Let $X_t$ be a Geometric Brownian Motion (GBM) with initial state $x_0$ at $t=0$ and dynamics:

$$ dX_t = \mu X_tdt + \sigma X_tdW_t $$

Let us define the GBM hitting time of level $x$:

$$ \begin{align} \tau & = \min\,\,\{t \geq 0: X_t=x\} \\[6pt] & = \min\left\{ t \geq 0:W_t+\left(\mu-\frac{\sigma^2}{2}\right)\frac{t}{\sigma}=\frac{1}{\sigma}\log\frac{x}{x_0}\right\} \end{align}$$

Under the following conditions:

$$ \begin{align} & (\text{A}) \quad \sigma > 0 \\[12pt] & (\text{B}) \quad \lambda > 0 \\[12pt] & (\text{C}) \quad x > x_0 \\[6pt] & (\text{D}) \quad \lambda\left(\theta+\frac{\lambda}{2}\right) >0 \end{align} $$

Do we have:

$$ \mathbb{P}\left(\tau < \infty\right)=1 \text{ ?}$$

By letting $\theta=(\mu-\sigma^2/2)/\sigma$ and $\hat{W}_t=W_t+\theta t$, I have defined the martingale $M_t$:

$$ M_t=e^{\lambda\hat{W}_t-\lambda\theta t-\frac{\lambda^2}{2}t} $$

Then by studying the limiting behaviour of the stopped (martingale) process $M_{\min(t,\tau)}$ for each case $\{\tau<\infty\}$ and $\{\tau=\infty\}$, noticing that $\mathbb{E}[M_{\min(t,\tau)}]=\mathbb{E}[M_0]=1$, and applying the dominated convergence theorem, I have concluded that:

$$ \mathbb{E}\underbrace{\left[\mathbf{1}_{\{\tau \,<\, \infty\}}e^{\lambda\hat{W}_{\tau}-\lambda\theta\tau-\frac{\lambda^2}{2}\tau}\right]}_{M_{\infty}}=1$$

Now, from $(\text{A})$, $(\text{C})$ and $(\text{D})$, I would be tempted to make $M_{\infty}$ dependent on $\lambda$, then showing that $M_{\infty}(\lambda)$ is bounded from above by $e^{(\lambda/\sigma) \log(x/x_0)}>1$ and converges to $\mathbf{1}_{\{\tau \, < \, \infty\}}$ to apply the dominated convergence theorem again to conclude that:

$$ \lim_{\lambda \rightarrow \infty}\mathbb{E}[M_{\infty}(\lambda)] = \mathbb{E}\left[\lim_{\lambda \rightarrow \infty}M_{\infty}(\lambda)\right] = \mathbb{P}(\tau < \infty) =1 $$

Would this be correct?

  • $\begingroup$ How is $\lambda$ to be chosen? $\endgroup$ – John Dawkins Jul 2 '17 at 22:06
  • $\begingroup$ @JohnDawkins just a positive real number. $\endgroup$ – Morris Fletcher Jul 4 '17 at 15:17

If you should take $\lambda\to0$ instead of $\lambda\to\infty$, your method works for $\theta\ge0$ without need for the LIL. Indeed, if $\theta\ge0$, then $M_\infty(\lambda)\le\exp(\lambda\hat W_\tau)\le\exp(\hat W_\tau)=(x/x_0)^{1/\sigma}$ for all $0<\lambda<1$, so the result follows by letting $\lambda\to0$ by the bounded convergence theorem. To prove that $\mathbb P(\tau<\infty)<1$ for $\theta<0$, you probably need to appeal to the LIL.

  • $\begingroup$ Thank you for the answer @Janson. Just a couple of questions: do we need $\theta \geq 0$, or is condition $(\text{D})$ enough? The 2nd one might be silly, but I thought the dominated convergence theorem worked when the convergence is towards $\infty$; how do we justify the applicability of the theorem when the convergence is towards $0$? $\endgroup$ – Morris Fletcher Jul 8 '17 at 10:55
  • $\begingroup$ You need $\theta\ge0$ - John Dawkins' answer demonstrates necessity. Condition (D) on its own doesn't make a lot of sense. Do you mean for all $\lambda>0$? Then it is equivalent to $\theta\ge0$. For some $\lambda>0$? Then this is true for any $\theta$. Taking $\lambda\to\infty$ doesn't work. Your dominating function depends on $\lambda$, and actually tends to $+\infty$ as $\lambda\to\infty$. The dominated convergence theorem works just fine if you let $\lambda\to0$ instead, as my answer shows. $\endgroup$ – Jason Jul 8 '17 at 16:45

Keep in mind that for large $t$, $W_t$ will frequently get as large as $\sqrt{bt\log\log t}$ (for each $0<b<2$) but not as large as $\sqrt{bt\log\log t}$, infinitely often, if $b>2$, by the law of the iterated logarithm. Therefore under your conditions (A) and (C), $W_t$ will hit a line with slope $(\mu-\sigma^2/2)/\sigma$ and intercept $\log(x/x_0)/\sigma$ (i) with probability 1 if $\mu-\sigma^2/2\le 0$, but (ii) with probability less than 1 if $\mu-\sigma^2/2>0$.

  • $\begingroup$ Thank you @JohnDawkins. However I am not sure I follow you, I am not familiar with the LIL. Could you be a bit more explicit please? $\endgroup$ – Morris Fletcher Jul 3 '17 at 18:48
  • 1
    $\begingroup$ The LIL states that, with probability 1, $\limsup_{t\to\infty}{W_t\over\sqrt{2t\log\log t}}=1$. For your purposes, the takeaway is that the Brownian motion will hit a line of positive slope. with intercept greater than $W_0$, only with probability less than $1$. $\endgroup$ – John Dawkins Jul 4 '17 at 16:29
  • $\begingroup$ I see thank you @JohnDawkins. Thus do you see where does my original argument go wrong? $\endgroup$ – Morris Fletcher Jul 5 '17 at 9:38
  • $\begingroup$ Your evaluation of the limit $\lim_{\lambda\to\infty}M_\infty(\lambda)$ is in doubt; should it not br $0$? $\endgroup$ – John Dawkins Jul 5 '17 at 16:24
  • 1
    $\begingroup$ You have your signs the wrong way around - $\tau$ is the first hitting time of the line $-\frac1\sigma(\mu-\sigma^2/2)t+\log(x/x_0)/\sigma$. $\endgroup$ – Jason Jul 7 '17 at 21:41

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.