15
$\begingroup$

As title. Can anyone supply a simple proof that

$$x \Phi(x) + \Phi'(x) \geq 0 \quad \forall x\in\mathbb{R}$$

where $\Phi$ is the standard normal CDF, i.e.

$$\Phi(x) = \int_{-\infty}^x \frac{1}{\sqrt{2\pi}} e^{-y^2/2} {\rm d} y$$


I have so far:

Defining $f(x) = x \Phi(x) + \Phi'(x)$ we get

$$ \begin{align} f'(x) & = \Phi(x) + x \Phi'(x) + \Phi''(x) \\ & = \Phi(x) + x\Phi'(x) - x\Phi'(x) \\ & = \Phi(x) \\ & >0 \end{align}$$

so it seems that if we can show

$$\lim_{x\to-\infty} f(x) = 0$$

then we have our proof - am I correct?

Clearly $f$ is the sum of two terms which tend to zero, so maybe I have all the machinery I require, and I just need to connect the parts in the right way! Assistance will be gratefully received.


In case anyone is interested in where this question comes from:

Bachelier's formula for an option struck at $K$ with time $T$ until maturity, with volatility $\sigma>0$ and current asset price $S$ is given by

$$V(S) = (S - K) \Phi\left( \frac{S-K}{\sigma S \sqrt{T}} \right) + \sigma S \sqrt{T} \Phi' \left( \frac{S-K}{\sigma S \sqrt{T}} \right) $$

Working in time units where $\sigma S\sqrt{T} = 1$ and letting $x=S-K$, we have

$$V(x) = x \Phi(x) + \Phi'(x)$$

and I wanted a simple proof that $V(x)>0$ $\forall x$, i.e. an option always has positive value under Bachelier's model.

$\endgroup$
  • 3
    $\begingroup$ $\Phi^\prime(x)$ is always $> 0$, and $x \Phi(x) \geq 0$ if $x \geq 0$, so your only worry is negative $x$... $\endgroup$ – J. M. is a poor mathematician Sep 23 '11 at 14:20
  • $\begingroup$ @SrivatsanNarayanan Yes, I just realized that and deleted my comment. Sorry $\endgroup$ – Sasha Sep 23 '11 at 14:26
  • $\begingroup$ $\Phi''(x) \ne -x\Phi(x)$. Maybe you're misremembering $\varphi'(x) = -x\varphi(x)$, where $\varphi=\Phi'$. $\endgroup$ – Michael Hardy Sep 23 '11 at 14:30
  • 4
    $\begingroup$ A search for "Miller's inequality" lists this page as the first relevant result :) $\endgroup$ – Chris Taylor Sep 23 '11 at 15:01
  • 2
    $\begingroup$ @Chris: that means it is all the more important that we check that what we post is accurate, for the benefit of future searchers. ;) $\endgroup$ – J. M. is a poor mathematician Sep 23 '11 at 15:22
22
$\begingroup$

Writing $\phi(x) = (2\pi)^{-1/2}\exp(-x^2/2)$ for $\Phi^{\prime}(x)$ $$ x\Phi(x) + \phi(x) = \int_{-\infty}^x x\phi(t)\mathrm dt + \phi(x) \geq \int_{-\infty}^x t\phi(t)\mathrm dt + \phi(x) = -\phi(t)\biggr\vert_{-\infty}^x + \phi(x) = 0 $$

$\endgroup$
  • 3
    $\begingroup$ Wow! ${}{}{}{}{}$ $\endgroup$ – Srivatsan Sep 23 '11 at 16:50
  • 2
    $\begingroup$ Thanks. The OP asked for a simple proof and this seemed to fit the bill. I had posted a different and more long-winded Answer earlier, but deleted it in view of the some of the comments that appeared while I was writing the Answer out. Then I came up with this one and so I revised my Answer and undeleted it. $\endgroup$ – Dilip Sarwate Sep 23 '11 at 16:55
  • $\begingroup$ Excellent - thanks a lot! This was just what I needed. $\endgroup$ – Chris Taylor Sep 23 '11 at 19:44
  • $\begingroup$ @Srivatsan: Yeah. Why didn't I read this before answering? $\endgroup$ – robjohn Sep 23 '11 at 22:23
6
$\begingroup$

As was pointed out by J.M in the comment, our worry is to show the inequality for $x<0$. We will begin with the following observations:

$$\Phi''(x)\geq 0,\:\text{ in }(-\infty,0).$$ Indeed $$\Phi''(x)=-xe^{-x^2/2}>0,\quad\forall x< 0.$$

Now, consider the function $$g(x)=-e^{-x^2/2}\Phi(x).$$ We rearrange the factors to obtain $$g(x)=-\sqrt{2\pi}\frac{1}{\sqrt{2\pi}}e^{-x^2/2}\Phi(x)=-\sqrt{2\pi}\Phi'(x)\Phi(x).$$ Moreover if we pick $z<x<0$, by our previous observations and the fact that both $\Phi'(x)$ and $\Phi(x)$ are greater than $0$ we can conclude $$g(x)-g(z)=-\sqrt{2\pi}(\Phi'(x)\Phi(x)-\Phi'(z)\Phi(z))<0.$$ But then $g$ is strictly decreasing on $(-\infty,0).$ Hence we can write $g'(x)<0.$ This relation translates in $$-(x\Phi(x)+\Phi'(x))e^{-x^2/2}<0.$$ Divide out by $-e^{-x^2/2}$ to obtain, for any $x<0$, $$x\Phi(x)+\Phi'(x)>0.$$ Hope this is correct.

$\endgroup$
4
$\begingroup$

I will push the OP's attempt to a complete proof. If $f(x) = x\Phi(x) + \Phi'(x)$, then, as the OP notes, $f'(x) \geq 0$ for all $x$; i.e., $f$ is monotonically increasing. So it suffices to show that $f(x) \to 0$ as $x \to -\infty$, which implies that $f(x) \geq 0$ for $x \in \mathbb R$, since: $$ \lim_{u \to -\infty} f(u) \leq f(x). $$

Now, to show $f(x) \to 0$ as $x \to -\infty$, it is enough to show that $x\Phi(x) \to 0$ and $\Phi'(x) \to 0$ as $x \to -\infty$.

  1. $\Phi'(x) = \exp\left(-\frac{1}{2} |x|^2 \right)$ clearly approaches $0$ as $x \to -\infty$.
  2. For $x \leq -2$, let $y = |x| \geq 2$. $$ |x \Phi(x)| = y \int_{-\infty}^x e^{-t^2/2} dt = y \int_{y}^\infty e^{-t^2/2} \leq y \int_{y}^\infty e^{-t} dt = y e^{-y} , $$ which approaches $0$ as $x \to -\infty$.

So we are done.


Notes on the name. The quantity $$ R(x) = e^{x^2/2} \int_x^{\infty} e^{-t^2/2} dt $$ is called Mill's ratio (see, for e.g., http://www.jstor.org/stable/2236360). I have also seen the inequality $$ \int_{x}^\infty e^{-t^2/2} dt \leq \frac{e^{-x^2/2}}{x} \ \ \ \ \ (x > 0), $$ sometimes referred to as Mill's inequality (see Theorem 6 in this lecture notes: http://www.stat.cmu.edu/~larry/=stat705/N3.pdf). Notice that by a change of variables $x \to -x$, this inequality can be seen to be equivalent to the one in the question.

$\endgroup$
  • $\begingroup$ Thanks, especially for the relationship with Mill's inequality. $\endgroup$ – Chris Taylor Sep 23 '11 at 19:46
  • $\begingroup$ I proved the equivalent inequality for $x<0$ in $(2)$ in my answer. $\endgroup$ – robjohn Sep 23 '11 at 21:44
  • $\begingroup$ (+1) Here is a simple proof of the inequality you state in your addendum. L'Hopital shows that it is asymptotically tight as $x \to \infty$. This inequality (and much more) was known to Laplace. $\endgroup$ – cardinal Sep 23 '11 at 22:05
2
$\begingroup$

I will concentrate at $\lim_{x \to -\infty} f(x) = 0$, where $f(x) = x \Phi(x) + \Phi'(x)$.

Consider $$ g(x) = \sqrt{2 \pi} f(-x) = \mathrm{e}^{-\frac{x^2}{2}} - x \int_{x}^\infty \mathrm{e}^{-\frac{y^2}{2}} \mathrm{d} y \qquad \text{for} \qquad x > 0. $$ Then $$ g(x) = \mathrm{e}^{-\frac{x^2}{2}} - x \int_{\frac{x^2}{2}}^\infty \mathrm{e}^{-t} \frac{\mathrm{d} t}{\sqrt{2t}} \, > \, \mathrm{e}^{-\frac{x^2}{2}} - \int_{\frac{x^2}{2}}^\infty \mathrm{e}^{-t} \mathrm{d} t = 0 $$ where $ x \int_{\frac{x^2}{2}}^\infty \mathrm{e}^{-t} \frac{\mathrm{d} t}{\sqrt{2t}} < \frac{x}{\sqrt{2 \frac{x^2}{2}}} \int_{\frac{x^2}{2}}^\infty \mathrm{e}^{-t} \mathrm{d} t = \int_{\frac{x^2}{2}}^\infty \mathrm{e}^{-t} \mathrm{d} t = \mathrm{e}^{-\frac{x^2}{2}}$ for $x>0$.

On the other hand $g(x) < \mathrm{e}^{-\frac{x^2}{2}}$ by definition, so $\lim_{x \to \infty} g(x) = \lim_{x \to \infty} \sqrt{2 \pi} f(-x)$ vanishes being sandwiched between 0 and $\mathrm{e}^{-\frac{x^2}{2}}$ which also tends to zero for large $x$.

$\endgroup$
1
$\begingroup$

Taking the derivative of $\Phi(x)$, $$ \Phi'(x)=\frac{1}{\sqrt{2\pi}}e^{-x^2/2}\tag{1} $$ For $x\ge0$, everything is non-negative, so $x\Phi(x)+\Phi'(x)\ge0$.

For $x<0$, $$ \begin{align} \Phi(x) &=\frac{1}{\sqrt{2\pi}}\int_{-\infty}^xe^{-t^2/2}\;\mathrm{d}t\\ &\le\frac{1}{\sqrt{2\pi}}\int_{-\infty}^xe^{-t^2/2}\frac{t}{x}\;\mathrm{d}t\\ &=-\frac{1}{x}\frac{1}{\sqrt{2\pi}}e^{-x^2/2}\\ &=-\frac{1}{x}\Phi'(x)\tag{2} \end{align} $$ Therefore, we also get $x\Phi(x)+\Phi'(x)\ge0$.

$\endgroup$
  • $\begingroup$ I just reviewed my answer and $(2)$ proves the whole thing for $x<0$. Doh! I could remove the rest, but then I think it would be the same as Srivatsan's answer. $\endgroup$ – robjohn Sep 23 '11 at 21:56
  • $\begingroup$ I removed the rest anyway. $\endgroup$ – robjohn Sep 23 '11 at 22:05

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.