This is a follow-up to my previous question on MathOverflow.

Is there a way to combine the Dambis-Dubins-Schwarz theorem and the Martingale Representation Theorem to get the following result?

Let $(M_t, t\ge 0)$ be a continuous $(\mathcal{F}_t)_{t \ge0}$ local martingale, then there exists some Brownian motion $B$ and some progressively measurable process $\xi$ on some probability space such that $$M_t = \int_0^t \xi_s dB_s = \large{ B_{\int_0^t \xi_s^2 ds}} \quad \forall t \in [0,\infty)$$

In particular it would follow that $\langle M \rangle_t=\int_0^t \xi_s^2 ds$ (by Ito's Isometry?).


This appears to be correct (at least for continuous local martingales such that $\langle M_t \rangle_{\infty}=\infty$), provided the following claim is true:

Claim If $(M_t, t\ge 0)$ is a continuous local $(\mathcal{F}_t)_{t \ge 0}$ martingale, then it is also a continuous local $(\mathcal{F}_{\tau_u})_{u \ge0}$ martingale, where $\tau_u:=\inf\{t: \langle M \rangle_t > u\}$.

The truth of this claim might be a direct corollary of the Dambis-Dubins-Schwarz Theorem, although I am not entirely certain of this.

Let's assume we have a continuous $(\mathcal{F}_t)_{t \ge 0}$ local martingale $(M_t, t\ge 0)$ such that $\langle M \rangle_{\infty}=\infty$. Then it is also a $(\mathcal{F}_{\tau_u})_{u \ge 0}$ local martingale, which by the Dambis-Dubins-Schwarz Theorem equals $B_{\langle M \rangle_t}$ for a standard Brownian motion on the same probability space $B_u$ whose natural filtration is $(\mathcal{F}_{\tau_u})_{u \ge 0}$. Then it follows from the martingale representation theorem that $(M_{\tau_u}, u\ge0)$ (the "same" continuous local martingale as before, but using the coarser filtration $(\mathcal{F}_{\tau_u})_{u \ge 0}$ as opposed to the original filtration) can be written as $M_{\tau_u}=M_0 + \int_0^u \xi dB$ for a predictable, $B$-integrable process $\xi$. It then follows (by Ito's Isometry?) that $\langle M \rangle_t = \int_0^t \xi_s^2 ds$.

(But why is this integral even necessarily defined? Just because $\xi$ is predictable doesn't necessarily mean that it is of finite variation, hence I don't see how we can necessarily conclude that the pathwise Lebesgue-Stieltjes integral is applicable, unless we can somehow guarantee that $\xi^2$ is a finite variation process, which seems dubious to me.)

In summary

Local Martingales as Both Time Changes of and Integrals of Brownian Motion Let $(M_t, t\ge 0)$ be a continuous local $(\mathcal{F}_t)_{t \ge 0}$ martingale. Define $\tau_u:=\inf\{t: \langle M \rangle_t > u \}$. Then there exists a standard $(\mathcal{F}_{\tau_u})_{u \ge 0}$ Brownian motion $(B_u, u \ge 0)$ and a predictable $B$-integrable process $\xi$ such that $$\langle M \rangle_t = \int_0^t \xi_s^2 ds$$ and $$M_{\tau_u} = B_{\langle M \rangle_t}=B_{\int_0^t \xi_s^2 ds}= M_0 + \int_0^{\tau_u} \xi dB.$$

According to Professor Marc Yor, we have:

Dambis-Dubins-Schwarz Theorem If $(M_t, t\ge 0)$ is a continuous local $(\mathcal{F}_t)_{t \ge 0}$ martingale such that $\langle M \rangle_{\infty}=\infty$, then the process $$B_u=M_{\tau_u}, \quad \tau_u=\inf\{t:\langle M \rangle_t > u\}$$ is a standard $(\mathcal{F}_{\tau_u})_{u \ge 0}$ Brownian motion, and $M$ may be represented as: $$M_t=B_{\langle M \rangle_t}, \quad t \ge 0$$

I believe that the validity of the representation $M_t=\beta_{\langle M \rangle_t}$ must have as a corollary that $M_t$ is a continuous local $(\mathcal{F}_{\tau_u})_{u \ge 0}$ martingale, although again I am not 100% sure about this.

According to George Lowther (note that his notation is not the same as Marc Yor's), we have:

Martingale Representation Theorem Let $B$ be a standard Brownian motion defined on a probability space $(\Omega, \mathcal{F}, \mathbb{P})$ and $( \mathcal{F}_t )_{t \ge 0}$ be its natural filtration. Then every local $(\mathcal{F}_t)_{t \ge 0}$ martingale $(M_t,t \ge 0)$ can be written as $$M_t=M_0 + \int_0^t \xi dB$$ for a predictable, $B$-integrable process $\xi$.


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.