The question is to find an example of a non-negative martingale $(X_n)$ with $EX_n=1$ for all $n$ such that $X_n$ converges almost surely to a random variable $X$ where $EX\neq 1$ and $\text{Var}(X)>0$.

My attempt

An example of a martingale that I thought could fit the bill was the product martingale with $X_n=\prod_{i=1}^n Y_i$ where $(Y_{i})$ are i.i.d non-negative random variables with mean $1$ and $P(Y_i=1)<1$. Unfortunately $X_n\to 0$ a.s and hence the limit is degenerate. Other examples, I tried to cook up (e.g. branching process with one individual) all had degenerate limits.

I am having trouble coming up with an example that does not have a degenerate limit.


Modifying your example a little, by letting $Y_0$ be any non-negative random variable independent of $\{Y_n\}_{n\ge 1}$ with $E(Y_0)=1$, $$ M_0=\frac12Y_0+\frac 12,\ \ M_n=\frac12\left(Y_0+\prod_{i=1}^nY_i\right),\ \ n\ge 1 $$ is a martingale converging to $\frac 12 Y_0$.



Convex combinations of martingales are martingale. Your example converges to $0$. modify if to get an example converging to something else, mix the two.


Let $X_n$ be the martingale you described; $X_n=\prod_{i=1}^n Y_i$, where $Y_i\stackrel{\text{iid}}\sim \operatorname{Unif}\{1/2,3/2\}$.

Let $(X_n')_{n\ge 1}$ be an independent copy of $(X_n)_{n\ge 1}$.

Finally, let $\xi$ be Bernoulli$(1/3)$, independent of the previous variables. Then $$ X_n\cdot{\bf 1} ({\xi=1})+(2-X_n')\cdot {\bf 1}(\xi=0) $$ is a martingale whose expectation is always one, and in the limit equals $0$ with probability $1/3$ and equals $2$ with probability $2/3$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.