# Expectation of time integral of Wiener process

I am trying to calculate $E(\int_0^T {W_s ds})$, where $W_s$ is a standard Brownian motion.

Now two approaches I can think of:

1) Take a partition of $[0,T]$. Calculate $E(\sum {W_{t_i}(t_{i+1} - t_i)})$ and take the limit as you shrink the size of the partition.

2) Calculate $\int_0^T { E(W_s)ds }$.

However, for approach 1), its not clear what function would dominate the absolute value of the terms inside the E() for all possible partitions, and that it would have a finite expectation. So, interchanging limit and expectation is dicey.

For approach 2), Fubini's theorem would require me to know a-priori that $E(\int_0^T {|W_s|ds})$ is finite, and I don't see how I could show that.

How can any of these approaches be fixed, if at all? Or is there another way to solve the problem?

-
Brownian motion is a martingale (en.wikipedia.org/wiki/Martingale_%28probability_theory%29); the expectation you want is always zero. Also voting to close as this would be better suited to another site mentioned in the FAQ. –  Steve Huntsman Oct 14 '10 at 3:28
If BM is a martingale, why should its time integral have zero mean ? (Although, yes, both approaches will give me an answer of 0). Also, the problems with the approaches I mentioned are valid for questions like calculating E(\Int_0_T {W_s^2 ds}) –  Cosmonut Oct 14 '10 at 3:56
If I have the distribution of Max{|W_s|:0<s<T}, I could try showing that its expectation if finite. This would let me validate both the approaches. If someone has a reference where this is calculated, please let me know. –  Cosmonut Oct 14 '10 at 4:04
Martingales are in $L^1$. –  Steve Huntsman Oct 14 '10 at 5:34
More appropriate for math.stackexchange.com –  zhoraster Oct 14 '10 at 6:34

For approach 2, Fubini's theorem works just as well if you show $$\int_0^T E|W_s|ds < \infty$$ which is easy. Indeed, perhaps even easier is to note $$\int_0^T E(|W_s|^2)ds = \int_0^T s ds = \frac{1}{2}T^2 < \infty$$ and use Jensen/Hölder/Cauchy-Schwarz.

-
This is the way to do it. Any invocation of martingales, running maximum, etc., is overkill for this problem. –  Tom LaGatta Oct 15 '10 at 1:03
Thanks, I like this. Looks like this approach can be generalized to work out the expectation of integrals of various functions of a Weiner process as well, such as exp(W_s). –  Cosmonut Oct 15 '10 at 4:19

It is possible to integrate by parts in $\int_0^T B(t) dt$ and obtain

$-B(t) (T-t)|_{t=0}^{t=T} + \int_0^T (T-t) dB(t) \overset{d}{=} \int_0^T (T-t) dB(t)$

The Wiener integral on the right has a normal distribution with mean $0$ and variance $\int_0^T (T-t)^2 dt = T^3/3$.

Edit: Sorry, I used $B$ instead of $W$ to denote Brownian motion.

-
This is really cool! I had worked out the variance of \Int_0_T {W_s ds} the hard way by partitioning [0, T] into n parts, calculating the variance of the sum and then letting n go to infinity. I got the correct answer, (thanks for verifying that :)) but was once again worried about limit interchange issues. Although, I'm guessing there must be some technical isses involved in justifying the integration by parts formula and equating the time integral to a stochastic integral. (I guess you are implicitly using Ito's lemma ?) –  Cosmonut Oct 15 '10 at 4:29
Slightly technical, but not bad: If h is a function with bounded variation on [a,b] and g is continuous on [a,b], then the Riemann-stieltjes integral \int_a^b g(x) dh(x) can be defined by approximating with step functions. Then, applying integration by parts, you can define the integral \int h(x) dg(x) as h(x)g(x)|_a^b - \int g(x) dh(x). Since W is almost surely continuous, and (T-t) has bounded variation, you apply this fact omega by omega to g(t) = W(t,\omega). See pages 12 and 13 in Kuo's book Introduction to Stochastic Integration for more detail. –  MarkV Oct 18 '10 at 15:22

Sorry if I ask a question in the answer section because my problem is related to the variance of the time integral of wiener process. I use the first approach of @cosmonut and come up with the variance equal limit of \sum {t_{t_i} (t_{i+1} - t_i)^2}+\sum\sum {min(t_t{t_i},t_t{t_j}) (t_{i+1} - t_i)(t_{j+1} - t_j)} and don't know how to reduce the sequence and get the result of T^3/3. Can anyone help me? Thanks.

-
Why don't you ask a real question if you look for an answer? People will not notice it here. By the way, where does your problem come from? –  András Bátkai Mar 29 '13 at 17:44