2
$\begingroup$

Definition from my textbook:

A stochastic process $X = (X_n, n \in \mathbb{N}_0)$ is called predictable (or previsible) with respect to the filtration $\mathbb{F} = (\mathcal{F}_n, n \in \mathbb{N}_0 )$ if $X_0$ is constant and if, for every $n \in \mathbb{N}$ $$X_n \text{ is } \mathcal{F}_{n-1}\text{-measurable.}$$

Okay, but that surely cannot mean that I can really predict the next outcome $X_{n+1}$ if I know $X_0, \ldots, X_n$ (regardless if $X$ is also adapted to $\mathbb{F}$), right?

$\endgroup$
  • 2
    $\begingroup$ No you cannot, rather this means that one can predict the next outcome $X_{n+1}$ if one knows $\mathcal{F}_n$ (and there is usually more information in this than just $X_0$, ..., $X_n$). $\endgroup$ – Did Mar 7 '15 at 9:46
  • $\begingroup$ I agree with @Did. An easy example for a filtration $\mathcal{F}_n$ which contains more information than $X_0,\ldots,X_n$ is $\mathcal{F}_n := \mathcal{A}$ where $\mathcal{A}$ denotes the $\sigma$-algebra on the probability space $\Omega$. $\endgroup$ – saz Mar 7 '15 at 10:01
  • $\begingroup$ Somehow I'm interested in the notion @Did mentioned. How can conditioning on the filtration and conditioning on the sample path differ? I guess conditioning on $X_{1} = x_{1},X_2 = x_2, \ldots, X_n = x_n$ is the same thing with conditioning on $\mathcal{F}_n = \sigma(X_1,\ldots,X_n)$, no? What is the alternative filtration which can cause the case mentioned above? $\endgroup$ – user48547 Aug 10 '16 at 20:46
  • $\begingroup$ @oeda Post this as a question on the site then. $\endgroup$ – Did Aug 10 '16 at 23:12
  • $\begingroup$ OK, it will also help me to make sense of the question. $\endgroup$ – user48547 Aug 11 '16 at 13:30
0
$\begingroup$

It is not clear -mathematically speaking- what do you mean by predict, but in a certain way, you can predict $X_n$ by knowing $ \mathcal{F}_{n-1} $ (which may be the same as knowing $X_0,X_1,...,X_{n-1}$ if $\mathcal{F}_n$ is the natural filtration).

The thing is, in applications, a previsible process has to do with a strategy (i.e. a decision you make taking into account the previous information, for example a gambling strategy) rather than with a system you are analysing.

See for example Probablity with Martingales , by Williams.

I'm sorry if this answer arrives late!

$\endgroup$
  • $\begingroup$ Am I right to think that a for a previsable process, one can in some sense estimate $X_{n}$ without having to taking avarages of values with some subsets. One can estimate it just as the immediate predecessor. $\endgroup$ – user415535 Sep 5 '17 at 11:05
  • $\begingroup$ Mmm I don't think so, that sounds to me as a Martingale!! $\endgroup$ – Max Sep 5 '17 at 12:40
  • $\begingroup$ hmm if we have a martingale then one takes the avarages of $X_{n}$ over the sets in $F_{n-1}$ which in turn have the same distriution as $X_{n-1}$. In this case this we dont take avarages but get the possible values and thier probabilites just by looking at the predecessor $X_{n-1}$ to get the value of $X_{n}$. Or have I slipped somewhere? $\endgroup$ – user415535 Sep 5 '17 at 13:10
  • 1
    $\begingroup$ You are right about the averages. On the other hand, take the trivial previsible (deterministic) process $X_n =n$. In which sense you estimate $X_{n+1}$ as $X_n$ ? $\endgroup$ – Max Sep 5 '17 at 13:14
  • $\begingroup$ Right, I was thinking that $C_{n}$ was a martingale aswell...So it more like "if we know what event in $\mathcal{F}_{n-1}$ did occur then we know the value of $X_{n}$? $\endgroup$ – user415535 Sep 5 '17 at 13:22
0
$\begingroup$

Technically, you can always predict in terms of the best available information but you will NOT necessarily know the value of $X_{n+1}$ if you know the values of $X_0, \ldots, X_n$. This is easy to show since based on the definition of a previsible process, if you are at $(n-1)^{th}$ index of the filtration (i.e. $\mathcal{F}_{n-1}$), you already know the values of $X_0, \ldots, X_n$ but you don't get to know the value of $X_{n+1}$. You will know $X_{n+1}$ when you are at $n^{th}$ index of filtration (i.e. $\mathcal{F}_{n}$). I think in the question there seems to be a confusion regarding what subscript $n$ means in $X_n$. It doesn't mean when you know $X_0, \ldots, X_n$ you are at index $n$ in the filtration. In fact, you know those values when you are at $\mathcal{F}_{n-1}$ which doesn't tell what $X_{n+1}$ is.

$\endgroup$

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.