After reading Larry Wassermans blog on the difference between Bayesian and Frequentist inference I started to appreciate that frequency guarantees can be desirable regardless of the inference method that you choose, since some Bayesian inference methods have frequency guarantees.

With that said, assume that I am developing a prediction model for a Business application that will perform a large number of predictions over the next year. Furthermore, the Business requires a guarantee of the performance of this prediction model in order to manage risk.

Is a guarantee of the performance of a (Bayesian or Frequentist) prediction model a frequency guarantee of that prediction model? If all guarantees of performance are frequency guarantees should frequency guarantees not always be desirable for both Bayesians and Frequentists in this Business context?

As an example: if I am learning from a class of (classification) prediction models with finite VC dimension and more data than the corresponding sample complexity is available I have guarantees on the performance of the prediction model. However, these results from statistical learning theory are derived from the risk which involves an integral over the data which is fundamentaly frequentist, i.e. a frequency guarantee. My intuition is that all guarantees about performance (future data) will involve an integral over the data and therefore Bayesians require frequency guarantees to guarantee performance, but I might be mistaken?