2
$\begingroup$

Here is my problem

Let $H(x)=\Phi(\frac{Ax+b}{\sqrt{2}})$ where $A$, $b$ are some non-negative constant and $\Phi$ is the CDf of a standard normal distribution.

I want to find

$$\int x \, dH(x)$$

I thought maybe integration by parts which yields:

$$x H(x)-\int H(x) \, dx$$

And now I'm not sure how to proceed or if it's possible to take the integral of a CDf

$\endgroup$
0
1
$\begingroup$

Let us do some probability here. Let $Z$ denote a standard normal random variable, with CDF $\Phi$, and $X$ a random variable with CDF $H$. Then, for every $x$, $H(x)=\Phi((Ax+b)/\sqrt2)$ implies that $\mathrm P(X\leqslant x)=\mathrm P(Z\leqslant(Ax+b)/\sqrt2)=\mathrm P((\sqrt2Z-b)/A\leqslant x)$. That is, $X$ is distributed as $(\sqrt2Z-b)/A$. In particular, $Z$ is centered hence $$ \int\limits x\mathrm dH(x)=\mathrm E(X)=\mathrm E((\sqrt2Z-b)/A)=(\sqrt2\mathrm E(Z)-b)/A=-b/A. $$

$\endgroup$
5
  • $\begingroup$ Why did you change your name? $\endgroup$ – Pedro Tamaroff Apr 10 '12 at 23:13
  • $\begingroup$ @PeterT.off Why do you ask? $\endgroup$ – Did Apr 10 '12 at 23:17
  • $\begingroup$ Just Curiosity. I added T.off when I saw many users named Peter, but you seem to have done the opposite. $\endgroup$ – Pedro Tamaroff Apr 10 '12 at 23:22
  • $\begingroup$ @PeterT.off Right, but not me. :-) $\endgroup$ – Did Apr 10 '12 at 23:38
  • $\begingroup$ @Dider I'm considering using another name. Pierre seems a little less abundant. $\endgroup$ – Pedro Tamaroff Apr 11 '12 at 0:45
1
$\begingroup$

Since $H$ is an everywhere differentiable function, we have $$ \int x \, dH(x) = \int x H'(x)\,dx = \int x \Phi'\left(\frac{Ax+b}{\sqrt{2}}\right) \frac{A}{\sqrt{2}}\, dx. $$ Let $u=\dfrac{Ax+b}{\sqrt{2}}$, so that $du = \dfrac{A\,dx}{\sqrt{2}}$ and $x= \dfrac{\sqrt{2}\; u - b}{A}$. Then the integral becomes $$ \int \frac{\sqrt{2}\;u-b}{A} \Phi'(u) \, du. $$ Then make this into $$ \frac{\sqrt{2}}{A} \int u \Phi'(u)\,du - \frac{b}{A} \int \Phi'(u)\,du. $$ The second integral above becomes $\Phi(u)+\text{constant}$. The first is $$ \int u \Phi'(u)\,du = \int u e^{-u^2/2} \, du = \int e^{-w}\,dw = -e^{-w}+\text{constant}. $$ Then convert back to an expression in $u$, then back to an expression in $x$.

If you had in mind a definite integral from $-\infty$ to $\infty$, then $$ \int_{-\infty}^\infty \Phi'(u)\,du = 1 $$ and $$ \int_{-\infty}^\infty u \Phi'(u) \, du = 0. $$ The second integral is $0$ because you're integrating an odd function over an interval that is symmetric about $0$. The first is $1$ since you're integrating a probability density function.

$\endgroup$

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.