- BerndSchmitz
**Posts:**242**Joined:**

Hi,Can anybody help me with deriving the distribution of [$]max_{0 \leq s \leq t} W_s - W_t[$]?Thanks,Bernd

- LocalVolatility
**Posts:**128**Joined:****Location:**Amsterdam-
**Contact:**

Let [$]M_t = \max_{0 \leq u \leq t} W_u[$]. You know that [$]M_t[$] and [$]W_t[$] have the joint density[$]f_{M_t, W_t}(m, w) = \frac{2 (2m - w)}{t \sqrt{2 \pi t}} \exp \left\{ -\frac{(2m - w)^2}{2 t} \right\}[$]. Then I'd just take it from there..[$]\mathbb{P} \left\{ M_t - W_t \geq x \right\} = \int_0^\infty \int_{-\infty}^{m - x} f_{M_t, W_t}(m, w) \mathrm{d}w \mathrm{d}m[$]

- BerndSchmitz
**Posts:**242**Joined:**

Thanks. I had hoped for some elegant trick though

- BerndSchmitz
**Posts:**242**Joined:**

QuotePr( max(ws) - wt > a) = Pr( max(ws) > a) How do you justify that?

Last edited by BerndSchmitz on March 16th, 2015, 11:00 pm, edited 1 time in total.

QuoteOriginally posted by: BerndSchmitzThanks. I had hoped for some elegant trick though :-)One elegant trick would note that it is well-known that [$]M_t - W_t \stackrel{d}{=} |W_t|[$], where M is the BM maximum process and |W| is reflected BM. The pdf for the latter is [$]p(t,x) = \frac{2 \, e^{-x^2/(2 t)}}{\sqrt{2 \pi t}}[$] on [$](0,\infty)[$] and so is also the density of the distribution you seek.

Last edited by Alan on March 16th, 2015, 11:00 pm, edited 1 time in total.

- BerndSchmitz
**Posts:**242**Joined:**

Thanks a lot Alan. This was the kind of answer I was looking for.Is it intuitive why [$]M_t-W_t=^d|W_t|[$]? Or can you recommend any good (online available?) ressources for that?

You're welcome. To me, it is certainly not "obvious", merely plausible. Certainly both sides have non-negative support andhit zero a lot. You'll have to google -- I looked it up in Bertoin's book on Levy processes.

The equality in law P(Mt) = P(Mt-Wt) (is this what you mean?) is what is true.The equality almost surely Mt - Wt = |Wt| could not be true, for instance with >0 prob Wt can be <0 and Mt >0.

Terminology is the same whether r.v. s or processes. To sum up, in continuous time:for processes: Mt - Wt ~ |Wt| , and for fixed t ~ Mtin discrete time (binomial tree)for fixed t: Mt - Wt ~ Mt !~ |Wt|In a trinomial tree at least |Wt| and Mt have the same support, so there is some hope of redefining things to make them have the same law.

Last edited by savr on March 24th, 2015, 11:00 pm, edited 1 time in total.

GZIP: On