Discrete-Time Martingales

A discrete-time martingale is a sequence of random variables, indexed by time and tied to a growing flow of information, whose best forecast of the next value given the past is always its current value.

Definition

A discrete-time martingale is an integrable sequence of random variables adapted to a filtration for which the conditional expectation of each term given the earlier information equals the immediately preceding term.

Scope

The topic covers filtrations and adapted, predictable processes, the definitions of martingale, submartingale, and supermartingale, the conditional-expectation property and its consequences, the Doob decomposition of a submartingale into a martingale and an increasing predictable part, martingale transforms representing the gains of a betting strategy, and standard examples such as sums of independent centered variables and likelihood-ratio processes.

Core questions

• What information structure does a filtration encode, and what does it mean for a process to be adapted?
• How do martingales, submartingales, and supermartingales differ?
• How does the Doob decomposition separate a process into a fair-game part and a trend?
• Why can no predictable betting strategy turn a martingale into a winning game?

Key concepts

• filtration
• adapted and predictable processes
• submartingale and supermartingale
• Doob decomposition
• martingale transform

Key theories

Doob decomposition

Any adapted integrable process splits uniquely into a martingale plus a predictable process starting at zero, and the process is a submartingale exactly when this predictable part is increasing, isolating the systematic trend from the fair-game fluctuations.

Martingale transform and fairness of fair games

The accumulated gains from a predictable betting strategy applied to a martingale form another martingale, so no strategy using only past information can produce a positive expected gain, the precise statement that a fair game cannot be beaten.

Clinical relevance

Discrete-time martingales formalize sequential information and fair betting, underpinning the sequential likelihood-ratio tests of statistics, the no-arbitrage condition in discrete financial models, and the construction of martingale difference sequences used to prove concentration inequalities and limit theorems for dependent data.

History

Ville introduced martingales to refute the existence of successful gambling systems, and Doob built the discrete-time theory with the decomposition that bears his name, making martingales a standard tool whose treatment in Williams' text became a model of exposition.

Key figures

• Joseph L. Doob
• Jean Ville
• Jacques Neveu

Seminal works

• williams1991

Frequently asked questions

What is a filtration?

A filtration is an increasing family of sigma-algebras, one for each time, representing the information available up to that time; a process is adapted to it when each value is known given the information at its own time.

What distinguishes a submartingale from a supermartingale?

A submartingale tends to increase in conditional mean, since its expected next value given the past is at least the present value, while a supermartingale tends to decrease; a martingale is exactly the borderline case where the conditional mean is unchanged.