product of martingales bounded in $L^2$

301 Views Asked by At

Let $(M_t)_t$ and $(M_t)_t$ be two càdlàg martingales on the same filtered probability space. We know that $M_{\infty}$ and $N_{\infty}$ are orthogonal in $L^2$. Is it true that $(M_t N_t)_t$ is a martingale? If so, is it uniformly integrable?

I suspect the answers are yes and no, but not sure on how to prove.

On the first statement: $MN$ is clearly adapted and càdlàg and the martingale equality should be equivalent to $$ \mathbb{E}_t [(M_{t+h}-M_t)(N_{t+h}-N_t)] = 0 \quad \forall t,h >0 $$

If possible, provide an answer that does not involve the process $[M,N]$.

1

There are 1 best solutions below

2
On BEST ANSWER

No, in general $(M_t \cdot N_t)_{t \geq 0}$ is not a martingale. Consider the (continous) $L^2$-martingales

$$M_t := \int_0^t f(s) \, dB_s \qquad N_t := \int_0^t g(s) \, dB_s$$

where $f,g$ are deterministic functions such that $$\int_0^{\infty} f(s)^2 \, ds < \infty \quad \int_0^{\infty} g(s)^2 \, ds < \infty$$

Then, by Itô's formula

$$\mathbb{E}(M_t \cdot N_t) = \mathbb{E} \left( \int_0^t f(s) \cdot g(s) \, ds \right) = \int_0^t f(s) \cdot g(s) \,ds$$

for any $t \in [0,\infty]$. This means that $M_{\infty}$ and $N_{\infty}$ are orthogonal if, and only if, $$\int_0^{\infty} f(s) \cdot g(s) \, ds = 0 \tag{1}$$

Obviously, we can choose $f,g$ such that $(1)$ is satisfied and

$$t \mapsto \int_0^t f(s) \cdot g(s) \, ds$$

is not constant. But this means that $\mathbb{E}(M_t \cdot N_t)$ is not constant; hence $(M_t \cdot N_t)_{t \geq 0}$ is not a martingale.