Cash Management and Control Band Policies for Spectrally One-sided Levy Processes

Kazutoshi Yamazaki

Abstract

We study the control band policy arising in the context of cash balance management. A policy is specified by four parameters (d,D,U,u). The controller pushes the process up to D as soon as it goes below d and pushes down to U as soon as it goes above u, while he does not intervene whenever it is within the set (d, u). We focus on the case when the underlying process is a spectrally one-sided Levy process and obtain the expected fixed and proportional controlling costs as well as the holding costs under the band policy.

Introduction

Effective cash balance management is a cornerstone of corporate finance and inventory control. The primary objective is to maintain a cash reserve that balances the opportunity costs of holding idle cash against the transaction costs and potential shortages associated with low liquidity. Traditionally, mathematical models for this problem have relied on Brownian motion to represent cash flow fluctuations. However, Brownian motion is continuous and fails to account for the discontinuous "shocks"—such as sudden large expenses or market crashes—that characterize real-world financial environments.

This research extends the classical cash management framework by modeling the underlying cash flow as a spectrally negative Lévy process. This class of stochastic processes incorporates both continuous fluctuations and random, downward jumps, providing a more robust mathematical representation of financial risk and liquidity shocks.

The (d, D, U, u) Control Band Policy

The paper analyzes a four-parameter intervention strategy known as a control band policy. The parameters $(d, D, U, u)$ define the thresholds for intervention:

  • Lower Bound ($d$): If the cash balance falls below $d$, the controller immediately injects capital to raise the balance to the target level $D$.
  • Upper Bound ($u$): If the cash balance exceeds $u$, the controller withdraws funds to reduce the balance to the target level $U$.
  • No-Intervention Zone: While the balance remains within the interval $(d, u)$, the process evolves naturally without interference.

Each intervention incurs a cost structure consisting of a fixed fee plus a proportional cost based on the amount of cash moved. The goal is to evaluate the long-term costs of maintaining this policy.

Methodology: Fluctuation Theory and Scale Functions

The primary challenge in analyzing Lévy processes with jumps is the complexity of their path behavior. To derive the expected Net Present Value (NPV) of future costs, the author employs the fluctuation theory of Lévy processes.

At the heart of the analytical framework are $q$-scale functions, denoted as $W^{(q)}$ and $Z^{(q)}$. These functions are instrumental in solving "first passage" problems—determining when and where a process will first exit a defined boundary. By leveraging the strong Markov property, the author decomposes the controlled process into segments between interventions, allowing for the systematic calculation of:

  1. Expected Upward Controlling Costs ($v_R$): The discounted costs of injecting cash at the lower boundary.
  2. Expected Downward Controlling Costs ($v_L$): The discounted costs of withdrawing cash at the upper boundary.
  3. Expected Holding Costs ($w$): The cumulative discounted cost of maintaining the balance within the band, given a general holding cost function.

Key Insights and Results

The research provides exact, semi-analytical expressions for the total expected cost under the $(d, D, U, u)$ policy. The significance of these results lies in their generality; the formulas are expressed entirely in terms of the scale functions, making them applicable to any spectrally negative Lévy process regardless of its specific Lévy measure.

Key findings include:

  • Unified Framework: The paper demonstrates that the inclusion of negative jumps does not prevent the derivation of clean, closed-form expressions for cost evaluation.
  • Analytical Tractability: By expressing costs through $W^{(q)}$ and $Z^{(q)}$, the author transforms a complex stochastic control problem into a more manageable system of algebraic expressions.
  • Realistic Modeling: By moving beyond Brownian motion, the model better captures the reality of financial systems where sudden drops in liquidity are a constant threat.

Limitations and Future Directions

While the paper successfully evaluates the costs of a given policy, it does not yet solve the optimization problem—identifying the specific values of $(d, D, U, u)$ that minimize the total expected cost. Finding these optimal parameters requires solving a system of four highly nonlinear equations, a task the author identifies as a significant mathematical hurdle.

Future research will focus on using these derived cost functions to verify the optimality of the band policy. This involves demonstrating that the value function satisfies the Quasi-Variational Inequalities (QVI) associated with impulse control, providing a rigorous proof that the $(d, D, U, u)$ policy is indeed the best possible strategy for managing cash flows with jumps.

Summary updated

Abstract

We study the control band policy arising in the context of cash balance management. A policy is specified by four parameters (d,D,U,u). The controller pushes the process up to D as soon as it goes below d and pushes down to U as soon as it goes above u, while he does not intervene whenever it is within the set (d, u). We focus on the case when the underlying process is a spectrally one-sided Levy process and obtain the expected fixed and proportional controlling costs as well as the holding costs under the band policy.
BibTeX
@misc{yamazaki2014cash,
  author = {Kazutoshi Yamazaki},
  title = {{Cash Management and Control Band Policies for Spectrally One-sided Levy Processes}},
  year = {2014},
  eprint = {1410.7490},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1410.7490},
}
Figure