4
$\begingroup$

Let $I(\cdot)$ be an indicator, and $B_{t}$ be an 1-dim standard Brownian motion in a nice filtered probability space $(\Omega, \mathcal{F}, P, \mathcal{F}_{t})$. We consider a random process $$Y_{t} = I(t\ge 1) B_{t-1}.$$ Obviously, $Y$ is an $\mathcal F_{t}$-adapted process with finite quadratic variation $\langle Y \rangle_{t} = (t-1) I(t\ge 1)$.

[Q1.] Is it a Ito process, i.e. there exists a representation of the form $$Y_{t} = \int_{0}^{t} b_{s}ds + \sigma_{s}d B_{s}, \quad \forall t\ge 0$$ for some $\mathcal F_t$-adapted processes $b$ and $\sigma$?

[Q2.] If No for [Q1.], then is it semi-martingale?

[Q3.] If the answers for the above are both NO, then it gives an example of finite QV which is not semi-martingale. Is there any other such an example which belongs to QV but not in semi-martingale?

$\endgroup$
4
  • $\begingroup$ I have not thought of this issue. Just let the filtration be any larger one which runs at least 1-d BM. $\endgroup$
    – kenneth
    May 9, 2015 at 8:09
  • $\begingroup$ If it were an Ito process, wouldn't we have to have $b_s = \sigma_s = 0$ for $0 \le s < 1$? Then it ought to be possible to show that $Y$ is $\sigma(B_t - B_1 : t \ge 1)$-measurable, hence independent of $\sigma(B_t : t \le 1\}$. That is absurd since $Y_{3/2} = B_{1/2}$. $\endgroup$ May 10, 2015 at 0:55
  • $\begingroup$ @NateEldredge Probably the original question was not precise. In fact, $b$ and $\sigma$ may not be deterministic, and I modified it now. For instance $\sigma_s = B(1/2)$ for all $s>1/2$, then the process $Y_t$ shall not be $\sigma(B_t-B_1: t\ge 1)$-measurable. $\endgroup$
    – kenneth
    May 10, 2015 at 5:25
  • $\begingroup$ @StephanSturm Yes, I agree. $\endgroup$
    – kenneth
    May 10, 2015 at 13:53

2 Answers 2

1
$\begingroup$

Note that for fixed $t_0$, we have by the martingale representation theorem that

$$Y_{t_0} = \int_0^{t_0} I\bigl( s \in [0,t_0-1) \bigr) \, dB_s.$$

In particular that for $t_0\leq 1$ the indicator yields $0$ trivially. However, if we want to consider the process $(Y_t)$, then the integrand $I\bigl( s \in [0,t-1) \bigr)$ is not progressively measurable in the Brownian filtration (i.e. not measurable w.r.t. the $\sigma$-field $\mathcal{B}([0,s]) \otimes \mathcal{F}_s$). Thus it is not an Ito process and also not a semi-martingale as the martingale representation is unique. This answers [Q1] and [Q2].

As for [Q3], one classical example for a process with finite quadratic variation that is not a semimartingale is fractional Brownian motion with Hurst-parameter $h<1/2$. (It has even quadratic variation constant $0$).

$\endgroup$
2
  • $\begingroup$ Thanks. However, I need one $\sigma$ for all $t\ge 0$, that is, $Y_t = \int_0^t \sigma(s, \omega) d B_s$. However, your representation is $\int_0^t \sigma(s, t, \omega) d B_s$. $\endgroup$
    – kenneth
    May 10, 2015 at 5:30
  • $\begingroup$ I got your point. Uniqueness of MRT actually tells us it's not a semi-martingale, and I agree. However, I do not quite understand, what do you mean by $I(s\in [0,t-1))$ being not progressively measurable? And how progressive measurability plays a role here? $\endgroup$
    – kenneth
    May 10, 2015 at 14:00
1
$\begingroup$

Let $\sigma_t=I(t\ge 1)$ and $b_t=0$ for all $t$. So $\sigma_t$ and $b_t$ are deterministic and certainly predictable, adapted, and integrable. Then the Ito process $$ \int_0^t b_d\,ds + \sigma_s\,dB_s = \int_0^t I(s\ge 1)\,dB_s = I(t\ge 1)(B_t-B_1) $$ is identically distributed (but not identical) with your process $I(t\ge 1)B_{t-1}$.

$\endgroup$
3
  • 1
    $\begingroup$ I think the question was for identical, not for identical law (that is, the OP really wants to rewrite $Y_t$ itself as a stochastic integral against $B$). Clearly, one must take $b_t=0$ for equality in law (to preserve the martingale property), and I suspect that no $\sigma_t$ will do the job, but at the moment do not see a convincing argument. $\endgroup$ May 9, 2015 at 20:17
  • $\begingroup$ I wants to know if $B(t-1)$ is Ito process or not. I agree that your representation is giving an identical distribution. But I am not sure it is sufficient to answer it is Ito process? $\endgroup$
    – kenneth
    May 10, 2015 at 5:35
  • $\begingroup$ @kenneth I guess it's not quite $\endgroup$ May 10, 2015 at 5:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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