AN EMPIRICAL ANALYSIS OF ALTERNATIVE STRATEGIES FOR MANAGING THE CASH BALANCE

JONATHAN ALAN SCOTT, Purdue University

Abstract

The daily cash balance control decision is one of the most important functions of the cash manager. A cash forecast is required because the manager does not know with certainty what the closing cash balance for the day will be. With this forecast the manager must decide how to adjust the cash balance to its target level before the day's end. This thesis will examine the cash balance control decision using data from an actual U.S. corporation by (1) evaluating various statistical forecasting methods and (2) analyzing the robustness of several mathematical decision models for adjusting the cash balance to its target level. The forecasting analysis will consider the application of dummy variable regression, time series, and adaptive filtering statistical models to the cash flow data. An evaluation of the models will be made from unconditional forecasts and will also include the costs of maintaining the model. For these data, the dummy variable regression model provides the most accurate forecast, although all the models exhibit similar forecast error patterns. When the cost of operating and maintaining the models is considered, the adaptive filtering model is the best. The problem of how much to adjust the cash balance before the day's end has received much theoretical treatment in the academic literature, but virtually no empirical testing of the various models promulgated through the years has been done. Two types of formal models have been developed: (1) control-limit models which assume net cash flows are randomly distributed and therefore cannot be forecasted effectively; and (2) linear programming models which assume cash flows can be forecasted perfectly. Obviously the real world lies between these two extremes, so that any test of these models will be analyzing the robustness of the model's assumptions when applied to actual data. This research will use two types of control-limit models and a linear programming model to construct a simulated cash balance decision strategy. The linear programming model will use unconditional forecasts made from a dummy variable regression forecasting model. An average daily profit will be computed for each model to be compared with the actual figure. The somewhat surprising result of the simulations is that the linear programming model, even with its imperfect forecasts, generates an average daily profit which far exceeds the control-limit models' average daily profit, as well as the actual figure. These results suggest, in the context of these data, that more attention should be paid to the statistical forecasting problem and the linear programming approach to cash balance control than to the control-limit approach.

Degree

Ph.D.

Subject Area

Finance

COinS