1
$\begingroup$

I am trying to read Karatzas/Shreve "Methods of Mathematical Finance". In ch. 1, Definition 5.5, a cumulative income process $\Gamma(t)=\Gamma^{\mathrm{fv}}(t)+\Gamma^\mathrm{lm}(t)$ (a semimartingale under the original measure $P$) is defined to be integrable if $$ E_0 \int_0^T \frac{d \hat\Gamma_0^{\mathrm{fv}}(u)}{S_0(u)} < \infty, E_0 \int_0^T\frac{d \langle \Gamma_0^{\mathrm{lm}} \rangle (u)}{S_0^2(u)} < \infty $$ where $E_0$ denotes expectation with respect to $d P_0 = Z_0\, dP, Z_0 =\exp \left[ - \int_0^t \theta'(s) dW(s) - \frac 1 2 \int_0^t \| \theta(s) \|^2 ds \right]$, and $\theta (\cdot )$ being the market price of risk process.

$d \hat\Gamma^{\mathrm{fv}}(u) = \vert d \Gamma^{\mathrm{fv}} (u)\vert $ denotes the absolute variation of the finite variation part of $\Gamma$. Moreover, the authors remind that, with respect to the new measure $P_0$, $$ d \Gamma_0^{\mathrm{fv}}(t) = d \Gamma^{\mathrm{fv}}(t) - \theta'(t) d \langle \Gamma^{\mathrm{lm}}, W \rangle (t) . $$

Now the authors claim in Remark 5.8 that with $H_0(t ) := Z_0 (t) / S_0 (t)$ (the state price density process), the integrability conditions for the cumulative income process can be rewritten as $$ E \int_0^T H_0(u) d \hat\Gamma^{\mathrm{fv}}(u) < \infty , $$ if $\Gamma^{\mathrm{lm}}(\cdot ) \equiv 0$.

I don't see how this is equivalent, and I don't know if it's supposed to be equivalent to both conditions or only the first one. I tried to rewrite the proposed condition. If I'm not mistaken,

$$ E \int_0^T H_0(u) d \hat\Gamma^{\mathrm{fv}}(u) = E_0 \left[ \int_0^T \frac{\exp \left( \int_{(u,T]} \theta'(s) dW(s) + \frac 1 2 \int_{(u,T]} \| \theta(s) \|^2 d s \right) }{S_0(u)} d \hat\Gamma_0^{\mathrm{fv}}(u) \right] $$

Now what do I do with the numerator?

$\endgroup$
1
$\begingroup$

Sketch of partial solution: Let $\pi_n := \lbrace 0 = \tau^n_0\leq \tau_1^n \leq \dots \leq \tau_{m_n}^n=T\rbrace$ be a sequence of partitions s.t. $\operatorname{mesh}(\pi_n)\to 0$ a.s. We have the following limit (strictly speaking, we have to choose a subsequence, plus some steps need justification): $$ E \int_0^T \frac{Z_0(t)}{S_0(t)}\,d\widehat\Gamma^{\mathrm{fv}} (t) = \lim_{n\to \infty} E\sum_{i = 1}^{m_n - 1} \vert \widehat\Gamma^{\mathrm{fv}}\vert\,([\![\tau^n_{i}, \tau^n_{i+1} [\![) \, \frac{Z_0(\tau_i^n)}{S_0(\tau_i^n)} $$ $$ = \lim_{n\to \infty} \sum_{i = 1}^{m_n - 1} E \left[ \vert \widehat\Gamma^{\mathrm{fv}} \vert ([\![\tau^n_{i}, \tau^n_{i+1} [\![)\,\frac{Z_0(\tau_i^n)}{S_0(\tau_i^n)} \right] = \lim_{n\to \infty} \sum_i E_0 \left[ \vert \widehat\Gamma^{\mathrm{fv}} \vert ([\![\tau^n_{i}, \tau^n_{i+1} [\![)\,\frac{Z_0(\tau_i^n)}{Z_0(\tau_{i+1}^n)S_0(\tau_i^n)} \right] $$ $$ = \lim E_0 \sum \dots = E_0 \int_0^T \frac{d \widehat \Gamma^{\mathrm fv} (t)}{S_0(t)} $$ Now, $$ E_0 \int_0^T \frac{d \widehat \Gamma^{\mathrm fv} (t)}{S_0(t)}\leq E_0 \int_0^T \frac{d \widehat\Gamma_0^{\mathrm{fv}} (t)}{S_0(t)} +E_0 \int_0^T \vert \theta (t) \vert \frac{d \vert \langle \Gamma^{\mathrm{lm}}, W \rangle \vert }{S_0(t)} \leq E_0 \int_0^T \frac{d \widehat\Gamma_0^{\mathrm{fv}} (t)}{S_0(t)} +\left( \int_0^T \| \theta (t) \|^2\,dt \right)^{1/2} \left( E_0 \int_0^T \frac{d \langle \Gamma^{\mathrm{lm}} \rangle }{S_0(t)^2} \right)^{1/2} $$ Here, the last step follows from the Kunita-Watanabe inequality (followed by Cauchy-Schwarz for $L^2 (P_0)$). In the same way, one can show $E_0 \int \dots \geq E_0 \int \dots - E_0 \int \dots \geq E_0 \int\dots - T^{1/2} (E_0 \int \dots )^{1/2}$. This shows that if one assumes $E_0 \int_0^T \frac{d \langle \Gamma^{\mathrm{lm}} \rangle }{S_0(t)^2} < \infty$, the other two conditions are equivalent. (EDIT: Note that $\langle \Gamma^{lm} \rangle = \langle \Gamma_0^{lm} \rangle$)

I'm still thinking about the rest.

I demand my bounty back.^^

| improve this answer | |
$\endgroup$

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.