The management of bank balance sheets is a major determinant of banks' long-term performance and sustainability. Therefore, it is one of the most critical problems in bank management. Using autoregressive processes for stochastic deposit flows and credit losses, and vector autoregressive processes for interest rates, we develop a simulation-optimization method that devises balance sheet policy functions based on state variables such as interest rates or leverage levels. Performance analysis on an independent testing set shows that the algorithm outperforms other established methodologies, delivering an increase in the return-to-risk ratio of up to 68% when compared to optimizing time-independent static allocations and an increase greater than 593% when compared to equal-weight and 60/40 policies. In addition to performance, we emphasize interpretability, developing an algorithm that allows the analysis of the parametric allocations and interpretation, and can thus be used in practice to support banks' senior management dynamic decisions.