Multivariate Gaussian integral of ratio of quadratic forms Given two real symmetric matrices $M,S$ is there a known answer for the Gaussian integral $\int d^Nz\frac{z^TMz}{z^TSz}$ where the integration is over N-dimensional Gaussian variable $z\sim N(\vec{0},I)$? 
This can also be written as $E_z[\frac{z^TMz}{z^TSz}]$, so it seems like a very simple expression, but I could not find any result on this.
 A: If I'm not mistaken, Magnus [1986] ("The exact moments of a ratio of quadratic forms in normal variables") has your answer.  See section 5, let s=1, for your case.
http://www.google.com/url?sa=t&source=web&cd=2&ved=0CCEQFjABahUKEwie29a07sLHAhXCNz4KHRj8Dg0&url=http%3A%2F%2Fannales.ensae.fr%2Fanciens%2Fn04%2Fvol4-05.pdf&rct=j&q=integral%20ratio%20quadratic%20forms&ei=XqfbVZ60DsLv-AGY-Lto&usg=AFQjCNH2P-MVDDyDAS0FjuvOUx1H7mrwCw
A: Per @GeoffreyEvans answer the general case of the s-th moment $E_z[(\frac{z^TMz}{z^TSz})^s]$ for $z\sim N(\mu,\Sigma)$ is discussed by [Magnus 1986]. A (complex) closed-form answer is available there, but for $s=1$, $\mu=\vec{0}$  and $\Sigma =I$ which the question targeted it simplifies to
$$\mathbb{E}\left[\frac{z^{T}Mz}{z^{T}Sz}\right]=\int_{0}^{\infty}dt\prod_{j}\frac{1}{\sqrt{1+2t\lambda_{j}}}\sum_{i}\frac{\left[U^TMM^TU\right]_{i}^{2}}{1+2t\lambda_{i}}$$
where $S=U\Lambda U^T$ for orthonormal matrix $U$ and diagonal matrix $\Lambda$ with diagonal elements $\{\lambda_i\}$.
