Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

The setting of my question is the following. We have a diffusion process

$$dS(t) = \mu S(t) \; dt + v(t,S(t)) \; dW(t)$$

where $W$ is a standard Brownian motion under an equivalent martingale measure $Q$ and $v$ satisfies all necessary regularity conditions for there to exist a unique solution (i.e. linear growth, local Lipschitz etc.). The Linear growth condition is the one important for my question:

$$v^2(t,y) \leq L(1+y^2) \text{ for every }t \in [0,T]$$

I'm reading an article where it's claimed that due to the linear growth assumption the marginal density function $p_t(\cdot) = \frac{d}{dy}Q[S(t) \leq y]$ must have second moment i.e. $E[S^2(t)]<\infty$.

It's just mentioned in a footnote so I get the feeling that it's a well known result, but I cannot prove it! Any help is appreciated!

Thank you in advance

share|cite|improve this question
$v(t,S(t))$ needs to have at most linear growth. For instance $v(t,S(t)) = \left( 1+ S(t)^2 \right)^{1/4}$ is regular, but has sub-linear growth. – Sasha Feb 21 '12 at 15:35
Thank you for your input Sasha. I have edited and clarified what I mean by linear growth. – DoubleTrouble Feb 21 '12 at 15:44
up vote 2 down vote accepted

Since $\mathrm d\langle S\rangle(t)=v^2(t,S(t))\mathrm dt$, Itô's formula for $X(t)=S(t)^2$ reads $$ \mathrm dX(t)=2\mu X(t)\mathrm dt+2v(t,S(t))S(t)\mathrm dW(t)+v^2(t,S(t))\mathrm dt. $$ Hence $x(t)=\mathrm E(X(t))=\mathrm E(S(t)^2)$ is such that $$ x'(t)=2\mu x(t)+\mathrm E(v^2(t,S(t))), $$ as long as $x(t)$ stays finite. Now, $v^2(t,s)\leqslant L\cdot(1+s^2)$ for some finite $L$, hence $$ x'(t)\leqslant2\mu x(t)+L\cdot(1+x(t)). $$ This implies that $y(t)=\mathrm e^{-(2\mu+L)t}x(t)$ is such that $y'(t)\leqslant L\mathrm e^{-(2\mu+L)t}$ for every $t\geqslant0$, from which the finiteness of $y(t)$, hence of $x(t)$, is direct. To be more specific, $$ x(t)\leqslant \mathrm e^{(2\mu+L)t}\left(x(0)+\frac{L}{2\mu+L}\right)-\frac{L}{2\mu+L}. $$

share|cite|improve this answer
Thank you Didier for your answer. Even though you differentiated wrong to get x'(t) your answer really helped me out! Thanks a lot! – DoubleTrouble Feb 21 '12 at 16:49
Yes. Corrected now. – Did Feb 21 '12 at 17:04
When going through the details I'm not sure that we know that the expectation of the Ito integral is zeros. So this approach basically boils down into showing that $E[\int_0^t S^2(u)v^2(u,S(u)) dt] < \infty$ – DoubleTrouble Feb 21 '12 at 19:39

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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