Review of Conditional Expectation and Martingale

Absolute Continuity

We say $\nu$ is absolutely continuous with respect to $\mu$, denoted by $\nu \ll \mu$, if $\mu (A)=0$ implies $\nu (A)=0$.

Radon-Nikodym Theorem

Let $\nu$ and $\mu$ be sigma-finite measures on $(\Omega ,\mathcal{F})$. If $\nu$ is absolutely continuous with respect to $\mu$, i.e., $\nu \ll \mu$, then there will be a function $f\in \mathcal{F}$ such that for all $A\in \mathcal{F}$, $$\nu (A)={\int }_{A}fd\mu .$$ $f$ is denoted by $\frac{d\nu }{d\mu }$ and called the Radon-Nikodym derivative.

Conditional Expectation

Consider a probability space $(\Omega ,\mathcal{F},P)$ and a random variable $X$ on $(\Omega ,\mathcal{F},P)$. Let ${\mathcal{F}}^{\ast }\subset \mathcal{F}$ be a sub-sigma field. Define the measure $\nu$ on ${\mathcal{F}}^{\ast }$:
$$\nu (A)={\int }_{A}XdP,\forall A\in {\mathcal{F}}^{\ast }.$$ If $X\ge 0$, it can be shown that $\nu$ is a measure on ${\mathcal{F}}^{\ast }$ by DCT. For the general case, we can write $X=X^{+}-X^{-}$. Since $\nu \ll P$, Radon-Nikodym says that there exists $Y\in {\mathcal{F}}^{\ast }$ such that $Y=\frac{d\nu }{dP}$. $Y$ is called the conditional expectation of $X$ given ${\mathcal{F}}^{\ast }$ and denoted by $Y=E[X|{\mathcal{F}}^{\ast }]$.

Conditional expectation holds nearly all the properties of expectation.

The Smaller $\sigma$-field Always Wins

If ${\mathcal{F}}_1\subset {\mathcal{F}}_2$, then $$E[E[X|{\mathcal{F}}_1]|{\mathcal{F}}_2]=E[X|{\mathcal{F}}_1]=E[E[X|{\mathcal{F}}_2]|{\mathcal{F}}_1].$$Let $Y$ be a random variable and $Z=g(Y)$ where $g$ is a mearsurable function from $(\mathbb{R},\mathcal{B})$ to $(\mathbb{R},\mathcal{B})$. Then $Y^{-1}{g}^{-1}(\mathcal{B})\subset Y^{-1}(\mathcal{B})$. As a result,
$E[E[X|Y]|Z]=E[X|Z]$. In addition, $E[Z|Y]=Z$.

Conditional Expectation is the Projection in $L_2$ Space.

Suppose $E{X}^2<\infty$, then $E[X|\mathcal{F}]$ is the element $Y\in \mathcal{F}$ that minimizes $E(X-Y{)}^2$.

Filtration

A filtration is an increasing sequence of sigma-fields.

Martingale

Let ${\mathcal{F}}_n$ be a filtration. A random sequence $X_n$ with

1. $E|X_n|<\infty$, (for finite n)
2. $X_n\in {\mathcal{F}}_n$,
3. $E[X_{n+1}|{\mathcal{F}}_n]=X_n$ for all n.

is said to be a martingale. If in the last equation, = is replaced by $\le$ or $\ge$, then { X n } \{X_n\} {Xn​} is said to be a supermartingale or submartingale.

Equivalent Conditions

For a supermartingale, $E[X_n|{\mathcal{F}}_m]\le X_m$ for all $n>m$. For a submartingale, $E[X_n|{\mathcal{F}}_m]\ge X_m$ for all $n>m$.

Examples of Martingale

1. Sum of independent zero-mean RVs, $S_n$.
2. Products of non-negative independent RVs of mean 1, ${\prod }_{i=1}^n{X}_i$.
3. Accumulating data about a random variable, $E[X|{\mathcal{F}}_n]$.

Previsible Process

A process $C_n$ is previsible if $C_n\in {\mathcal{F}}_{n-1}$ for all n.

Martingale Transformation

$(C\cdot X{)}_n={\sum }_{k=1}^n{C}_k(X_k-X_{k-1})$ is the martingale transform of $X$ by $C$.

You Can’t Beat the System!

If $C_n$ are non-negative and bounded, $X$ is the supermartingale, then $(C\cdot X{)}_n$ is a supermartingale.

Stopping Time

For countable state space, $T$ is a stopping time IFF { T = n } ∈ F n , ∀ n \{T=n\}\in\mathcal{F}_n, \forall n {T=n}∈Fn​,∀n. That is to say, $T$ is a time when you can decide to stop the game or not after the $n$-th game based on the history up to time n.

A Proposition

If $N$ is a stopping time, $X_n$ is a supermartingale, then $X_{n\wedge N}$ is a supermartingale.

Martingale Convergence Theorem

If $X_n$ is a submartingale with ${\sup }_{n}E{X}_n^{+}<\infty$, $X_n$ converges a.s. to a limit $X$ with $E|X|<\infty$. A followed collary is:
If $X_n\ge 0$ is a supermartingale, then $X_n\to X$ a.s. and $EX\le E{X}_0$.

Doob’s Decomposition

Any submartingale can be written in a unique way as $X_n=M_n+A_n$ where $M_n$ is a martingale and $A_n$ is a predictable increasing sequence with $A_0=0$. In fact, $A_n={\sum }_{k=1}^{n}E[X_k-X_{k-1}|{\mathcal{F}}_{k-1}]$

$L_p$ Convergence Theorem

If $X_n$ is a martingale with $\sup E|X_n{|}^p<\infty$ where $p>1$, then $X_n$ converges a.s. and in $L_p$.

Uniform Integrability

A collection of { X i , i ∈ I } \{X_i,i\in I\} {Xi​,i∈I} is said to be uniformly integrable IFF
$$\underset{M\to \infty }{\lim }\underset{i\in I}{\sup }{\int }_{|X_i|>M}|X_i|=0.$$

An Example

Given a probability space $(\Omega ,{\mathcal{F}}_0,P)$ and $X\in L_1$, { E [ X ∣ F ] : F  is a  σ -field ⊂ F 0 } \{E[X|\mathcal{F}]:\mathcal{F} \text{ is a } \sigma\text{-field}\subset\mathcal{F}_0\} {E[X∣F]:F is a σ-field⊂F0​} is uniformly integrable.

Equivalence of Uniform Integrability I

If $X_n\to X$ in prob. and $E|X_n|<\infty$ for all n, then the following are equivalent:

1. { X n } \{X_n\} {Xn​} is uniformly integrable.
2. $X_n\to X$ in $L_1$.
3. $E|X_n|\to E|X|<\infty$.

Equivalence of Uniform Integrability II

For a submartingale, the following are equivalent:

1. uniform integrability.
2. a.s. convergence and convergence in $L_1$.
3. convergence in $L_1$.

A Lemma

If a martingale $X_n$ converges to $X$ in $L_1$, then $X_n=E[X|{\mathcal{F}}_n]$.

Equivalence of Uniform Integrability III

For a martingale, the following are equivalent:

1. uniform integrability.
2. a.s. convergence and convergence in $L_1$.
3. convergence in $L_1$.
4. There is an integrable random variable $X$ such that $X_n=E[X|{\mathcal{F}}_n]$.

The Limit of $E[X|{\mathcal{F}}_n]$

If ${\mathcal{F}}_n$ is an increasing sequence of sigma-field with limit F ∞ = σ { ∪ n F n } \mathcal{F}_\infty=\sigma\{\cup_n\mathcal{F}_n\} F∞​=σ{∪n​Fn​}, then
$$E[X|{\mathcal{F}}_n]\to E[X|{\mathcal{F}}_{\infty }]$$ a.s. and in $L_1$.

Levy’s 0-1 Law

If ${\mathcal{F}}_n\uparrow {\mathcal{F}}_{\infty }$ and $A\in {\mathcal{F}}_{\infty }$, then $E[1_A|{\mathcal{F}}_n]\to 1_A$ a.s…

A Theorem

If $X_n$ is a submartingale and $N$ is a stopping time with $pr(N\le k)=1$, then $E{X}_0\le E{X}_N\le E{X}_k$.
For a uniformly integrable submartingale $X_n$ and any stopping time $N$ we have $E{X}_0\le E{X}_N\le E{X}_{\infty }$

Reference

Durrett, Rick. Probability: theory and examples. Vol. 49. Cambridge university press, 2019.
S.R.S.Varadhan, Probability Theory (Courant Lecture Notes), 2000