This is not a homework question so please be kind not to remove it right away. I am working on some research but have to justify the following argument: Assume $S_t$ is a continuous stochastic process, don't want to make an assumption about distribution, think about something like a smooth function of Brownian motion. I define another process $$Y_t=\frac{1}{t} \int_0^t S_u du$$ Now I am interested in the limit of $Y_t$ as $t$ approaches zero. I would like to know in what sense the argument would hold, the guess is that the limit is $S(0)$. Please suggest a solution or the way to approach this problem.

| cite | improve this question | | | | |
  • 1
    $\begingroup$ How is S defined? That is if really is a continuous function of BM then there is a lot you can say. So, more detail please. $\endgroup$ – BSteinhurst Nov 20 '12 at 3:21
  • $\begingroup$ ok, let's take a particular case, where $S_t$ is according to the following dynamics $dS_t=adt+bdB_t$ where $B_t$ is a Brownian motion, so that $S_t$ has a lognormal distribution. $\endgroup$ – Kamil Nov 20 '12 at 3:59
  • 1
    $\begingroup$ @ Kamil : Unless mistaken you can think on a path by path basis as long as your process is continuous for every path it is integrable on a fixed time interval with respect to Lebesgue measure "ds", then applying Lebesgue differentiation theorem (wiki) you have that the limit is $S_0$ (it is not a research level question but a good question for math exchange this is why I answer it in comment). Regards. $\endgroup$ – The Bridge Nov 20 '12 at 9:51
  • $\begingroup$ @The Bridge: what if $S_u=u^{1/2}$, don't I have a problem then? Is the statement holds for "any" continuous process $S_u$ regardless how wild it is? $\endgroup$ – Kamil Feb 6 '13 at 3:44

You can prove it reasoning $\omega$ by $\omega$. I mean, if you are working in a probability space $\Omega$, the continuous path $(S_t)_{t\geqslant 0}$ depends on $\omega\in \Omega$. But for all $\omega$, $t\mapsto S_t(\omega)$ is continuous, so the following convergence holds : $$ Y_t(\omega) \rightarrow S_0(\omega) $$

It is enough to conclude that $Y_t$ converges almost surely to $S_0$ as $t$ goes to zero.

| cite | improve this answer | | | | |
  • 2
    $\begingroup$ actually The Bridge had already answered in the same way in the comments. $\endgroup$ – Guillaume Nov 20 '12 at 18:36
  • 1
    $\begingroup$ @ Guillaume : 1up for honesty ;-) $\endgroup$ – The Bridge Nov 20 '12 at 19:24
  • $\begingroup$ yes, thanks the Bridge and Guillaume, I think the answer now is complete as I was looking for some omega mentioning. Thus, I have the convergence for every single omega and therefore convergence in a.s. sense. $\endgroup$ – Kamil Nov 21 '12 at 13:30
  • $\begingroup$ related to time $0$ question: If I assume $dS_t = adt+bdB_t$ and from above $dY_t = \frac{1}{t}(S_t-Y_t)$ then I can use Ito formula for $u(t,S_t,Y_t)$. I would like that to be a martingale and thus require $u_t+0.5b^2u_{ss}+au_s+\frac{1}{t}(s-y)u_y=0$. The last term is not defined at $t=0$ but from above answer as $t$ approaches $0$ I have $\frac{1}{t}(s-y)=\frac{0}{0}$. Is there anything can be said about the last term in the pde? $\endgroup$ – Kamil Nov 22 '12 at 16:14
  • $\begingroup$ @kamil : something is missing in your sde for $Y$, moreover once modification done can you show that the sde has a solution ? Best regards $\endgroup$ – The Bridge Nov 22 '12 at 20:46

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.