4
$\begingroup$

I would first of all like to say that I am an analyst, and so I am familiar with probabilistic methods only on a basic level.

My initial situation is the following. Consider two stochastic differential equations: \begin{align} dX_t &= f(X_t) dt + \sigma(X_t) dW_t\\ dY_t &= f(Y_t) dt + \tilde\sigma(Y_t) dW_t, \end{align} where $W_t$ is the standard 1D-Brownian motion, the initial data for $Y$ and $X$ are the same and the drift $f$ and the diffusion coefficients $\sigma,\tilde\sigma$ have sufficiently nice properties. Now assume that (for some reason) we know that the measures corresponding to this processes are mutually absolutely continuous.

Question: What can I conclude about the diffusion coefficients $\sigma,\tilde\sigma?$. Do they have to be equal (maybe only up to some symmetries or something like that)?

I appreciate any hint or reference. Thanks in advance.

$\endgroup$
3
$\begingroup$

Well, if there are regions on the real line where the processes never get (which can be the case, e.g. if $f\equiv 1$ and $\sigma$ has zeroes with fast enough decay), then, clearly, you cannot say anything about the $\sigma$'s there.

Otherwise the quadratic variation is the invariant that ensures $\sigma=\pm\tilde{\sigma}$. Suppose that $|\sigma(x)| <a<b<|\tilde{\sigma}(x)|$ and let $\tau_x=\min\{t:X_t=x\}$. If $\tau_x<\infty$ with positive probability, then for a small $\epsilon>0$ also $|\sigma(X_t)| <a<b<|\tilde{\sigma}(X_t)|$ for $\tau_x<t<\tau_x+\epsilon$ with positive probability. But on this event $$\langle X_t\rangle_{[\tau_x,\tau_x+\epsilon]}=\int_{\tau_x}^{\tau_x+\epsilon}\sigma^2(X_t)dt<a^2\epsilon$$ and similarly $\langle X_t\rangle_{[\tau_x,\tau_x+\epsilon]}>b^2\epsilon$, which is a contradiction.

$\endgroup$

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.