3
$\begingroup$

I need to calculate the following integral

$$\int_{\mu+c}^{\infty} y\cdot \frac{1}{\sigma\sqrt{2\pi}}e^{(y-\mu-w)^2/2\sigma^2}dy$$

So essentially $y\sim N (\mu+w, \sigma^2)$ and im trying to calculate its expected value from $\mu+c$ to $\infty$

I know I will have to use integration by parts, but don't know how to proceed.

More importantly however, when I calculate the integral, will I get an expression which includes $w$? I really need the $w$ in there.

Improve this question
$\endgroup$
2
  • $\begingroup$ The result will definitely be a function of $w$. $\endgroup$
    – Glen_b
    Sep 13, 2014 at 2:08
  • $\begingroup$ The integral does not converge: presumably you intend for the argument of $\exp$ to be negated. $\endgroup$
    – whuber ♦
    Sep 13, 2014 at 21:45

2 Answers 2

Reset to default
3
$\begingroup$

Hint: setting $\sigma^2=1$ for convenience, we have that

$$\begin{align} \frac{\mathrm d}{\mathrm dy} \exp(-(y-\mu-w)^2/2) &= -(y-\mu-w)\exp(-(y-\mu-w)^2/2)\\ &= -y\exp(-(y-\mu-w)^2/2 \\ &\qquad + (\mu+w)\exp(-(y-\mu-w)^2/2 \end{align}$$ and so you should be able to write the integrand as $(\mu+w)$ times a Gaussian density less a perfect differential, and thus get the integral to work out to something involving $\Phi(\cdot)$ and an $\exp(g(w))$ term where $g(w)$ is a quadratic function of $w$.

Improve this answer
$\endgroup$
2
  • $\begingroup$ i dilip thanks for that. Are you hinting at using integration by parts? If i integrate the expression you'vve provided i will end up getting $exp^(-(y-\mu-w)^2/2)$ rather than the integral of it. Won't i? $\endgroup$
    – user55405
    Sep 14, 2014 at 3:35
  • 1
    $\begingroup$ @user55405 I am not hinting at using integration by parts. I am saying outright that $$\begin{align}\int y\exp(-(y-\mu-w)^2/2\,\mathrm dy&=-\int\frac{\mathrm d}{\mathrm dy} \exp(-(y-\mu-w)^2/2)\,\mathrm dy+\int(\mu+w)\exp(-(y-\mu-w)^2/2\,\mathrm dy\\&=-\exp(-(y-\mu-w)^2/2)+(\mu+w)\int\exp(-(y-\mu-w)^2/2\,\mathrm dy\end{align}$$ from which, when you change to definite integrals instead of antiderivatives as in the above expression, you get an exponential in $w$ from the first term and something involving the Gaussian CDF $\Phi(\cdot)$ from the second term on the right. $\endgroup$
    – Dilip Sarwate
    Sep 14, 2014 at 11:54
0
$\begingroup$

The conditional expectation for a normal random variable with mean $\tau$ and standard deviation $\sigma$ can be found from this formula, which is available, for example, here: www.actuaries.org/LIBRARY/ASTIN/vol35no1/189.pdf ‎

$$ E[X | X > q ]={\tau + {{\sigma} \phi \left({{q-\tau} \over \sigma} \right) \over {1-\Phi \left( {{q-\tau} \over \sigma}\right)} } },$$

where $\Phi$ is the CDF and $\phi$ is the PDF for the standard normal. The general formula that can be used to derive these type of expectations is given by

$$ E[X | X > q ]={ {\int_q^\infty xf(x)dx} \over {P[X>q]}}$$

Putting these together with your parameters, we have

$$E[Y | Y > \mu+c ]={ {{{\int_{\mu+c}^\infty {y \over {\sigma \sqrt{2\pi}}} e^{{-{\left ( {y-\mu-w} \right)^2}} \over {2 \sigma^2}} dy} }\over {1-\Phi \left( {{c-w} \over \sigma}\right)} }=\mu+w +\left[{\sigma {\phi \left( {c-w} \over \sigma \right) } \over {1-\Phi\left( {c-w} \over \sigma \right)}} \right] }$$

Finally, then, we can solve for the integral you want:

$${\int_{\mu+c}^\infty {y \over {\sigma \sqrt{2\pi}}} e^{{-{\left ( {y-\mu-w} \right)^2}} \over {2 \sigma^2}} dy}=\left(\mu+w \right) \left[{1-\Phi\left( {c-w} \over \sigma \right)} \right]+\sigma {\phi \left( {c-w} \over \sigma \right) }$$

Improve this answer
$\endgroup$

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.