Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Consider a probability filtred space $ (\Omega, \mathcal F, \mathcal F_ t, \mathbb P)$ and a continuous $\mathcal F _t$-martingal starting from $0$, $ M = (M_t)_{t \geq 0}$, such that $\left \langle M \right \rangle_\infty \leq 1$ $\mathbb P$-ps. Now, we define by recurence $ \forall n \in \mathbb{N}$ $$ I^{(o)}_t \equiv 1, \ I^{(n+1)}_t \int _0 ^t I^{(n)}_s d M_s \ , \ t \geq 0 $$

The question: How to show that $I^{(n)}$ is a $\mathcal F _t $-martingal with $I^{(n)}_t$ in $L^2(\Omega, \mathcal F, \mathbb P)$ and that we have

$$ \mathbb{E} \left \{ \sup_ {t \geq 0} \left | I^{(n)}_t \right | ^2 \right \} \leq 4^n $$

Element of answer: for the last inequality we can maybe start using Burkholder-Davis-Gundy inequality, then we'll have :

\begin{align} \mathbb{E} \left \{ \sup_ {0 \leq t \leq T} \left | I^{(n)}_t \right | ^2 \right \} & \overset{BDG}{\leq} C\mathbb{E} \left \{ \left \langle I^{(n)} \right \rangle_T \right \} \\&= C\mathbb{E} \left \{ \int_0 ^T \left |I^{(n-1)}_s\right | ^2 d\left \langle M\right \rangle _s \right \} \\ & \overset{Fubini}{=} C \int_0 ^T \mathbb{E} \left \{\left |I^{(n-1)}_s\right | ^2 \right \}d\left \langle M\right \rangle _s \\ & \overset{Ito's isometry}{=} C \int_0 ^T \mathbb{E} \left \{\int _0 ^s\left |I^{(n-2)}_s\right | ^2 d\left \langle M\right \rangle _{s_1}\right \}d\left \langle M\right \rangle _s \\ &=(...) \\& = C \int_0 ^T \int_0 ^s \int_0 ^{s_1} ... \int_0 ^{s_n-2} 1 \ d\left \langle M\right \rangle _{s_{n-2}} ... d\left \langle M\right \rangle _{s_2} d\left \langle M\right \rangle _{s_1} d\left \langle M\right \rangle _{s}\end{align}


Did sugestion: Could you please, check if I understood ?

You should notice that all the steps where the Fubini's theorem was applied are wrong!

\begin{align} \mathbb{E} \left \{ \sup_ {0 \leq t \leq T} \left | I^{(n)}_t \right | ^2 \right \} & \overset{BDG}{\leq} 4\mathbb{E} \left \{ \left \langle I^{(n)} \right \rangle_T \right \} \\&=4\mathbb{E} \left \{ \int_0 ^T \left |I^{(n-1)}_s\right | ^2 d\left \langle M\right \rangle _s \right \} \\ & \overset{Fubini}{=} 4 \int_0 ^T \mathbb{E} \left \{\left |I^{(n-1)}_s\right | ^2 \right \}d\left \langle M\right \rangle _s \\ & \overset{}{\leq} 4 \int_0 ^T \mathbb{E} \left \{\sup_{0\leq u \leq s}\left |I^{(n-1)}_u\right | ^2 \right \}d\left \langle M\right \rangle _s \\ & \overset{BDG}{=} 4 ^2 \int_0 ^T \mathbb{E} \left \{\int _0 ^s\left |I^{(n-2)}_s\right | ^2 d\left \langle M\right \rangle _{s_1}\right \}d\left \langle M\right \rangle _s \\ &=(...) \\& \leq 4^n \int_0 ^T \int_0 ^s \int_0 ^{s_1} ... \int_0 ^{s_n-2} 1 \ d\left \langle M\right \rangle _{s_{n-2}} ... d\left \langle M\right \rangle _{s_2} d\left \langle M\right \rangle _{s_1} d\left \langle M\right \rangle _{s}\\ &\leq 4^n\end{align} because $\left \langle M \right \rangle = (\left \langle M \right \rangle_t)_{t\geq 0}$ is an increasing process with $\left \langle M \right \rangle_\infty \leq 1$

share|cite|improve this question
@Did: I've forgoten to accept the answer, sorry. – Paul Mar 18 '13 at 12:15
up vote 1 down vote accepted

This is direct if one recalls that $C=4$ and if one replaces $\left |I^{(n-1)}_s\right | ^2$ by $\sup\limits_{s\geqslant 0}\left |I^{(n-1)}_s\right | ^2$ as soon as possible (and note that the localization by $t\leqslant T$ is not needed).

share|cite|improve this answer
Can you detail the calculation please? – Paul Dec 31 '12 at 1:13
It IS detailed (just read). – Did Dec 31 '12 at 1:20
Could you please, check at the end of the question if I understood ? I still do not understand how to complete the proof neither why it is not necessary to localizate by $t\leq T$. – Paul Dec 31 '12 at 1:38
Edit part, second line of displayed equations: $C=4$. Third line and every subsequent line: this is not Fubini, watch out for random variables vs constants. Everywhere: please follow closely my hint. – Did Dec 31 '12 at 9:05
OP: Why not trying to follow my suggestions? One of the first things to do is to correct the identity in the third line (the one topped by the mention Fubini), which is invalid because $d\langle M\rangle_s$ is random, hence it must stay inside the integral. – Did Jan 1 '13 at 15:43

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.