2

Suppose we have the density and distribution of the standard normal. How can one calculate the integral:

$\int_{-\infty}^{\infty} \Phi (a + bX) \phi (c + eX) dx$

Note this is not included in the Wikipedia list of integrals of Gaussian functions.

Further, this is not the same as How can I calculate $\int^{\infty}_{-\infty}\Phi\left(\frac{w-a}{b}\right)\phi(w)\,\mathrm dw$ because that solution only works for a standard normal second term, i.e. $\phi (X)$ whereas this problem includes a coefficient and additional term. One cannot put into the same form as that case because $c$ and $e$ cannot be brought out of the normal density due to their involvement in the exponential component.

Community
  • 1
SG2013
  • 31
  • 4
  • 1
    Not sure if this helps, but we know that $$\int_{-\infty}^{\infty}\Phi(a+bX) \phi( X ) dX=\Phi\left(\frac{a}{\sqrt{1+b^2}}\right)$$ –  Nov 11 '15 at 05:29
  • @ZERO Thanks, I am familiar with that integral and have seen short proofs on here, but unfortunately dealing with the second linear term is still unclear. – SG2013 Nov 11 '15 at 05:40
  • Are you sure it has an analytical closed form? – Zhanxiong Nov 11 '15 at 06:01
  • It was suggested to me that it does, but I am not sure. – SG2013 Nov 11 '15 at 06:12
  • I think that if the OP doesn't see how the other one answers this one (as was already established), then it's not really a dupe, since the question needs as an answer the explanation that connects one to the other. – Glen_b Nov 11 '15 at 22:13
  • 1
    @Glen_b In which case it really ought to be migrated to [math.se], because it's about changing variables in integration. :-) – whuber Nov 12 '15 at 00:03
  • @whuber I am in two minds about that; there's an extent to which the context means it's on topic (otherwise the original arguably should be migrated there as well; it's simply an integration problem). And on a similar basis ("it's really a mathematics question - it's just integration/differentiation/matrix algebra/change of variable...") a lot of the questions here would migrate. There's not a very clear dividing line about when it really counts as sufficiently statistical in context to count here. What I think just tips it over here is that the answer on which it relies is here. – Glen_b Nov 12 '15 at 00:17

1 Answers1

7

But if you know the form discussed in comments

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

it's quite straightforward!

Just do the substitution $Z = c+eX$ (don't forget the Jacobian), and use the fact there that you already know (but in $z$ rather than $x$).

Glen_b
  • 257,508
  • 32
  • 553
  • 939