Let $(B_{t})_{t\geq 0}$ be a Brownian motion and $(\mathcal{F}_{t})_{t\geq 0}$ its canonical filtration. Find a Stochastic Process $(A_{t})_{t\geq 0}$ such that $(B_{t}^{4}+A_{t})_{t\geq 0}$ is a Martingale.

I don't want to settle for a single $(A_{t})_{t\geq 0}$ process, I would like to be able to find a family of processes.

My attempt: Let $X_{t}:=B_{t}^{4}+A_{t}$ and $t>s$, then we have \begin{align} \mathbb{E}[X_{t}-X_{s}|\mathcal{F}_{s}]&= \mathbb{E}[B_{t}^{4}+A_{t}-B_{s}^{4}-A_{s}|\mathcal{F}_{s}]\\ &=3(t-s)^{2}+6(t-s)B_{s}^{2}+\mathbb{E}[A_{t}-A_{s}|\mathcal{F}_{s}] \end{align} If $(X_{t})_{t\geq 0}$ is a martingale, then we should have $$0=3(t-s)^{2}+6(t-s)B_{s}^{2}+\mathbb{E}[A_{t}-A_{s}|\mathcal{F}_{s}]$$ and and we should also have that $(A_{t})_{t\geq 0}$ is adapted to filtration $(\mathcal{F}_{t})_{t\geq 0}$. Therefore, we have $$A_{s}=3(t-s)^{2}+6(t-s)B_{s}^{2}+\mathbb{E}[A_{t}|\mathcal{F}_{s}] \tag{$\bigstar$}$$

Therefore, it all comes down to solving equation ($\bigstar$).

  • $\begingroup$ How about Ito's formula? You are not allowed to use it ? $\endgroup$ – Paresseux Nguyen Nov 19 at 3:26
  • $\begingroup$ @ParesseuxNguyen No, this exercise is in a section where Ito's formula is not covered. $\endgroup$ – Diego Fonseca Nov 19 at 3:27
  • $\begingroup$ Are there any conditions ruling out the trivial solutions $A_t = -B_t^4 + c$ for any $c \in \mathbb{R}$? $\endgroup$ – user6247850 Nov 19 at 3:40

(I take it that you are looking for a non-trivial solution - otherwise you could simply take $A_t := -B_t^4$, as was pointed out in a comment.)

Using that $(B_t)_{t \geq 0}$ has independent and stationary increments and the fact that $B_t \sim N(0,t)$, it follows that

$$M_t^{(\lambda)} := \exp \left(\lambda B_t - \frac{1}{2} \lambda^2 t \right)$$

is a martingale for any $\lambda \in \mathbb{R}$. Equivalently,

$$\int_F M_s^{(\lambda)} \, d\mathbb{P} = \int_F M_t^{(\lambda)} \, d\mathbb{P} \tag{1}$$

holds for all $s \leq t$ and $F \in \mathcal{F}_s$. Now the idea is to differentiate $(1)$ with respect to $\lambda$. For the first derivative, we get

$$\int_F (B_s-\lambda s) M_s^{(\lambda)} \, d\mathbb{P} = \int_F (B_t-\lambda t) M_t^{(\lambda)} \, d\mathbb{P}. \tag{2}$$

Since this holds in particular for $\lambda=0$, we see that $\int_F B_t \, d\mathbb{P} = \int_F B_s \, d\mathbb{P}$ for all $s \leq t$ and $F \in \mathcal{F}_s$, which means that $(B_t)_{t \geq 0}$ is a martingale (no surprise here; we know this already). By calculating higher derivatives, we can increase the "power" of the martingale. E.g. by differentiating $(2)$ once more again with respect to $\lambda$, we find that

$$\int_F ((B_s-\lambda s)^2-s) M_s^{(\lambda)} \, d\mathbb{P} = \int_F ((B_t-\lambda t)^2-t) M_t^{(\lambda)} \, d\mathbb{P}.$$

Putting $\lambda=0$, we see that $(B_t^2-t)_{t \geq 0}$ is a martingale. To get a martingale involving $B_t^4$ we need two differentiate two more times. Doing so yields

$$\int_F (B_s-\lambda s)((B_s-\lambda s)^2-3s) M_s^{(\lambda)} \, d\mathbb{P} = \int_F (B_t-\lambda t)((B_t-\lambda t)^2-3t) M_t^{(\lambda)} \, d\mathbb{P}$$

(which shows, for $\lambda=0$, that $(B_t^3-tB_t)_{t \geq 0}$ is a martingale) and

\begin{align*} &\int_F \left[((B_s-\lambda s)^2-s)((B_s-\lambda s)^2-3s)-2s(B_s-\lambda s)^2\right] M_s^{(\lambda)} \, d\mathbb{P} \\ &= \int_F\left[((B_t-\lambda t)^2-t)((B_t-\lambda t)^2-3t)-2t(B_t-\lambda t)^2 \right] M_t^{(\lambda)} \, d\mathbb{P},\end{align*}

which shows, for $\lambda=0$, that

$$(B_t^2-t)(B_t^2-3t)-2t B_t^2=B_t^4-6tB_t+3t^2$$

is a martingale.

Remark: There is a close connection to Hermite polynomials, see e.g. this question.

| cite | improve this answer | |

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.