Here, I give a quick review of the concept of a Martingale. A Martingale is a sequence of random variables satisfying a specific expectation conservation law. If one can identify a Martingale relating to some other sequence of random variables, its use can sometimes make quick work of certain expectation value evaluations.

This note is adapted from Chapter 2 of Stochastic Calculus and Financial Applications, by Steele.


Often in random processes, one is interested in characterizing a sequence of random variables \(\{X_i\}\). The example we will keep in mind is a set of variables \(X_i \in \{-1, 1\}\) corresponding to the steps of an unbiased random walk in one-dimension. A Martingale process \(M_i = f(X_1, X_2, \ldots X_i)\) is a derived random variable on top of the \(X_i\) variables satisfying the following conservation law

\begin{align} \tag{1} E(M_i | X_1, \ldots X_{i-1}) = M_{i-1}. \end{align}

For example, in the unbiased random walk example, if we take \(S_n = \sum_{i=1}^n X_i\), then \(E(S_n) = S_{n-1}\), so \(S_n\) is a Martingale. If we can develop or identify a Martingale for a given \(\{X_i\}\) process, it can often help us to quickly evaluate certain expectation values relating to the underlying process. Three useful Martingales follow.

  1. Again, the sum \(S_n = \sum_{i=1}^n X_i\) is a Martingale, provided \(E(X_i) = 0\) for all \(i\).
  2. The expression \(S_n^2 - n \sigma^2\) is a Martingale, provided \(E(X_i) = 0\) and \(E(X_i^2) = \sigma^2\) for all \(i\). Proof:
    \begin{align} \tag{2} E(S_n^2 | X_1, \ldots X_{n-1}) &= \sigma^2 + 2 E(X_n) S_{n-1} + S_{n-1}^2 - n \sigma^2\\ &= S_{n-1}^2 - (n-1) \sigma^2. \end{align}
  3. The product \(P_n = \prod_{i=1}^n X_i\) is a Martingale, provided \(E(X_i) = 1\) for all \(i\). One example of interest is
    \begin{align} \tag{3} P_n = \frac{\exp \left ( \lambda \sum_{i=1}^n X_i\right)}{E(\exp \left ( \lambda X \right))^n}. \end{align}
    Here, \(\lambda\) is a free tuning parameter. If we choose a \(\lambda\) such that \(E(\exp(\lambda X)) = 1\) for our process, we can get a particularly simple form.

Stopped processes

In some games, we may want to setup rules that say we will stop the game at time \(\tau\) if some condition is met at index \(\tau\). For example, we may stop a random walk (initialized at zero) if the walker gets to either position \(A\) or \(-B\) (wins \(A\) or loses \(B\)). This motivates defining the stopped Martingale as,

\begin{align} M_{n \wedge \tau} = \begin{cases} M_n &\text{if } \tau \geq n \\ M_{\tau} &\text{else}. \tag{4} \end{cases} \end{align}

Here, we prove that if \(M_n\) is a Martingale, then so is $M_{n \wedge \tau} $. This is useful because it will tell us that the stopped Martingale has the same conservation law as the unstopped version.

First, we note that if \(A_i \equiv f_2(X_1, \ldots X_{i-1})\) is some function of the observations so far, then the transformed process

\begin{align} \tag{5} \tilde{M}_n \equiv M_0 + \sum_{i=1}^n A_i (M_i - M_{i-1}) \end{align}

is also a Martingale. Proof:

\begin{align} \tag{6} E(\tilde{M}_n | X_1, \ldots X_{n-1}) = A_n \left ( E(M_n) - M_{n-1} \right) + \tilde{M}_{n-1} = \tilde{M}_{n-1}. \end{align}

With this result we can prove the stopped Martingale is also a Martingale. We can do that by writing \(A_i = 1(\tau \geq i)\) — where \(1\) is the indicator function. Plugging this into the above, we get the transformed Martingale,

\begin{align} \nonumber \tag{7} \tilde{M}_n &= M_0 + \sum_{i=1}^n 1(\tau \geq i) (M_i - M_{i-1}) \\ &= \begin{cases} M_n & \text{if } \tau \geq n \ M_{\tau} & \text{else}. \end{cases} \end{align}

This is the stopped Martingale — indeed a Martingale, by the above.

Example applications

Problem 1

Consider an unbiased random walker that takes steps of size \(1\). If we stop the walk as soon as he reaches either \(A\) or \(-B\), what is the probability that he is at \(A\) when the game stops?

Solution: Let \(\tau\) be the stopping time and let \(S_n = \sum_{i=1}^n X_i\) be the walker’s position at time \(n\). We know that \(S_n\) is a Martingale. By the above, so then is \(S_{n \wedge \tau}\), the stopped process Martingale. By the Martingale property

\begin{align} \tag{8} E(S_{n \wedge \tau}) = E(S_{i \wedge \tau}) \end{align}

for all \(i\). In particular, plugging in \(i = 0\) gives \(E(S_{n \wedge \tau}) = 0\). If we take \(n \to \infty\), then

\begin{align} \tag{9} \lim_{n \to \infty} E(S_{n \wedge \tau}) \to E(S_{\tau}) = 0. \end{align}

But we also have

\begin{align} \tag{10} E(S_{\tau}) = P(A) * A - (1 - P(A)) B. \end{align}

Equating (9) and (10) gives

\begin{equation} \tag{11} P(A) = \frac{B}{A + B} \end{equation}
Problem 2

In the game above, what is the expected stopping time? Solution: Use the stopped version of the Martingale \(S_n^2 - n \sigma^2\).

Problem 3

In a biased version of the random walk game, what is the probability of stopping at \(A\)? Solution: Use the stopped Martingale of form \(P_n = \frac{\exp \left ( \lambda \sum_{i=1}^n X_i\right)}{E(\exp \left ( \lambda X \right))^n}\), with \(\exp[\lambda] = q/p\), where \(p = 1-q\) is the probability of step to the right.

Like this post? Share on: Twitter | Facebook | Email

Jonathan Landy Avatar Jonathan Landy Jonathan grew up in the midwest and then went to school at Caltech and UCLA. Following this, he did two postdocs, one at UCSB and one at UC Berkeley.  His academic research focused primarily on applications of statistical mechanics, but his professional passion has always been in the mastering, development, and practical application of slick math methods/tools. He currently works as a data-scientist at Stitch Fix.