Let $(M_t)_{t \in \mathbb R^+}$ be a continuous martingale with regard to a filtration $(\mathcal F_t)$ generated by a continuous stochastic process $(Y_t)$.

Is it true that there exists a Brownian motion $(B_t)$ adapted to $(\mathcal F_t)$ and a stochastic process $(X_t)$ such that $M_t=M_0+\int_0^t X_s d B_s$ ?

This is nearly the martingale representation theorem: Let $(B_t)$ be a standard Brownian motion defined on a probability space $(\Omega, \mathcal{F}, \mathbb{P})$ and $\left\{\mathcal{F}_{t}\right\}_{t \geq 0}$ be its natural filtration. Then, every $\left\{\mathcal{F}_{t}\right\}$-local martingale $M$ can be written as \begin{equation*} M_t=M_{0}+\int_0^t \xi_s d B_s \end{equation*} for a predictable, $B$-integrable, process $\xi$ (statement taken from: https://almostsuremath.com/2010/05/25/the-martingale-representation-theorem/).

In my statement, $(\mathcal F_t)$ is not necessarily generated by Brownian motion. Does it still work?

  • 3
    $\begingroup$ My intuition says no. $\endgroup$
    – user545426
    Commented Feb 14, 2022 at 19:05

1 Answer 1


If a continuous martingale $(M_t)_{t\in\mathbb{R}_+}$ could be written as a stochastic integral with respect to Brownian motion $(B_t)$: \begin{equation*} M_t=M_0+\int^t_0 X_s\,\text{d}B_s, \end{equation*} where $X$ is predictable. Then \begin{equation*} \langle M\rangle_t=\int_0^tX_s^2\,\text{d}s. \end{equation*} Furthermore, from above equality and Radon-Nikodym theorem, for Lebesgue measure $\lambda$, \begin{equation*} \text{d}\langle M\rangle \ll \text{d}\lambda. \quad(\text{absolute continuous of measures})\tag{1} \end{equation*} Now suppose $F$ is a continuous increasing function with $F(0)=0$ and $\text{d}F\bot\text{d}\lambda$(mutual singular), then there exist $\{Z(t), t\ge 0\}$ a continuous Gaussian process with \begin{equation*} \mathsf{E}[Z_t]=0,\qquad \mathsf{E}[Z_sZ_t]=F(s\wedge t), \quad s,t\ge0. \end{equation*} Since $Z$ is a continuous process with independent increments and $\mathsf{E}[Z_t]=0 $, hence $Z$ is also a continuous martingale with $\langle Z\rangle =F $, it doesn't satisfy (1), so continuous martingale $Z$ doesn't be written as a stochastic integral w.r.t. BM.

  • $\begingroup$ What does $\mathrm{d} F \perp \mathrm{d} \lambda$ mean? Also, where can I find a proof of $(1)$ ? $\endgroup$
    – W. Volante
    Commented Feb 15, 2022 at 14:44
  • $\begingroup$ Thanks, I add some interpret. For Radon-Nikodym theorem, pls refer R. Durrett, Probability: Theory and Examples, 5th ed, Th.A.4.8, p.406- $\endgroup$
    – JGWang
    Commented Feb 16, 2022 at 1:26
  • $\begingroup$ Thank you, could you give an example of $F$ that satisfy your conditions ? $\endgroup$
    – W. Volante
    Commented Feb 16, 2022 at 1:35
  • $\begingroup$ $F$ --Cantor-Lebesgue function, c.f. Royden, H. L., Real Analysis, 4th Ed, (2010), p.50. $\endgroup$
    – JGWang
    Commented Feb 16, 2022 at 3:56

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .