Convergence of a conditional expectation

837 Views Asked by At

Consider a filtered probability space ($\Omega,\mathcal{F}_t,\mathcal{F},\mathbb{P}$) and a random variable $X$ defined on $\Omega$ with values in a set $E$.

We consider the process $X_t = \mathbb{E}[ X | \mathcal{F}_t]$ and the question is the following:

What conditions should be put on $\mathcal{F_t}$ so that $X_t$ has continuous paths $\mathbb{P}$-a.s. That is to say

$X_t = \lim_{s\rightarrow t} X_s$ a.s

Thank you!

1

There are 1 best solutions below

8
On BEST ANSWER

Obviously, $(X_t,\mathcal{F}_t)_{t \geq 0}$ is a uniformly integrable martingale. Fix $t>0$. By the martingale convergence theorem, there exists $Y \in L^1$ such that

$$X_s = \mathbb{E}(X \mid \mathcal{F}_s) \to Y, \qquad s \uparrow t$$

in $L^1$ and almost surely. Using the tower property, we find

$$\mathbb{E}(Y \mid \mathcal{F}_s) = X_s = \mathbb{E}(X_t \mid \mathcal{F}_s), \qquad s<t$$

and therefore

$$\int_F Y \, d\mathbb{P} = \int_F X_t \, d\mathbb{P}$$

for any $F \in \mathcal{F}_s$. This implies

$$\int_F Y \, d\mathbb{P} = \int_F X_t \, d\mathbb{P}$$

for any $F \in \sigma(F_s; s <t)=: \mathcal{F}_{t-}$. As $Y$ is $\mathcal{F}_{t-}$-measurable, we conclude

$$Y = \mathbb{E}(X_t \mid \mathcal{F}_{t-})$$

This yields the (necessary and sufficient) condition for left-continuity of the process $(X_t)_{t \geq 0}$:

$$X_t \stackrel{!}{=} Y=\mathbb{E}(X_t \mid \mathcal{F}_{t-}) = \mathbb{E}(X \mid \mathcal{F}_{t-}) \tag{1}$$

A similar reasoning shows that

$$X_t \stackrel{!}{=} \mathbb{E}(X \mid \mathcal{F}_{t+}) \tag{2}$$

with $\mathcal{F}_{t+} := \bigcap_{s>t} \mathcal{F}_s$ ensures the right-continuity of $(X_t)_{t \geq 0}$.

Consequently, we conclude that $(X_t)_{t \geq 0}$ is (a.s.) continuous if and only if $(1)$ and $(2)$ are satisfied. This is in particular the case if

$$\mathcal{F}_{t-} = \mathcal{F}_t = \mathcal{F}_{t+}$$

for any $t \geq 0$.