Let $B_t$ be a brownian motion, $\delta \in \mathbb{R}$. I have to prove if $e^{\delta B_t - \frac{\delta^2 t}{2}}$ is also a brownian motion.

I can't not use any Ito's stuff because it is not part of the course this problem comes from. But I can use any other stuff from martingales, Levy characterization theorem, and brownian motion properties and theorems.

This is what I've tried:

A brownian motion must have independent stationary increments with normal distribution, and continuous paths. And it must start at a value in $\mathbb{R}$. The function $e^{x+y}$ is a continuous Borel-Medible so, if $B_t - B_s$ has normal distribution with continous paths, the process $e^{\delta (B_t - B_s) - \frac{\delta^2 (t-s)}{2}}$ is also normally distributed with continous paths. It starts at value 1.

The problem comes when proving independence. If I choose carefuly the values of the intervals $[t,s]$ I can have the intervals of the function $e^{\delta (B_t - B_s) - \frac{\delta^2 (t-s)}{2}}$ overlapping, then the resulting value will be a function of values of different intervals (no more independence!).

Then I tried the Levy characterization theorem:

The process $e^{\delta B_t - \frac{\delta^2 t}{2}}$ has continous paths and it is a martingale (for the proof see of the book Brzezniak Z., Zastawniak T. - Basic Stochastic Processes, A Course Through Exercises, Springer, 2002. Solution to exersice 6.35)

Meh, I blatantly copy/pasted the proof for you to enjoy:

enter image description here

This time my problem comes from proving that $e^{2*\delta B_t - \delta^2 t}-t$ is a martingale. If I make $\alpha = 2* \delta$ and repeat the solution 6.35 I have no way to get rid of the $-t$ term.

As long as I can prove here, the process $e^{\delta B_t - \frac{\delta^2 t}{2}}$is a martingale but it is not a brownian motion. Please, tell me if I did something wrong.

Edit: I fix a typo because I wrote plus sign in the exponential instead of a minus sign. I'm almost sure (not pun intended) the exponential martingale is not a brownian motion, I was getting confused with the geometric brownian motion, but they are different processes.


it's clearly not a Brownian motion. $$ P(B_1 <0 ) =0.5. $$ $$ P(e^{\delta B_t - \frac{\delta^2 t}{2}} < 0 )=0. $$ It is a martingale: for $s < t$ $$ e^{\delta B_t - \frac{\delta^2 t}{2}}=e^{\delta B_s - \frac{\delta^2 s}{2}}e^{\delta (B_t-B_s) - \frac{\delta^2 (t-s)}{2}} $$ Taking a conditional expectation on time $s$ the second factor has expectation $1$ and the result follows.

  • $\begingroup$ You're right! I had a typo, I'm fixing it! $\endgroup$ – Broken_Window Dec 2 '16 at 22:20
  • $\begingroup$ Why does the second factor have expectation 1? $E[B_{t} - B_{s}] = 0$ and $t - s$ is deterministic so stays as is. Or can you not simply take the expectation inside the exponential? Even if you use Jensen's inequality I don't see how it reduces to 1. $\endgroup$ – Omar Haque Jan 29 '18 at 20:38

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.