1
$\begingroup$

I'm trying to find $\int_{\frac{a-b}{B}}^\infty\Phi\left(tA+ABx\right)\phi(x)\,dx$ where

$A = \frac{\sqrt{\gamma_{3}+\sigma_3^2}}{\gamma_{3}},\ B = \frac{\gamma_{2}}{\sqrt{\gamma_{2}+\sigma_{2}^2}},\ t=b-c$.

This integral arrives from trying the find $E[\max\{X,Y,a\}]$, where $X\sim N(b,B^2) $, $Y\sim N(c,\frac{1}{A^2})$, independent. I've tried differentiating w.r.t. $A$ but after solving the integral, I can't integrate back w.r.t. $A$. This is trying to follow the strategy from How can I calculate $\int^{\infty}_{-\infty}\Phi\left(\frac{w-a}{b}\right)\phi(w)\,\mathrm dw$.

Is there a closed form solution for this?

I realise that it is similar to rainbow options where they calculate $\max\{X,Y,k\}$ but $X$ and $Y$ follow lognormal distribution. The closed formed solution for that involves the bivariate normal distribution.

$\endgroup$
2
$\begingroup$

You have $tA + ABx$, while the answer you linked is differentiating with $\frac{x-\mu}{\sigma}$.

The answer you linked provides the solution (when integratingn over $-\infty, \infty$)

$$\int_{-\infty}^\infty \Phi \left( \frac{x-\mu}{\sigma} \right) \phi (x) dx = \Phi \left( \frac{-\mu}{\sqrt{1+\sigma^2}} \right)$$

It seems to me you can just transform your integral into that form and apply the answer they provided.

Write

$$tA + ABx = ABx - (-tA) = \frac{x - (-tA)(AB)^{-1}}{(AB)^{-1}} = \frac{x-\mu}{\sigma}$$

Where

$$\mu = (-tA)(AB)^{-1}$$

$$\sigma = (AB)^{-1} = \frac{1}{AB}$$

Then you now have an integral of the form used in that question. So following the logic of that answer, the integral should be

$$\Phi \left( \frac{tA(AB)^{-1}}{\sqrt{1+(AB)^{-2}}} \right) = \Phi \left( \frac{tA}{\sqrt{(AB)^2+1}} \right)$$

Now, in your example you have a lower bound that is not $-\infty$. So, using the transformation above, you should be able to replicate the steps taken in that answer (or similar, as their are multiple on this site) with your lower bound, provided a closed form solution exists.

Perhaps you can run some basic simulations in R or Python to verify this is correct.

$\endgroup$
  • $\begingroup$ Hey! Thanks for the reply! I have a few questions regarding your solution. Are you allowed to differentiate with respect to $\mu$ if $\mu$ is a function of $\sigma$? I'm not sure if that is possible or correct. Since you are treating $\sigma$ as a constant. Also, the solution provided seems to be integrating from $-\infty$ to $\infty$ which is not as the same as mine. $\endgroup$ – icecream_sandwich07 Oct 13 '18 at 10:24
  • 1
    $\begingroup$ The integral the OP is asking about has a different lower limit than the integral in the cited answer whose method is being applied here. $\endgroup$ – Dilip Sarwate Oct 13 '18 at 11:44
  • $\begingroup$ Good catch @DilipSarwate the linked answer had excluded the bounds so it confused me $\endgroup$ – Xiaomi Oct 13 '18 at 11:47

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.