Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

I know that if we have $E[\int_0^1 |X_t|dt] < \infty$ we may apply Fubini's theorem and compute $E[\int_0^1 X_tdt] = \int_0^1 E[X_t]dt$. Is there a similar version that allows the exchange of integral and conditional expectation? I want to say $E[\int_0^1 X_tdt|F_s] = \int_0^1 E[X_t|F_s]dt$ but I don't know why it should be true.

share|improve this question
3  
Take $F\in\mathcal F_s$ and compute $E[\chi_F\int_0^1X_tdt]$ with Fubini's theorem. –  Davide Giraudo Feb 16 '13 at 19:52

3 Answers 3

up vote 2 down vote accepted
+100

Let $(\Omega,\cal A,\mu)$ be a probability space, and assume that $X\colon [0,1]\times\Omega\to \Bbb R$ is measurable for the product $\sigma$-algebra and is integrable with respect to the product measure. In particular, we don't need the assumption of path-continuity.

We can approximate $X$ in $L^1(\lambda \otimes\mu)$ and almost everywhere by simple function, i.e. finite linear combinations of elements of the form $\chi_A$, where $A\in \mathcal{A}\otimes \mathcal{B}([0,1])$. As such an element can be approximated in $L^1(\lambda\otimes\mu)$ by a finite linear combination of elements of the form $\chi_A\chi_B$, $A\in\cal A$, $B\in\mathcal B[0,1]$, we can consider linear combinations $\sum_kc_k\chi_{A_k}\chi_{B_k}$. We will denote $\{X_n\}$ this approximating sequence, which can be chosen in order to make $|X_n|\leqslant |X|$ almost everywhere.

Using dominated convergence theorem for conditional expectation, we just have see prove the equality for $X_n$ instead of $X$. We can indeed take $E(|X(t,\cdot)| \mid\mathcal F)$ as a dominating function, since $|X_n|\leqslant X$.

Note that by dominated convergence, $$\int_0^1E[X(t,\cdot)\mid\mathcal F]dt=\lim_{n\to +\infty}\int_0^1E[X_n(t,\cdot)\mid\mathcal F]dt,$$ and $\int_0^1E[X_n(t,\cdot)\mid\mathcal F]dt$ is $\mathcal F$-measurable.

By linearity of conditional expectation, we will be done once we show that for any $A\in\cal A,$ $B\in\mathcal B[0,1]$, we have $$E\left[\int_0^1\chi_A(\cdot)\chi_B(t)dt\mid \mathcal F\right]=\int_0^1E\left[\chi_A(\cdot)\chi_B(t)\mid \mathcal F\right]dt.$$ It's the case since $\chi_B(t)$ doesn't depend on $\omega$.

share|improve this answer
    
When using Dominated Convergence Theorem, it's important to know the dominating function. In your proof, can you show that with probability one, –  Coiacy Feb 22 '13 at 4:27
    
When using Dominated Convergence Theorem, it's important to know the dominating function. In your proof, can you show that with probability one, $ |\mathbb{E}(X_n|\mathcal{F})|\le |\mathbb{E}(X|\mathcal{F})|$? And can you give more details on the $\mathcal{F}$-measurability of $\int_0^1\mathbb{E}(X_n|\mathcal{F})\mathrm{d}t$? Say why $\int_{[t_k,t_{k+1}]}\mathbb{E}(\chi_A |\mathcal{F})$ is $\mathcal{F}$-measurable? Thanks! –  Coiacy Feb 22 '13 at 4:41
    
We have $|E(X_n\mid\mathcal F)|\leqslant E(|X_n|\mid\mathcal F)\leqslant E(|X|\mid \mathcal F)$ as we took $|X_n|\leqslant X$ almost surely. –  Davide Giraudo Feb 22 '13 at 10:34
    
Another thing is that If $\{X_n\}$ is a sequence of $r.v.'s$ and $X_n \to X\ a.s.$, then $\mathbb{E}(X_n|\mathcal{F})\ may\ not\ converge\ to\ \mathbb{E}(X|\mathcal{F})$. There exists such counter examples. Well, this is just for random variables, and here you approximate a process, so I wonder whether it works. Maybe it's better to approximate $\mathbb{E}(X_n(t,\omega)|\mathcal{F})$ instead. –  Coiacy Feb 23 '13 at 23:24
    
@Coiacy The fact follows because there is convergence in L^1$ (and also almost everywhere). I agree that almost everywhere convergence alone wouldn't give the result. –  Davide Giraudo Feb 24 '13 at 10:17

So you are considering a continuous-time stochastic process $\{X_t: t\ge 0\}$. $\mathcal{F}_s$ looks like an element in a filtration $\{\mathcal{F}_t: t\ge 0\}$, and I suggest you put a general $\sigma$-field $\mathcal{F}$ instead. The following are some points we need pay attention to:

  1. First we need make sure $X_t$ is intergrable on $[\ 0,1\ ]$, say the path of $X$ is continuous a.s.
  2. Both $\mathbb{E}\left[\,\int_0^1 X_t\;\mathrm{d}t\,|\mathcal{F}\,\right]$ and $\int_0^1 \mathbb{E}[\,X_t|\mathcal{F}\,]\mathrm{d}t$ are random variables, and $\mathbb{E}[\,\int_0^1 X_t\;\mathrm{d}t\,|\mathcal{F}\,]$ is $\mathcal{F}-$measurable, can you show that $\int_0^1 \mathbb{E}[\,X_t|\mathcal{F}\,]\mathrm{d}t$ is also $\mathcal{F}-$measurable? (I am not sure)
  3. Of course we can check by definition of conditional expectation. I think it's true $$ \forall F\in \mathcal{F},\quad \mathbb{E}\left(\chi_F\cdot \int_0^1X_t\ \mathrm{d}t\right)=\mathbb{E}\left(\chi_F\cdot \int_0^1 \mathbb{E}(X_t|\mathcal{F})\ \mathrm{d}t\right)$$

In fact, $$ \mathbb{E}\left(\chi_F\cdot \int_0^1X_t\ \mathrm{d}t\right)=\mathbb{E}\left(\int_0^1 X_t\cdot \chi_F\ \mathrm{d}t\right)=\mathbb{E}\left(\int_0^1\mathbb{E}(X_t\cdot \chi_F|\mathcal{F}\right)\ \mathrm{d}t)=\mathbb{E}\left(\chi_F\cdot \int_0^1\mathbb{E}(X_t|\mathcal{F})\ \mathrm{d}t\right).$$

share|improve this answer
    
I'm stuck trying to show $\int_0^1 E[X_t|\mathcal{F}]$ is $\mathcal{F}$-measurable. –  nullUser Feb 18 '13 at 2:51

Here is a way to show $\int_0^1 \mathbb{E}[X_t|\mathcal{F}]\,\mathrm{d}t$ is $\mathcal{F}$-measurable:

define $I_n(\omega)=\sum_{k=1}^n \mathbb{E}[X_{\frac{k}{n}}|\mathcal{F}](\omega)\cdot \frac{1}{n}$, then $I_n$(sum of finitely many $\mathcal{F}$-measurable $r.v.$') is $\mathcal{F}$-measurable. If we know $\mathbb{E}[X_t|\mathcal{F}]$ is integrable on $[0,1]$, for example, $X_t$ is continuous $a.s.$ such as Brownian motion, then it's Riemann integrable and

$$ \int_0^1 \mathbb{E}[X_t|\mathcal{F}]\,\mathrm{d}t=:I=\lim_{n \rightarrow \infty}I_n \quad a.s.$$

Note this is a pathwise property.

Now since $I_n\in \mathcal{F}$, we obtain $I=\lim_{n \rightarrow \infty}I_n \in \mathcal{F}$.(because the limit of $\mathcal{F}$-measurable functions is still $\mathcal{F}$-measurable)

share|improve this answer
    
You are right, here I just assume it's continuous out of convenience. I think you know the fact every continuous-time martingale or submartingale $X_t$(which we usually take interest in) has a version that is right continuous with left limits(then its Lebesgue integral equals Riemann integral). But it still remains to be seen whether $\mathbb{E}(X_t|\mathcal{F})$ is continuous when $X_t$ is continuous. –  Coiacy Feb 22 '13 at 4:15

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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