This paper considers the stochastic cash balance problem. A dynamic simple policy (DSP) is proposed to minimise transaction costs, under a general cost structure, when the cash flows are not independently or identically distributed. The validity of the approach is demonstrated using the scenario of double exponentially distributed cash flows considered by Penttinen. A data set from a large multinational is used to demonstrate the practical application of the DSP. To provide conditional expectations of future cash flows, a time series model is developed to provide forecasts. The performance of the DSP on these cash flows achieved a 16% saving in average transaction cost (over the Penttinen model). The saving due to the forecasting model over the assumption that the cash flows exhibit a constant mean is a function of the time series structure of the relevant cash flows. For the example data, the benefit of using the bespoke forecasting model over the constant mean was only 2%.