Bayesian network management

We formulate some general network (and risk) management problems in a Bayesian context, and point out some of the essential features. We argue and demonstrate that, when one is interested in rare events, the Bayesian and frequentist approaches can lead to very different strategies: the former typically leads to strategies which are more conservative. We also present an asymptotic formula for the predictive probability of ruin (for a random walk with positive drift) for large initial capital and large number of past observations. This is a preliminary investigation which raises many interesting questions for future research. This revised version was published online in June 2006 with corrections to the Cover Date.