What is wrong with the following logic:

let $0\leqslant s \leqslant t \leqslant u$, find $E[W_t | W_s, W_u]$

\begin{align*} E[W_t | W_s, W_u] &= E\left.\left[W_t - \frac{t}{u} W_u + \frac{t}{u}W_u\right|W_s,W_u\right]\\ &=E\left.\left[W_t - \frac{t}{u} W_u \right|W_s\right] + \frac{t}{u}W_u\\ &=E\left[W_t|W_s\right] - \frac{t}{u} E\left[W_u|W_s\right] + \frac{t}{u} W_u\\ &=W_s+\frac{t}{u}(W_u-W_s). \end{align*}

I know this is is the wrong answer, but I am struggling to identify which part of this argument is incorrect?

  • $\begingroup$ why did the conditioning on $W_u$ suddenly disappear in the second equality? Also why are you multiplying by $t/u$? $\endgroup$ – Alex R. Mar 28 '15 at 4:11
  • $\begingroup$ @AlexR. I am using the fact that $(W_t - \frac{t}{u} W_u) \bot W_u$, so I remove it from the conditioning in the second equality $\endgroup$ – dimebucker Mar 28 '15 at 4:14

@AlexR. is correct: Your argumentation fails in the second line. You claim that

$$\mathbb{E} \left( W_t - \frac{t}{u} W_u \mid W_s, W_u \right) = \mathbb{E} \left( W_t - \frac{t}{u} W_u \mid W_s \right)$$

because $(W_t- \frac{t}{u})$ and $W_u$ are independent. This equality does not hold because $W_u$ and $W_s$ are not independent. The following statement is correct

Let $X,Y,Z$ be random variables such that $X,Y$ are jointly independent from $Z$. Then $$\mathbb{E}(f(X) \mid Y,Z) = \mathbb{E}(f(X) \mid Z).$$

whereas this statement is (in general) not correct:

Let $X,Y,Z$ be random variables such that $X$ and $Y$ are independent. Then $$\mathbb{E}(f(X) \mid Y,Z) = \mathbb{E}(f(X) \mid Z).$$

  • $\begingroup$ How can I prove this, if for example I want to see what the covariance here, how does one go about solving: $$ Cov ( W_t - \frac{t}{u} W_u , (W_s, W_u) ) $$ $\endgroup$ – dimebucker Mar 28 '15 at 8:44
  • $\begingroup$ @dimebucker91 What do you mean by an expression of the form $\text{cov}\, (X, (Y,Z))$ for random variables $X,Y,Z$...? $\endgroup$ – saz Mar 28 '15 at 8:57
  • $\begingroup$ I guess it doesn't make sense. Im just trying to solve this expectation by using the 'trick' of adding and subtracting an increment. For example if I instead started off with: \begin{align*} E(W_t | W_u,W_s) & = E(W_t - W_s + W_s | W_u,W_s)\\ &=E( W_t-W_s| W_u,W_s) +W_s\\ &=E( (W_t-W_s) -\frac{t-s}{u-s}(W_u-W_s) + \frac{t-s}{u-s}(W_u-W_s)| W_u,W_s) +W_s\\ &=E( (W_t-W_s) -\frac{t-s}{u-s}(W_u-W_s) | W_u,W_s)+ \frac{t-s}{u-s}(W_u-W_s) +W_s\\ \end{align*} The part outside of the expectation is the answer given in the book. That means the expectation is zero, but im not sure how to show this.... $\endgroup$ – dimebucker Mar 28 '15 at 9:12
  • $\begingroup$ @dimebucker91 I would calculate the conditional expectation using the joint density of $(W_s,W_t,W_u)$. However, right now, I'm not sure how to do this using this increment-trick. Might be better to ask a new question. $\endgroup$ – saz Mar 28 '15 at 10:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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