Consider for simplicity the following brownian market:

$$dS^0_t= r S^0_tdt$$

$$dS^1_t= S^1_t(r dt + dW^1_t + dW^2_t) $$

where the filtration is generated by $W^1,W^2$

Consider now $W_t:= \frac{1}{2}(W^1_t + W^2_t)$, which is a Brownian motion, too.

When I substitute $W_t$ in the above financial market, I would have $1$ Brownian motion and $1$ risky asset, so I could conclude that the market is complete. But that is not correct, since completness depends on martingale representation, which depends again on the underlying filtration.

But why do I have to pick another filtration in this context, since the filtration generated by $W^1,W^2$ is the same as being generated by $W$?

  • 1
    $\begingroup$ It is not the same filtration. Knowing the sum of two components is not same as knowing the individual parts. $\endgroup$
    – fes
    Commented Sep 13, 2021 at 11:41
  • $\begingroup$ I see. Thank you for this inuitive comment. How could I proof that? $\endgroup$ Commented Sep 13, 2021 at 12:52
  • $\begingroup$ Prove that $\sigma\{(W_s^1,W_s^2)\,s\leq t \} \neq \sigma\{W_s^1+W_s^2\,s\leq t \}$ $\endgroup$
    – fes
    Commented Sep 13, 2021 at 13:41
  • $\begingroup$ How could I do that? I do not know exatly an element of the set will look like? $\endgroup$ Commented Sep 13, 2021 at 13:46

1 Answer 1


tl;dr: informally: You cannot know the value of the difference of two random variables by knowing their sum.

Consider the following set $A =\{ \omega \in \Omega | W_1(t)(\omega) - W_2(t)(\omega) \in [0,1] \}$ This set is of course in the Filtration generated by $W_1$ and $W_2$ since the addition of measurable functions is measurable. Is this set in the (augmented) generated Filtration of $W$? For that either $P(A) = 0$ (because we consider the augmented filtration) or it exists a borel set $B$ (a set that can be build from intervals through countable set operations) s.t. $$\{ \omega \in \Omega | W(\omega)(t) \in B \} =A \ \mathbb{P}-a.s.$$ (you don't need to consider past time points since the brownian filtration is increasing ).

$P(A) > 0 $ since $W_1(t)- W_2(t)\sim \mathcal{N}(0,2t) .$

Assume it exists such a $B$. Then

$$ \{ \omega \in \Omega |W_1(t)(\omega) + W_2(t)(\omega)) \in B \}=\{ \omega \in \Omega | W_1(t)(\omega) - W_2(t)(\omega) \in [0,1] \} $$ (we absorb the 0.5 factor into the set B) Which is equivalent to:

$$ X + Y \in B \Leftrightarrow X-Y \in [0,1] \ \mathbb{P}-a.s. \ (\#)$$ where $X,Y$ are iid normally distributed. This is equivalent to

$$ Y \in B-X \Leftrightarrow Y \in [-1,0] \ \mathbb{P}-a.s.$$ Thus:

$$ B-X =[-1,0], \mathbb{P}-a.s.$$

inputting into LHS of (#):

$$ Y\in [-1,0] \ \mathbb{P}-a.s.$$

Which is wrong since $ Y$ is normally distributed.

Thus the filtrations are not equal

As an example for incompleteness: The derivative $H(t) = \mathbb{1}_{A}\mathbb{1}_{t}$ cannot be replicated.


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.