with one-unit wagers. Discounted asset prices under the risk-neutral measure are martingales — that's what risk-neutral valuation means.

The Optional Stopping Theorem says that for a "nice" stopping time τ\tau, the expectation at the stopping time equals the starting value: E[Xτ]=X0E[X_\tau] = X_0. This sounds obvious but yields powerful results — the "double-or-nothing" gambler's ruin, the expected hitting time of barriers, and many option-pricing identities.

The trap: optional stopping can fail when τ\tau is unbounded and the martingale isn't uniformly integrable. The classic example is doubling your bet until you win — the strategy looks profitable but requires unlimited capital.