Skip to content

Martingale

Definition

Process X is a martingale if for all n:

\[ E[X_{n+1}|F_n] = X_n \]

Where

  • Fn is the history of Xn (called filtration)

Which means, on the n+1 step, the expectation of X shall be the same as any step before.

A martingale may be thought of as a “fair game”, because given the information in current and previous plays (Fn), you don't expect to change your total winning (X).

Example

Here are two practical example to help you understand it.

Partial sum process

A simple coin game, in \(i\) turn we bet \(X_i\), and \(S_n\) is our winning money after n turn.

Let Xi be independent, define the partial sum process:

\[\begin{aligned} S_0 &= 0, \\ S_n &= \sum_{ i=1 }^n X_i, n=1, 2,... \end{aligned}\]

Sn is a martingale iff:

\[E(X_i) = 0\]

Proof

With Xi being independent, and E(Xi) = 0, we have:

\[\begin{aligned} E[S_{n+1}|X_1,...X_n] &= E[S_n + X_{n+1} | X_1,...X_n] \\ &= S_n + E[X_{n+1} | X_1,...X_n] \\ &= S_n + E[X_{n+1}] \\ &= S_n \end{aligned}\]

Gambler's Ruin Problem

A classic problem on martingale.

Consider a gambler who starts with an initial fortune of $1 and then on each successive gamble either wins $1 or loses $1 independent of the past with probabilities p and q = 1−p respectively. Let Rn denote the total fortune after the n th gamble. The gambler’s objective is to reach a total fortune of $N, without first getting ruined (running out of money). If the gambler succeeds, then the gambler is said to win the game.

Let \(P_i\) denote the probability that the gambler wins when the initial money \(R_0 = i\), we have:

\[P_i = pP_{i+1} + qP_{i-1}\]

This is because P_i can only lead to two states:

  • Winning $1 with probability p to state \(P_{i+1}\) 
  • Losing $1 with probability q to state \(P_{i-1}\)

Subtract P_{i-1} from both sides of the equation, we get:

\[P_i = (p + q) P_i = pP_{i+1} + qP_{i-1}\]

i.e.,

\[{P_{i+1} - P_i}= \frac{ q }{ p }(P_i - P_{i-1})\]

Thus

\[\begin{aligned} P_{i+1} - P_1 &= \sum_{k=1}^i (P_{k+1} - P_k) \\ &= \sum_{k=1}^i (\frac{q}{p})^k P_1 \end{aligned}\]

We have

\[P_{i+1}= \begin{cases} P_1 \frac{1-(q/p)^{i+1}}{1-(q/p)} &, p \neq q\\ P_1 (i+1) &, p=q \end{cases}\]

To solve P_1, pick i = N-1 and use the fact that P_N = 1

\[P_1= \begin{cases} \frac{1-(q/p)}{1-(q/p)^N} &, p \neq q\\ 1/N &, p=q \end{cases}\]

Substitute P_1 and we have:

\[P_i= \begin{cases} \frac{ 1-(q/p)^i}{ 1-(q/p)^N } &, p \neq q \\ i/N &, p=q \end{cases}\]

Reference

  1. martingales.dvi (rice.edu)
  2. 4700-07-Notes-GR.pdf (columbia.edu)