Definition
Let $X$ be an integrable random variable and let $(\mathcal F_n)$ be a filtration. Define
$$ M_n=\mathbb E[X\mid \mathcal F_n]. $$Then $(M_n)$ is called Doob’s martingale.
Why it is a martingale
Using the tower property,
$$ \mathbb E[M_{n+1}\mid \mathcal F_n]
\mathbb E[\mathbb E[X\mid \mathcal F_{n+1}]\mid \mathcal F_n]
\mathbb E[X\mid \mathcal F_n]
M_n. $$
Important consequence
Doob’s martingale is automatically uniformly integrable. Therefore, for any stopping time $T$,
$$ \mathbb E[M_T]=\mathbb E[M_0]. $$Interpretation
$M_n$ is the best prediction of $X$ based on the information available at time $n$.
Source Links
- STA447
- Phase 5.2-5.5 Atomic Reading Order
- Phase 5.5 — Doob’s Maximal Inequality + Wald’s Identity
- Previous atom: Wald’s Identity
- Next atom: none