1
$\begingroup$

It is well known that for the diffusions

\begin{align*} dX&=f(X)dt+&cdB\\ dY&=&cdB \end{align*}

the density of the law of $X$ with respect to the law of $Y$ is

\begin{align*} \frac{d\mu_X}{d\mu_Y}(c B)&=\exp\left(\int_0^T\frac{f(cB(t))}{c}dB(t)-\frac12\int_0^T\frac{f^2(cB(t))}{c^2}dt\right)\\ &=\exp\left(\frac{1}{c^2}\left(\int_0^Tf(cB(t))d(cB(t))-\frac12\int_0^Tf^2(cB(t))dt\right)\right)\\ &:=\exp\left(\frac{1}{c^2}L(cB(t))\right),\end{align*}

where $L(cB)=\int_0^Tf(cB(t))d(cB(t))-\frac12\int_0^Tf^2(cB(t))dt$. My interest now is in making $\frac{1}{c^2}L(cB(t))$ defined pathwise. Applying integration by parts yields that

$$\frac{1}{c^2}L(cB)=\frac{1}{c^2}\left(F(cB(T))-\frac{c^2}{2}\int_0^T f'(cB(t))dt-\frac12\int_0^T f^2(cB(t))dt\right):=\frac{1}{c^2}L^c(cB).$$

I guess my question is - why does the "pathwise" functional depend on $c$ where the nonpathwise functional doesn't? What is the term $\int_0^T f'(B(t))dt$ in relation to the original $L$? Is there a nice expression?

If we define $L_1(B)=\int_0^Tf(B(t))d(B(t))-\frac12\int_0^Tf^2(B(t))dt$ and $L_2(B)= F(B(T))-\frac{1}{2}\int_0^T f'(B(t))dt-\frac12\int_0^T f^2(B(t))dt\ $. Then $L_1(B)=L_2(B)$ however seemingly $L_1(cB)\neq L_2(cB)$. Edit:actually this comes from the fact that Ito's lemma only holds a.s. and cB is singular with B for $c\neq \pm 1$.

$\endgroup$

1 Answer 1

2
$\begingroup$

In a nutshell you are asking the following:

When $F(x)=\int_0^xf(y)\,dy$ we get from Ito $$ F(cB(t))=\int_0^Tf(cB(t))\,d(cB(t))+\frac{1}{2}\int_0^Tf'(cB(t))\,c^2\,dt\, $$ so that the expressions \begin{align} &F(cB(t))-\frac{c^2}{2}\int_0^Tf'(cB(t))\,dt\,,\tag{1}\\ &\int_0^Tf(cB(t))\,d(cB(t))\tag{2} \end{align} are equal. Then why (2) depends only on $cB(t)$ and not separately on $c^2$ and $cB(t)$ like (1) seems to do?

Good question. I think the answer is that (1) can also be written as \begin{align} &F(cB(t))-\frac{1}{2}\int_0^Tf'(cB(t))\,d\big\langle cB\big\rangle_t\,,\tag{1'} \end{align} and gone is the isolated $c^2$.

$\endgroup$
0

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.