2
$\begingroup$

I have the following equation: $ x = \frac{1-\Phi(x)}{a\phi(x)}$, where $\Phi$ is the cdf and $\phi$ is the pdf of the standard normal distribution.

How can one solve for $x$? Is there an analytical approach? Or can this only be done numerically?

These might be rather stupid questions (it's quite a while ago that I last was exposed to this), but any suggestions are appreciated. Thanks a lot!

$\endgroup$
10
  • $\begingroup$ First of all, have you tried to plot for example the 2 curves $y=a x \phi(x)$ and $y=1-\Phi(x)$ and see where they intersect in order, at least, to have an idea of where the root/roots is/are situated? A very coarse sketch shows for example that there are no negative roots $\endgroup$
    – Jean Marie
    Jun 21 '17 at 15:28
  • $\begingroup$ I am almost certain that there are no analytical approaches yielding an exact solution. $\endgroup$
    – Jean Marie
    Jun 21 '17 at 15:52
  • 1
    $\begingroup$ @JeanMarie" How fast time goes by ... $\endgroup$ Mar 1 '20 at 13:46
  • 1
    $\begingroup$ @callculus As said by Heinrich Heine: "Ich hab' euch im besten Juli verlassen / Und find' euch wieder im Januar ; / Ihr sasset damals so recht in der Hitze / Jetzt sei ihr gekhült und kalt sogar..." (Die Heimkehr, 1823) $\endgroup$
    – Jean Marie
    Mar 1 '20 at 14:18
  • 1
    $\begingroup$ @JeanMarie Wow. I didn´t expect that. C´est incroyable. I´ve to admit I have no idea about a similar lyric in french. It gives my the opportunity to concern with german literature and maybe with french literature-but translated. $\endgroup$ Mar 1 '20 at 14:55
2
$\begingroup$

Use the series expansions. You can see here how they can be derived. Let $a=1$.

The equation is $x\cdot \phi(x)=1-\Phi(x)$. The equation can be multipied by $\sqrt{2\cdot \pi}$. It becomes

$$x\cdot e^{-x^2/2}=\sqrt{2\cdot \pi}-\left(0.5\cdot \sqrt{2\cdot \pi}+\int_0^x e^{-t^2/2} \,dt \right)$$

Using the series expansion the approximated equation is

$$x\cdot \left(1-\frac{x^2}{2}+\frac{x^4}{8}-\frac{x^6}{48}+\frac{x^8}{384}\right)$$ $$=\sqrt{2\cdot \pi}-\left(0.5\cdot \sqrt{2\cdot \pi}+\frac{x}{1\cdot 1}-\frac{x^3}{2\cdot 3}+\frac{x^5}{8\cdot 5}-\frac{x^7}{48\cdot 7}+\frac{x^9}{384\cdot 9}\right)$$

This equation can be solved with Wolfram alpha. The result is $\boxed{x=0.751781}$

To check the result use calculators for the pdf and the cdf. I get

$$x\cdot \phi(x)=0.751781\cdot 0.30073521=0.226087...\approx 22.609\% $$

$$1-\Phi(x)=1-0.77391=0.22609=22.609\%$$

It looks like that the approximation is fine.

$\endgroup$
2
  • $\begingroup$ Thank you very much for your fast help. This is very interesting, and I will have a closer look at this. Quick follow-up question: The factors in the series are $1,2,8,48,384,...$ rather than $1,2,8,48,484,...$, or am I confused? (The context of my original question is a simple economic model, in which x is quantity in a profit function.) $\endgroup$
    – Carl
    Jun 22 '17 at 7:12
  • $\begingroup$ @Carl No, you´re not confused. Good catch. Luckely the typo was in the last summand. So it doesn´t change the result much. I´ve edited the answer, including the WA-Link. .If you find the answer helpful you can accept it $\endgroup$ Jun 22 '17 at 10:09

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.