4
$\begingroup$

Online I found the asymptotic behavior property of geometric Brownian Motion $X_t$as:

If $\mu$ (drift parameter) is $\ge$ $\sigma^2/2$ where $\sigma$ is the volatility parameter, then $X_t \rightarrow \infty$ as $t \rightarrow \infty$

If $\mu < \sigma^2/2$, then $X_t \rightarrow 0$ as $t \rightarrow \infty$

If $\mu = \sigma^2/2$, then $X_t$ has no limit as $t \rightarrow \infty$

While this makes sense, how would the proof look like for this property? I'm not really sure how to approach it at the moment. Any help is appreciated.

$\endgroup$

2 Answers 2

5
$\begingroup$

For any $\alpha > 0$, \begin{align*} \lim_{t\rightarrow\infty}P\left(e^{\big(u-\frac{\sigma^2}{2}\big) t +\sigma W_t} > \alpha \right) &= \lim_{t\rightarrow\infty}P\left(\big(u-\frac{\sigma^2}{2}\big) t +\sigma W_t > \ln \alpha \right)\\ &=\lim_{t\rightarrow\infty}P\left(\frac{W_t}{\sqrt{t}} > \frac{\ln\alpha- \big(u-\frac{\sigma^2}{2}\big) t }{\sigma \sqrt{t}} \right)\\ &=\lim_{t\rightarrow\infty}\Phi\left(\frac{\big(u-\frac{\sigma^2}{2}\big) t -\ln\alpha}{\sigma \sqrt{t}} \right)\\ &= \begin{cases} 1, &\mbox{ if } u>\frac{\sigma^2}{2},\\ \frac{1}{2}, &\mbox{ if } u=\frac{\sigma^2}{2},\\ 0, &\mbox{ if } u < \frac{\sigma^2}{2}. \end{cases} \end{align*} The conclusion now follows immediately.

Edit based on comments below.

Let $X_t = e^{\big(u-\frac{\sigma^2}{2}\big) t +\sigma W_t}$. Note that \begin{align*} \left(\omega:\, \lim_{t\rightarrow \infty} X_t = \infty \right) &= \cap_{m=1}^{\infty}\cup_{n=1}^{\infty} \cap_{t\ge n}(\omega:\,X_t > m). \end{align*} The almost surely convergence, for the case where $u>\frac{\sigma^2}{2}$, follows immediately.

$\endgroup$
2
  • $\begingroup$ This answer was flagged with the following comment from @antonzm 'It seems that what is proven here is different from the statement that is required. The general statement is, for instance, that 𝑋(𝑑)β†’βˆž a.s. as π‘‘β†’βˆž when 𝑒>𝜎22, however, what is shown in here is that 𝑋(𝑑)β†’βˆž in probability as π‘‘β†’βˆž. To show the former result, an argument with the law of iterated expectations must be done to argue that the 𝑑 term dominates the π‘Šπ‘‘ term in the limit.' I have copied the flag contents here to allow Gordon chance of rebuttal or amendment. $\endgroup$
    – Attack68
    Dec 4, 2019 at 17:39
  • $\begingroup$ @Attack68: I will have a look. In fact, antonzm can actually make comments here directly. $\endgroup$
    – Gordon
    Dec 4, 2019 at 18:25
4
$\begingroup$

Write $X_t = \exp(\mu B_t + (\mu - \frac {\sigma^2}{2})t )$. To prove the statement you can use the law of large numbers for Brownian motion which states that $\lim_{t \to \infty} \frac {B_t}{t} = 0$. Then rewrite $X_t$ as $$X_t = \exp(t (\mu \frac {B_t}{t} + (\mu - \frac{\sigma^2}{2})).$$ Using these two properties, you can analyze the convergence behaviour of $X_t$.

$\endgroup$

Your Answer

By clicking β€œPost Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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