0
$\begingroup$

Let $(W_t)_{t \geq 0}$ be a Brownian motion and let $\tau=$inf{$t\ge0:W_t=a+bt$} with $a<0$.

I want to compute $E[\tau]$ (and I already know that $E(\tau)<\infty$).

I want to say that I already read Expectation of the first passage time of $T_{a,b}$ [duplicate]

and Distribution of hitting time of line by Brownian motion

The reason I am not satisfied with the other posts is, that computing the distribution seems like an overkill to me (the exercise is from an exam and thus I think computing the distribution first is too time consuming). Furthermore was the solution only valid for $a>0$, which does not hold for my case. Additionally does the suggested solution from the first post: $E[\tau]=\frac{a}{|b|}e^{-ab}$ not hold for $a<0$ since this would imply a negative expectation for a strictly positive random variable.

$\endgroup$
  • $\begingroup$ Using martingale techniques it is not too difficult to compute $$f(\lambda) := \mathbb{E}e^{-\lambda \tau}, \qquad \lambda>0$$ see this question: math.stackexchange.com/q/2107249/36150. Once you have computed $f$, you can use that $$\mathbb{E}(\tau) = - \frac{d}{d\lambda} f(\lambda) \big|_{\lambda=0}.$$ $\endgroup$ – saz Jan 31 '18 at 11:37
  • $\begingroup$ When I am using that approach I am getting a negative expectation: $E[e^{-\lambda \tau}]=e^{-a(b+\sqrt{b^2+2\lambda})}$ with $\lambda=-a\alpha+1/2\alpha^2>0.$ Thus $E[\tau]=\frac{a}{\sqrt{b^2}}e^{-ab}<0.$ $\endgroup$ – Andrew Jan 31 '18 at 12:20
  • $\begingroup$ Or is it than we choose $\sqrt{b^2}=-b$ (if b is positive) such that the expectation is bigger than 0. Than the sign of $a$ still has a large impact on how $e^{-ab}$ explodes which seems a bit wierd to me but I guess it is fine. $\endgroup$ – Andrew Jan 31 '18 at 12:27
  • $\begingroup$ What do you mean by "choose "$\sqrt{b^2} = -b$ (if b is positive)"... Obviously, $\sqrt{b^2} = -b$ is plainly wrong if $b>0$. If you follow the approach, you get the correct result, see the answer below. $\endgroup$ – saz Jan 31 '18 at 20:17
1
$\begingroup$

If $b \leq 0$ then $$\tau \geq \inf\{t \geq 0; W_t = a\} =: T_a$$ implying $\mathbb{E}(\tau) \geq \mathbb{E}(T_a)=\infty$. Therefore we can assume from now on that $b>0$. If we set

$$X_t := W_t-bt$$

then

$$\tau = \inf\{t \geq 0; X_t=a\}.$$

Using that

$$M_t := \exp \left( - \alpha X_t - \alpha bt - \frac{1}{2} \alpha^2 t \right)$$

is a martingale for any $\alpha>0$ we find from the optional stopping theorem that

$$\mathbb{E}(M_{t \wedge \tau}) = \mathbb{E}(M_0)=1.$$

As $|M_{t \wedge \tau}| \leq e^{\alpha |a|}$ it follows from the dominated convergence theorem that $\mathbb{E}(M_{\tau})=1$, and so

$$\mathbb{E} \exp \left( - \left[ \alpha b + \frac{1}{2} \alpha^2 \right] \tau \right) = e^{\alpha a}.$$

Thus,

$$f(\lambda) := \mathbb{E}e^{-\lambda \tau} = \exp \left(- a \left[ b- \sqrt{b^2+2\lambda} \right] \right), \qquad \lambda>0.$$

Differentiating the expression with respect to $\lambda$ we find

$$\mathbb{E}(\tau) = - \frac{d}{d\lambda} f(\lambda) \bigg|_{\lambda =0} = \frac{-a}{b}.$$

$\endgroup$
  • $\begingroup$ I think what confused me is that you need two different approaches depending if $a>0$ or $a<0$, in particular for $a>0$ you use a "classic" GBM and for $a<0$ you use that $-W_t$ is a Brownian Motion as well. I understand how you derive the solution, but I still have two problems: 1) Why is your solution wrong for a>0? (In particular I don't see where we are using that a<0) 2) Consider the two stopping times $\tau_1=$inf{$t\ge0:W_t=t-1$}, $\tau_2=$inf{$t\ge0:W_t=10t-1$}. It holds that $\tau_2\le \tau_1.$ But from your result we have that $E[\tau_1]=e<E[\tau_2]=\frac{1}{10}e^{10}$ $\endgroup$ – Andrew Feb 1 '18 at 9:46
  • $\begingroup$ I got it. I think your solution is not correct in the last step. After differentiating and setting $\lambda=0$, we get $e^{-ab+a\sqrt{b^2}}=1$ , in particular $E[\tau]=-\frac{a}{b},$ if $a<0$. $\endgroup$ – Andrew Feb 1 '18 at 9:53
  • $\begingroup$ @Andrew 1) We are using $a<0$ to conclude that $X_{t \wedge \tau} \geq a$, and so $|M_{t \wedge \tau}| \leq e^{\alpha |a|}$. 2) Yeah, you are right about that, the $e^{-ab}$ cancels... $\endgroup$ – saz Feb 1 '18 at 10:57

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.