An Approximate Dynamic Programming Approach for Weapon System Financial Execution Management

Abstract

Each year, the Department of Defense (DoD) fiscal calendar starts on October 1 and ends on September 30. Once a fiscal year (FY) begins, weapon system program offices, agencies, and other divisions throughout the DoD serve as the stewards for their budgets. In this role, these offices are tasked with the responsibility of ensuring that congressionally appropriated funding is allocated efficiently over the entirety of the 12-month FY cycle. Furthermore, the DoD financial execution process operates under use-or-lose budgetary regulations. As the calendar moves closer to the end of the FY, September 30, DoD offices undergo a FY closeout review. Dollars that are not adequately spent are at risk of being pulled-back or “swept-up.” In other words, funding can be taken away from an office that is underspending and essentially removed from their FY appropriated budget. During the FY closeout process each year, considerable time and energy is invested in assessing cash utilization levels (disbursements) across the DoD and then implementing where necessary the required “sweep-up” actions. In this research, we investigate the construct of using a learning algorithmic approach known as approximate dynamic programming (ADP) for modeling use-or-lose budgetary systems. ADP is a prescriptive analytics approach used to model sequential decision-making problems under uncertainty. In the context of use-or-lose budgets, we look to leverage ADP in order to generate an efficient month-to-month cash allocation policy in order to minimize the amount of both underspending and overspending that occurs during the FY closeout period. The research presents a framework for modeling and simulating use-or-lose budgets using ADP and discusses the computational complexity and the implications for leveraging the ADP approach in practice.