2
$\begingroup$

Question: Solve the stochastic differential equation:

$$ dX_t=X^3_t\,dt-X^2_t\,dW_t $$ where: $$ X_0=1 $$

My Attempt:

Using Ito's with: $$ f(x)=\log(x) $$ I get that: $$ d\log(X_t)=dt\left(0+\left(\frac 1{X_T}\right)X_T^3+\left(\frac {-1}{X_T^2}\right)X^4_T\right)+\left(\frac 1{X_T}\right)\left(-X_T^2\right)\,dW_t $$ which gives me that: $$ d\log X_t=-X_t\,dW_t $$ and integrating from 0 to t: $$ X_t=\exp\left(-\int_0^tX_s\,dW_s\right) $$ Is this correct? I don't have too much experience with SDEs so would appreciate any hints/advice.

Cheers

|cite|improve this question
$\endgroup$
  • $\begingroup$ Note that your result gives $(X_t)$ as a function of itself. Also, $\log$ is not $C^2$ on $\mathbb R_+$ so you have to be careful when applying Itô's formula to $\log$ when you don't know if $(X_t)$ hits $0$ or not. $\endgroup$ – Ian May 21 '14 at 0:53
  • $\begingroup$ So, could I use another function instead? $\endgroup$ – dimebucker May 21 '14 at 0:57
  • $\begingroup$ Usually, you either consider the process stopped at $T_0$ the hitting time of zero, and then check that $T_0=+\infty$ almost surely, or you solve "as if it were $C^2$", and then check that the derived solution indeed solves the SDE. Aside from this technicality, you have proved that considering $\log$ does not help you derive a solution in this case. $\endgroup$ – Ian May 21 '14 at 0:59
2
$\begingroup$

If $X_t=F(W_t)$, one knows that $\mathrm dX_t=F'(W_t)\mathrm dW_t+\frac12F''(W_t)\mathrm dt$. If ever there exists some function $F$ such that $$F(0)=1,\qquad F'(w)=-F(w)^2,\qquad F''(w)=2F(w)^3,$$ the proof is complete. Can you identify such a function $F$? Be aware though that there might be no solution $(X_t)$ defined for every nonnegative $t$.

|cite|improve this answer
$\endgroup$
  • $\begingroup$ $$F(W_t)=\frac 1{W_t}$$but why those conditions? $\endgroup$ – dimebucker May 21 '14 at 7:34
  • $\begingroup$ You want $X$ to solve the SDE $dX=X^3dt-X^2dW$. Hence, if $dX=F'(W)dW+\frac12F''(W)dt$, you need that... // Note that $1/W$ is not defined at time $0$ and that you need $F(0)=1$ (as written in my answer) hence no, $F(w)=1/w$ is not a solution. $\endgroup$ – Did May 21 '14 at 7:37
  • $\begingroup$ thanks, this makes sense now, I get that $F(W_t)=\frac {1}{W_t+1}$ $\endgroup$ – dimebucker May 29 '14 at 13:39

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.