6
$\begingroup$

Is there a closed form solution of the following price formula? Assuming $dS_t=rSdt+\sigma S_t dW_t$ under the Q dynamics

$e^{-r(T-t)}\mathbb{E}_t^\mathcal{Q}[(\frac{(\int_0^T S_u du)}{T})^2]$

I know that the integral of a geometric brownian motion has no nice distribution, but is it the same with the square of the integral?

$\endgroup$
3
$\begingroup$

Yes it is known in closed form. See https://www.rocq.inria.fr/mathfi/Premia/free-version/doc/premia-doc/pdf_html/asian_doc/asian_doc.html section 5.1 which references an older Geman-Yor paper.

$\endgroup$
  • $\begingroup$ This gives a closed form for the Laplace transform of the Asian call option. Computing the second moment is a much simpler task. $\endgroup$ – AFK Mar 17 '15 at 18:59
  • $\begingroup$ Agree the moment calc is simpler than laplace transform. The reference I linked gives both; equation for moments is (11) in section 5.1. $\endgroup$ – q.t.f. Mar 17 '15 at 19:10
  • $\begingroup$ My bad. Good link. $\endgroup$ – AFK Mar 17 '15 at 19:40
2
$\begingroup$

A few tips. First note that $e^{-rt}S_t$ is a martingale. So make it appear and then integrate by part to rewrite $\int S_u du$ as a stochastic integral. Finally use the Ito isometry property.

$\endgroup$
2
$\begingroup$

The $P$ dynamics of the underlying asset are: \begin{align*} dS=S(\mu dt+\sigma dB_t) \end{align*} That has the following solution under the $\mathcal{Q}$ dynamics: \begin{align*} S_t=S_0 e^{(r-\frac{\sigma^2}{2})t+\sigma W_t} \end{align*} Where $W_t$ is the equivalent martingale with respect to the original geometric brownian motion. Define $Y_t=\int_0^t S_u du$, then according to Feynman-Kac the value of the replicating portfolio is given by \begin{align*} V_t&=e^{-r(T-t)}\frac{1}{T^2}\mathbb{E}_t^\mathcal{Q}[(Y_t+\int_t^T S_udu)^2)\\ &=\frac{e^{-r(T-t)}}{T^2} \mathbb{E}_t^\mathcal{Q}[Y_t^2 + 2Y_t \int_t^T S_u du + (\int_t^T S_u du)^2]\\ \end{align*} So: \begin{align*} \mathbb{E}_t^\mathcal{Q}[\int_t^T S_u du]&=S_t \mathbb{E}_t^\mathcal{Q}[\int_t^T \frac{S_u}{S_t} du]\\ &=S_t \int_t^T e^{(r-\frac{\sigma^2}{2})(u-t)}\mathbb{E}^\mathcal{Q}[e^{\sigma (W_u-W_t)}] du\\ \end{align*} Since $W_u-W_t$ follows $\mathcal{N}(0, u-t)$ under $Q$. According to the m.g.f. \begin{align*} &=S_t \int_t^T e^{r(u-t)}du\\ &=\frac{S_t}{r}(e^{r(T-t)}-1)\\ &=\frac{x}{r}(e^{r(T-t)}-1) \end{align*} Lets compute now the expectation of the square of the integral: \begin{align*} \mathbb{E}_t^\mathcal{Q}[(\int_t^T S_u du)^2]&=S_t^2 \mathbb{E}_t^\mathcal{Q}[\int_t^T \frac{S_u}{S_t}du \int_t^T \frac{S_v}{S_t}dv ]\\ &= S_t^2 \int_t^T \int_t^T \mathbb{E}^\mathcal{Q}[\frac{S_u}{S_t} \frac{S_v}{S_t}]dv du\\ \end{align*} Lets focus on: $\mathbb{E}^\mathcal{Q}[\frac{S_u}{S_t} \frac{S_v}{S_t}]$, lets assum $t \leq v \leq u$ \begin{align*} \mathbb{E}^\mathcal{Q}[\frac{S_u}{S_t} \frac{S_v}{S_t}&=\mathbb{E}^\mathcal{Q}[e^{(r-\frac{\sigma^2}{2})(u-t)+\sigma(W_u-W_t))}e^{(r-\frac{\sigma^2}{2})(v-t)+\sigma(W_v-W_t))}]\\ &=e^{(r-\frac{\sigma^2}{2})(u-t)+(r-\frac{\sigma^2}{2})(v-t)}\mathbb{E}^\mathcal{Q}[e^{\sigma(W_u-W_v)+2\sigma (W_v-W_t)}]\\ &=e^{(r-\frac{\sigma^2}{2})(u-t)+(r-\frac{\sigma^2}{2})(v-t)}\mathbb{E}^\mathcal{Q}[e^{\sigma(W_u-W_v)}]\mathbb{E}^\mathcal{Q}[e^{2\sigma(W_u-W_t)}]\\ &=e^{(r-\frac{\sigma^2}{2})(u-t)+(r-\frac{\sigma^2}{2})(v-t)}e^{\frac{\sigma^2}{2}(u-v)}e^{2\sigma^2(v-t)}\\ &= e^{ur}e^{u(r+\sigma^2)}e^{-t(2r+\sigma^2)} \end{align*} So: \begin{align*} \mathbb{E}_t^\mathcal{Q}[(\int_t^T S_u du)^2]&=2S_t^2 \int_t^T \int_t^u e^{ur}e^{u(r+\sigma^2)}e^{-t(2r+\sigma^2)}dv du\\ &= \frac{2x^2}{r+\sigma^2}(\frac{1}{2r+\sigma^2}e^{(2r+\sigma^2)(T-t)}-\frac{1}{r}e^{r(T-t)}+\frac{r+\sigma^2}{(2r+\sigma^2)r}) \end{align*} Mixing everything together we get: \begin{align*} V(t,x,y)&=\frac{y^2}{T^2}e^{rt-rT}+\frac{1}{T^2}\frac{2xy}{r}(1-e^{rt-rT})\\ &+ \frac{1}{T^2}\frac{2x^2}{r+\sigma^2}(\frac{1}{2r+\sigma^2}e^{(r+\sigma^2)(T-t)}-\frac{1}{r}+\frac{r+\sigma^2}{(2r+\sigma^2)r}e^{-r(T-t)}) \end{align*}

$\endgroup$
  • $\begingroup$ Are you sure that your derivation is correct? It doesn't look so to me. $\endgroup$ – SmallChess Mar 19 '15 at 23:24
  • $\begingroup$ I can check it, but it seems ok to me. It satisfies the SDE $V_t + rxV_x+\frac{1}{2}\sigma^2 x^2 V_{xx} +x V_y=rV$ and boundary conditions. Why it does not look correct to you? $\endgroup$ – Juan Imbet Mar 20 '15 at 23:41

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.