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.

Let $W_t$ be a standard one dimension Brownian Motion with $W_0=0$ and $X_t=\int_0^t{W_sds}$. With the help of ito formula, we could get $$E[(X_t)^2]=\frac{1}{3}t^3$$ $$E[(X_t)^3]=0$$

When I try to employ the same method to calculate the general case $E[(X_t)^n]$, I got stuck. I guess $X_t$ should be normal distribution since it could be the limit of the following $$\lim_{n\rightarrow \infty}{\sum_{i=0}^{n-1}{W_{t_i}(t_{i+1}-t_i)}},$$

where $ W_{t_i}\sim norm(0,\sqrt{\frac{t_i}{n}}).$

If it is true, the problem would be trivial.

Update: Thanks for all the suggestions. Now I believe $X_t$ is a Gaussian process.

How about for this integral $$Y_t=\int_0^t{f(W_s)ds}$$ if we assume that $f$ is some good function, say polynomial or exponential, i.e $$Y_t=\int_0^t{e^{W_s}ds}$$ $$Y_t=\int_0^t{[a_n(W_s)^n+a_{n-1}(W_s)^{n-1}+...+a_0]ds}$$

share|improve this question
    
Re your update: are you really asking whether $Y_t$ is Gaussian? Wow. –  Did Mar 19 '11 at 0:39
    
Did you get something out of one of the answers below? –  Did Apr 7 '11 at 8:11

3 Answers 3

As a linear functional of the Gaussian process $(W_s)_{0\le s\le t}$, $X_t$ is a Gaussian random variable. You indicate yourself that $X_t$ is centered and has variance $\sigma^2_t=\frac13t^3$, hence $X_t$ is distributed like $\sigma_tN$ with $N$ standard Gaussian.

Thus, for every $n$, $E((X_t)^{2n+1})=0$ and $E((X_t)^{2n})=(\sigma_t^2)^nE(N^{2n})$. If you know the moments of a standard Gaussian (and you should...), you are done.

share|improve this answer
    
@Didier Piau: $X_t$ is the limit of a serial of Gaussian random variables. We need to show that the limit is Gaussian. –  Jun Deng Mar 18 '11 at 20:43
    
@Jun Deng Each $X_t$ is the mean square limit of some linear combinations of $(W_s)_{0\le s\le t}$. The Gaussian property is preserved under linear combinations and mean square limits, thus $X_t$ is a Gaussian random variable. And in fact the process $(X_t)_t$ itself is Gaussian. –  Did Mar 18 '11 at 21:04
    
@Didier Piau: Thanks. I should study limit theory later.^_^ –  Jun Deng Mar 18 '11 at 21:19
    
@Didier: Do you do stuff with stochastic analysis? –  Jonas Teuwen Mar 18 '11 at 22:22
    
It is possible to check all this (moment + gaussian nature of $X_t$) by studying the properties of the Fourrier transform of $X_t$ (or its Laplace transform if you like it better). Regards. –  TheBridge Mar 18 '11 at 22:34

The random variable $X_t$ is Gaussian for the reasons in Didier's answer.

You can calculate the variance directly (without Ito's formula) as follows:

$$\mathbb{E}\left[\left( \int^t_0 W_s ds \right)^2\right] = \int^t_0 \int^t_0 \mathbb{E}(W_r W_s) dr ds = \int^t_0 \int^t_0 (r\wedge s) dr ds ={t^3\over 3}.$$

share|improve this answer
    
Thanks. Your approach is much easier and easy to extend to high dimension. –  Jun Deng Mar 18 '11 at 21:25
    
@Jun Deng You're welcome. I'm not sure how easy it is to extend this approach to higher moments. In order to work out the fourth moment, I needed to first deduce the formula $$\mathbb{E}(W_{t_1}W_{t_2}W_{t_3}W_{t_4})=t_{(1)} (2t_{(2)}+t_{(3)}),$$ where the $t_{(i)}$s are the order statistics of $t_i$. It works, but it's not that easy, and I'm not sure what the general pattern is. –  Byron Schmuland Mar 18 '11 at 22:12
    
Let say for triple case, $$1=\sum_{\sigma(t1,t2,t3)}I_{t_1\leq t_2\leq t_3},$$ where $\sigma(t1,t2,t3)$ means all the permutation of $t_1,t_2,t_3$. Hence, we could get the integral by symmetry. Am I right? –  Jun Deng Mar 18 '11 at 22:29
    
Yes. You are right. It can be done; but it is not obvious for the general formula. I will try again for the general case. –  Jun Deng Mar 18 '11 at 22:48
    
By the discussion before, we can suppose that $t_1<t_2<...<t_n.$ put $\xi_1=W_{t_1}, \xi_2=W_{t_2}-W_{t_1},...,\xi_n=W_{t_n}-W_{t_{n-1}}$, which are independent Gaussian variables. Then, we have $$E[W_{t_1}W_{t_2}...W_{t_n}]=E[\xi_1(\xi_1+\xi_2)...(\xi_1+\xi_2+...+\xi_n)],$$ Which could be done. But I don't know the explicit formula. –  Jun Deng Mar 19 '11 at 5:05

Note that $X_T = \int_0^T W_t\,dt = T W_T - \int_0^T t\,dW_t$. Thus $X_T$ is a linear combination of normally distributed random variables. (The latter integral may be easier to show that it is normal.)

share|improve this answer

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.