A Mysterious Consequence of Discrete Optional Sampling Theorem

107 Views Asked by At

I'm trying to understand the proof of optional stopping theorem for continuous Martingales. They begin by recalling the equivalent fact for a Martingales that take values on a discrete set:

Recall the optional sampling theorem for stopping times $\alpha, \beta: \Omega \rightarrow {t_j}_{j \geq 1}$ ̨with values in a discrete set: $$(X(t_j),F_{t_j})_{j \geq 1} \text{ submartingale, } \alpha \leq \beta \leq C \implies X(\alpha) \leq E[X(\beta)|F_{\alpha}]$$

They proceed to use the statement in the proof of the continuous case:

Using the discrete version of optional sampling theorem with $\alpha = \sigma_j \wedge k$ and $\beta = \tau_j \wedge k$ reveals $X(\sigma_j \wedge k), X(\tau_j \wedge k) \in L^1(P)$ and $X(\sigma_j \wedge k) \leq E[X(\tau_j \wedge k)|F_{\sigma_j \wedge k}]$.

I see where the inequality comes from, but what about the assertion that $X(\sigma_j \wedge k), X(\tau_j \wedge k) \in L^1(P)$?

1

There are 1 best solutions below

4
On BEST ANSWER

Here is a direct way to see integrability using the definition of $\tau_j$. The definition of $\tau_j$ given in your book (included here for completeness) is $$\tau_j(\omega) = \begin{cases} \tau(\omega) \qquad \text{ if } \tau(\omega) = \infty, \\n2^{-j} \qquad \text{ if }(n-1)2^{-j} \leq \tau(\omega) < n2^{-j} \end{cases}$$ where $\tau$ is some stopping time (the case of $\sigma_j$ will be similar so I just deal with $\tau_j$ here).

In particular, $\tau_j \wedge k \in \{ 2^{-j}n : n \leq 2^j k\}$. Hence $$|X(\tau_j \wedge k)| \leq \sum_{n=1}^{2^jk} |X(2^{-j}n)| \in L^1(P)$$ since, by assumption, $X(2^{-j}n) \in L^1(P)$ for each $n$. So $X(\tau_j \wedge k) \in L^1(P)$.