I'd be astounded if this isn't a duplicate, but I've yet to find anything equivalent. Whenever I've seen the natural filtration of a Brownian Motion stated explicitly, it's been in the form of $\mathcal{F}_t=\sigma\{W_s:s\leq t\}$ for any $t\geq0$. In other words, the natural filtration of a Brownian Motion up to some point $t$ is the sigma-field generated by all of the Brownian Motions from every time not after $t$.

My question is, why is that last bit necessary? Why do we not just have $\mathcal{F}_t=\sigma\{W_t\}$ for any $t\geq0$? By definition $W_t$ is $\sigma\{W_t\}$-measurable and $W_t$ can clearly take all of the values that any $W_s$ (with $s\leq t$) can, so I can't see what is gained by adding the fields generated by the "earlier" random variables. Even in cases where we want to talk about some point $b\leq t$, I can't see why we would use $\mathcal{F}_b=\sigma\{W_a:a\leq b\}$ when we could use $\mathcal{F}_b=\sigma\{W_b\}$. Indeed, I can't see why these two fields would not be identical. Is there an obvious element that I'm missing?

  • $\begingroup$ Yes, that's a filtration, but it loses information compared to the "natural" one. Keep in mind that events are evaluated with respect to a sigma-algebra, so losing the past means all you know is where the process is right now. $\endgroup$ – nomen Feb 25 at 17:06
  • $\begingroup$ @nomen Isn't that a contradiction? You've just said that my proposed filtration is a filtration that is smaller than the natural one. $\endgroup$ – J. Mini Feb 25 at 17:08
  • $\begingroup$ There are lots of filtrations... the constantly empty filtration is the smallest. The smallest one that captures the "past" is called the natural filtration. $\endgroup$ – nomen Feb 25 at 17:11
  • $\begingroup$ @nomen Ahh, so you're saying that the process is not adapted to the filtration that I've proposed? $\endgroup$ – J. Mini Feb 25 at 17:13

I think I've got it, or at least something along the right lines.

The proposed filtration is not a filtration at all. If it was, then we would have $\mathcal{F}_s\subseteq\mathcal{F}_t$ for all $s\leq t$. However, this is clearly not the case. Consider $\{W_0=0\}\subseteq\mathcal{F}_0$ where $\mathcal{F}_0=\sigma\{W_0\}$. Clearly this set is in $\mathcal{F}_1$ if we take $\mathcal{F}_1=\sigma\{W_s:s\leq 1\}$ (and therefore we have $\mathcal{F}_0\subseteq\mathcal{F}_1$), but this is not the case if we take $\mathcal{F}_1=\sigma\{W_1\}$ (and therefore we have $\mathcal{F}_0\not\subseteq\mathcal{F}_1$), meaning that the proposed filtration is not a filtration.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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