Data-driven decision-making and its application to the corporate cash management problem

Author: Francisco Salas Molina
Advisor: Francisco Martin, Juan A. Rodríguez-Aguilar
Year: 2017

This thesis investigates the cash management problem from a multidimensional perspective. Cash management focuses on finding the balance between cash holdings and short-term investments. Typically, cash managers make decisions based usually on a firm’s optimal cash balance for operational and precautionary purposes. We here explore the opportunities for improved decision-making derived from modeling cash flow uncertainty with the help of data-driven procedures within a multiobjective context. On the one hand, cash managers may achieve cost savings by forecasting future cash flows. To this end, we perform an empirical analysis of daily cash flow time-series to take advantage of modern machine learning techniques as a key step to connect data analysis and optimization methods in cash management. On the other hand, cash managers may be interested not only in the cost but also in the risk associated to decision-making. Thus, we address the cash management problem from a multiobjective perspective focusing on both cost and risk. In addition, under the current situation of time-varying financial circumstances, the selection of cash management models according to operating conditions and its robustness are worth considering questions. We also show the utility of forecasts through a new cash management model which outperforms the state-of-the-art by guaranteeing optimal solutions. Since most firms usually deal with cash management systems with multiple accounts, we develop a framework to formulate and solve the multiple bank accounts cash management problem. Finally, in an attempt to fill the gap between theory and practice, we also provide a software library in Python for practitioners interested in building decision support systems for cash management.