Quadratic variation and elementary processes

64 Views Asked by At

The context here is the construction of the stochastic integral. In this setting, one defines elementary processes that can be written as: $$ K_t=\sum_{i=0}^{k}X_{a_i}\mathbb{1}_{t \in (a_i,a_{i+1}]} $$ with $X_{a_i}$ being $\mathcal{F}_{a_i}$-measurable and in $L^2$. Now, given a continuous martingale $N$ bounded in $L^2$, and an elementary process $K$, the following formula holds $\forall t \geq 0$:

$$ \langle I(K),N \rangle_t = \int_0^{t} K_s d\langle B,N\rangle_s $$ where $I(K)$ indicates the stochastic integral. I want to prove this equality, but I can't even write down what the left hand side would actually be in this specific case.

1

There are 1 best solutions below

4
On BEST ANSWER

$I(K)_t = \sum_{i=0}^kX_{a_i}(B_{a_{i+1}\wedge t}-B_{a_i\wedge t})=\sum_{i=0}^kX_{a_i}(B^{a_{i+1}}-B^{a_i})_t$, where we denote $Y^a_t:=Y_{a\wedge t}$. Then $$ \begin{align*} \langle I(K),N\rangle_t&=\sum_{i=0}^k\langle X_{a_i}(B^{a_{i+1}}-B^{a_i}),N\rangle_t\\ &=\sum_{i=0}^kX_{a_i}\langle B^{a_{i+1}}-B^{a_i},N\rangle_t\\ &=\sum_{i=0}^kX_{a_i}(\langle B,N\rangle_{a_{i+1}\wedge t}-\langle B,N\rangle_{a_i\wedge t})\\ &=\sum_{i=0}^kX_{a_i}\int_0^t1_{(a_i,a_{i+1}]}(s)\,d\langle B,N\rangle_s\\ &=\int_0^t\sum_{i=0}^kX_{a_i}1_{(a_i,a_{i+1}]}(s)\,d\langle B,N\rangle_s\\ &=\int_0^tK_s\,d\langle B,N\rangle_s. \end{align*} $$

EDIT: let's clarify why the second equality hold. Let $Z:=B^{a_{i+1}}-B^{a_i}$ and $M_t=Z_tN_t-\langle Z,N\rangle_t$. So we want $\langle X_{a_i}Z,N\rangle=X_{a_i}\langle Z,N\rangle$, which amounts to say that $X_{a_i}ZN-X_{a_i}\langle Z,N\rangle=X_{a_i}M$ is a local martingale, or equivalently, $\langle X_{a_i},M\rangle=0$. The latter is straightforward, since for any stopping time $T$, $\langle X_{a_i},M^T\rangle=\langle X_{a_i}^T,M\rangle$. Therefore, for $T=a_i$, and noticing that $M^{a_i}=0$, we get $\langle X_{a_i},M\rangle=0$, which proves the claim.

For the sake of clarity I deliberately not mentionned measurability issues. If you want the proof to be complete, you should also verify that the process $X_{a_i}\langle Z,N\rangle$ is indeed adapted. It is for instance a consequence of the fact that $$ X_{a_i}\langle Z,N\rangle_t=X_{a_i}(\langle B,N\rangle_{a_{i+1}\wedge t}-\langle B,N\rangle_{a_i\wedge t})=X_{a_i}1_{\{t>a_i\}}(\langle B,N\rangle_{a_{i+1}\wedge t}-\langle B,N\rangle_{a_i\wedge t}). $$