Why are martingales described as fair games?

Because the defining property says that, given everything known so far, the expected future value equals the current value; there is no predictable drift up or down, exactly the condition for a game in which neither player has an edge.

What makes martingales so useful beyond gambling?

Their convergence theorems, optional-stopping theorem, and maximal inequalities apply under very weak assumptions, so many quantities in probability, statistics, and finance can be analyzed simply by recognizing or constructing an appropriate martingale.

Martingales

A martingale is a model of a fair game: a sequence of random variables whose expected next value, given all past information, equals its present value, a structure that yields some of the most powerful tools in probability.

Definition

A martingale is a sequence of integrable random variables adapted to a filtration such that the conditional expectation of each term given the past equals the previous term, formalizing a fair game in which no betting strategy yields a systematic gain.

Scope

The area covers filtrations and adapted processes, the definitions of martingales, submartingales, and supermartingales, the Doob decomposition, stopping times and the optional stopping theorem, the martingale convergence theorems and uniform integrability, Doob's maximal and Lp inequalities, and the role of martingales as a unifying device throughout modern probability.

Sub-topics

Core questions

What does it mean for a process to be a fair game relative to an information flow?
How does the optional stopping theorem constrain the value of a martingale at a random time?
Under what conditions does a martingale converge, and in what sense?
How do martingale inequalities control the maximum of a process?

Key theories

Optional stopping theorem: Under suitable conditions on a stopping time, the expected value of a martingale at that random time equals its initial value, formalizing the impossibility of beating a fair game and providing a versatile computational tool for hitting probabilities and expected durations.
Martingale convergence theorem: A martingale that is bounded in the first mean converges almost surely, and under uniform integrability it also converges in the first mean and is closed by its limit, a result of remarkable generality that subsumes many convergence statements.

Clinical relevance

Martingales are the mathematical backbone of arbitrage-free pricing in mathematical finance, where discounted asset prices are martingales under a risk-neutral measure; they also underlie sequential analysis and optional-stopping arguments in statistics, the analysis of randomized algorithms through concentration inequalities, and stochastic approximation.

History

The word martingale entered probability through Jean Ville's 1939 work on gambling systems, and Joseph Doob developed the systematic theory in the 1940s and 1950s, including the convergence and optional-stopping theorems and the maximal inequalities that made martingales a central tool of the field.

Key figures

Joseph L. Doob
Paul Levy
Jean Ville
David Williams

Seminal works

doob1953
williams1991

Frequently asked questions

Why are martingales described as fair games?: Because the defining property says that, given everything known so far, the expected future value equals the current value; there is no predictable drift up or down, exactly the condition for a game in which neither player has an edge.
What makes martingales so useful beyond gambling?: Their convergence theorems, optional-stopping theorem, and maximal inequalities apply under very weak assumptions, so many quantities in probability, statistics, and finance can be analyzed simply by recognizing or constructing an appropriate martingale.