The paper “Inferring Covariances for Probabilistic Programs” by Benjamin Kaminski, Joost-Pieter Katoen and Christoph Matheja has been accepted at the Int. Conf. on Quantitative Evaluation of SysTems (QEST) in Quebec City, Canada.  The paper shows that approximating (co)variances is computationally more difficult than approximating expected values and presents invariant-based techniques for approximating both upper and lower bounds.