Gaussian conditional expectation
ID: gaussian-conditional-expectation
Let the joint random vector have a multivariate normal distribution, with expectations , covariance matrices and cross-covariance matrix . If is invertible, the conditional expectation is the affine functionand the conditional covariance matrix is .
To prove both assertions, define . Direct covariance calculation gives . Since has a multivariate normal distribution, this zero cross-covariance implies independence. The decomposition of into its affine part and therefore gives the conditional expectation and the remaining covariance matrix. This is the conditional multivariate normal distribution calculation.
If is singular, replace by its Moore-Penrose inverse . The formulas hold for in its affine probability support, ; values away from that probability support do not determine a unique version of the conditional expectation. For example, if and for a standard normal random variable , then is the all-ones matrix. Although does not exist, on the probability support .
New to topics? Read the docs here!