1
$\begingroup$

Let $\lambda$ be a probability measure on $\Omega$ (finite), with filtration $\{\mathcal{F}_t\}$. Define $\nu(X) = \lambda\left(X\frac{d\nu}{d\lambda}\right)$, where $\frac{d\nu}{d\lambda}$ is a random variable i.e., $\nu(\omega) = \lambda(\omega)\frac{d\nu}{d\lambda}(\omega)$, all $\omega\in\Omega$. Show that $$E\nu[X|\mathcal{F_t}] = \frac{E_{\lambda}\left[X\frac{d\nu}{d\lambda}|\mathcal{F_t}\right]}{E_{\lambda}\left[\frac{d\nu}{d\lambda}|\mathcal{F}_t\right]}$$

Recall from the second fundamental theorem of asset pricing $$\frac{d\nu}{d\lambda} = \frac{S_T^{0}}{\lambda(S_T^{0})}$$ if $S_T^{0}$ is a constant then $$\frac{d\nu}{d\lambda} = 1 \ \ \Rightarrow \ \ \lambda = \nu$$ The change of measure formula is $$E_{\nu}[X] = E_{\lambda}\left[X\frac{d\nu}{d\mu}\right]$$

For some attainable claim $X$ let $\phi$ be a self financing strategy replicating $X$ then by the first fundamental theorem of asset pricing $$V_t(\phi) = E_{\nu}\left[X\frac{S_t^{0}}{S_T^{0}} |\mathcal{F_t}\right]$$

I am pretty sure the result will follow from one of these fundamental theorems of asset pricing but I am not sure where to go from here. Sorry for the messy start, also if you need me to write the three fundamental theorems I would be happy to do so. Any comments or suggestions is greatly appreciated.

Alternative Solution - For all $\omega\in \Omega$, let $\mathcal{F}_t(\omega) = \mathcal{F}_t$ be the partition element containing $\omega$. Then

\begin{align*} E_{\nu}[X|\mathcal{F}_t](\omega) &= \frac{\sum_{\omega\in\mathcal{F}_t(\omega)} X(\omega)\nu(\omega)}{\sum_{\omega\in\mathcal{F}_t(\omega)} \nu(\omega)}\\ &= \frac{\sum_{\omega\in\mathcal{F}_t(\omega)} X(\omega)\lambda(\omega)\frac{d\nu}{d\lambda}(\omega)}{\sum_{\omega\in\mathcal{F}_t(\omega)}\lambda(\omega)\frac{d\nu}{d\lambda}(\omega)}\\ &= \frac{\left( \frac{\sum_{\omega\in\mathcal{F}_t(\omega)} X(\omega)\lambda(\omega)\frac{d\nu}{d\lambda}(\omega)}{\sum_{\omega\in\mathcal{F}_t(\omega)} \lambda(\omega)} \right )}{\left(\frac{\sum_{\omega\in\mathcal{F}_t(\omega)} \lambda(\omega)\frac{d\nu}{d\lambda}(\omega)}{\sum_{\omega\in\mathcal{F}_t(\omega)} \lambda(\omega)} \right )}\\ &= \frac{E_{\lambda}\left[X\frac{d\nu}{d\lambda}|\mathcal{F}_t\right](\omega)}{E_{\lambda}\left[\frac{d\nu}{d\lambda}|\mathcal{F}_t\right](\omega)} \end{align*}

$\endgroup$
  • $\begingroup$ If $\lambda$ and $\nu$ are measures, then $\lambda(\omega)$ and $\nu(\omega)$ do not make good sense. What do you mean $\lambda(S_T^0)$? Is $S_t^0$ a numeraire process? $\endgroup$ – Gordon Oct 26 '16 at 16:21
  • $\begingroup$ @Gordon I may have written the question down wrongly I will double check. $S_t^{0}$ is a numeraire process. Apologies if I made some notation mistake, I will correct it. $\endgroup$ – Wolfy Oct 27 '16 at 0:18
  • $\begingroup$ @Gordon in regards to the $\lambda(S_T^{0})$ question, it is just a normalization factor to ensure that $\nu$ has total mass $1$. We define $\nu$ by means of the Radon-Nikodym derivative. $\endgroup$ – Wolfy Oct 27 '16 at 1:02
  • $\begingroup$ See also here. $\endgroup$ – Gordon Oct 27 '16 at 13:49
  • $\begingroup$ @Gordon posted the solution but the latex generator here is messed up $\endgroup$ – Wolfy Nov 2 '16 at 15:57
5
$\begingroup$

Let define $\mathbb{Q}$ and $\mathbb{P}$ two equivalent probabilities on a filtered space $(\Omega,(\mathcal{F}_t)_{t\geq 0})$

Let define $Z_T=\frac{d\mathbb{Q}}{d\mathbb{P}}$ restricted to $\mathcal{F}_T$ measurable events.

It means that for $X_T$ being $\mathcal{F}_T$ measurable we have: $$\mathbb{E}^{\mathbb{Q}}[X_T] = \mathbb{E}^{\mathbb{P}}\left[Z_TX_T\right]$$


Let $t\leq T$.

We want to define the change of probability measure on $\mathcal{F}_t$. i.e we want to find $Z_t$ being $\mathcal{F}_t$ measurable such that for $X_t$ being $\mathbb{F}_t$ measurable, we have:

$$\mathbb{E}^{\mathbb{Q}}[X_t]= \mathbb{E}^{\mathbb{P}}\left[Z_tX_t\right]$$

By definition of $Z_T$, and since $X_t$ is also $\mathcal{F}_T$ measurable, we have: $$\mathbb{E}^{\mathbb{Q}}[X_t]= \mathbb{E}^{\mathbb{P}}\left[Z_TX_t\right]$$

i.e

for any $X_t$ being $\mathcal{F}_t$ measurable we have $Z_t$ being $\mathcal{F}_t$ measurable such that:

$$\mathbb{E}^{\mathbb{P}}[Z_T X_t]=\mathbb{E}^{\mathbb{P}}[Z_t X_t]$$

so $Z_t = \mathbb{E}^{\mathbb{P}}[Z_T|\mathcal{F}_t]$ by definition of conditional expectation.


Let $Y_T$ being $\mathcal{F}_T$ measurable, then we want to compute $\mathbb{E}^{\mathbb{Q}}[Y_T|\mathcal{F}_t]$.

We denote $Y_t = \mathbb{E}^{\mathbb{Q}}[Y_T|\mathcal{F}_t]$

We look for $Y_t$ such that for any $X_t$ being $\mathcal{F}_t$ measurable, we have :

$$\mathbb{E}^{\mathbb{Q}}[Y_TX_t]=\mathbb{E}^{\mathbb{Q}}[Y_t X_t]$$

By definition of $Z_T$ we have $\mathbb{E}^{\mathbb{Q}}[Y_TX_t]=\mathbb{E}^{\mathbb{P}}[Z_TY_TX_t]$

By definition of $Z_t$ we have $\mathbb{E}^{\mathbb{Q}}[Y_tX_t]=\mathbb{E}^{\mathbb{P}}[Z_tY_tX_t]$

so we have:

$$\mathbb{E}^{\mathbb{P}}[Z_TY_TX_t]=\mathbb{E}^{\mathbb{P}}[Z_tY_tX_t]$$

and again by definition of conditional expectation, we have:

$$\mathbb{E}^{\mathbb{P}}[Z_TY_T|\mathcal{F}_t]=Z_tY_t$$

we can now conclude using the definition of $Y_t$ and $Z_t$.

$$\mathbb{E}^{\mathbb{Q}}[Y_T|\mathcal{F}_t] = \frac{\mathbb{E}^{\mathbb{P}}[Z_TY_T|\mathcal{F}_t]}{\mathbb{E}^{\mathbb{P}}[Z_T|\mathcal{F}_t]}$$

$\endgroup$
  • $\begingroup$ Very good, indeed. $\endgroup$ – Gordon Oct 26 '16 at 11:23
  • $\begingroup$ I wish this was clear for me haha $\endgroup$ – Wolfy Oct 28 '16 at 20:15

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.