Forecasting Daily Demand in Cash Supply Chains

INTRODUCTION

Cash supply chains play a vital role for the efficiency of banking systems. This complex string of interrelations required to circulate currency from central banks to end users and back, is a special case of closed-loop supply chains. Currency circulates freely between different stages, such as central bank, Cash In Transit (CIT) providers, banks, Automated Teller Machines (ATM), corporate and private customers, incurring transaction and holding costs. Cash management decisions at each stage are interrelated, but mostly beyond the control of the central bank. Hence, currency in cash supply chains represents one of the largest autonomous liquidity factors in banking systems.

This study considers the co-variability of daily cash demand series within cash supply chains in addition to seasonality and calendar effects. The goal is to evaluate the overall forecasting accuracy of daily cash demand and the potential of joint forecasting in cash supply chains.

An analysis of forecasting daily currency demand in cash supply chains is interesting and important for mainly three reasons. From an economic perspective, it enhances the understanding of the demand for currency by considering seasonality and calendar effects in higher frequency data. Traditionally, data availability restricts studies to lower frequency data, such as monthly, quarterly and annual aggregates. From a supply chain perspective, it provides empirical evidence on the accuracy of forecasting demand in cash supply chains, which is essential for integrating autonomous cash management decisions across the cash supply chain, coordinating currency needs and increasing efficiency of banking systems. Similarly, from a business strategy perspective, the study provides further insights into the role of information sharing. In particular, the present study accounts for partnering, outsourcing and privatization that are currently transforming cash supply chains into a mature and optimized businesses.

Forecasting demand in cash supply chains requires an underlying theoretical framework that can reasonably explain the demand for currency. On theoretical grounds, the nature and structure of the demand for cash has been intensely debated by academics in the past. Research has focused on the long run determinants of currency demand. Topics include estimation of income and interest rate elasticity, structural changes from currency change over, influence of alternative payment systems, money hoarding and informal economic activity. Surprisingly little is known about the short run determinants of currency demand.

Recently, modeling and forecasting of daily cash demand in cash supply chains has gained prominence by two papers. Cabrero et al. consider the daily series of banknotes in circulation in the context of liquidity management of the European Central Bank (ECB). The authors analyze and compare the forecasting accuracy of an ARIMA model with a structural time series model based on the Root Mean Squared Forecast Error (RMSE) and the forecasting accuracy test by Diebold and Mariano. Their empirical results suggest the two econometric models explain large parts of the variations in the daily series with the ARIMA model yielding a lower accuracy over forecasting horizons of up to 4 days and higher accuracy for forecasting horizons of more than 4 days compared to the structural time series model. Brentnall et al. take a different approach by studying the temporal process of cash withdrawals for individuals. The developed point process model describes the occurrence of individual cash withdrawals over time. Both studies provide strong evidence for seasonality and calendar effects in the series of daily demand for cash, but do not account for the potentially interrelated nature of demand in cash supply chains. The present study differs from previous work by (1) considering co-variability and (2) by focusing on points of cash withdrawal, not individual customers or macroeconomic aggregates.

Supply chain management is an approach to efficiently integrate the upstream and downstream relationships to provide goods in the right quantities, to the right locations, at the right time in order to minimize system wide costs while meeting service level requirements. Hence, cash supply chain management adopts a system wide view by emphasizing the interrelated nature of currency demand on a microeconomic level. Integration is rendered difficult by the variability in the supply chain due to the uncertainty of demand.

Clearly, independent forecasts isolate individual demand series and potentially amplify disturbances. One commonly suggested way to improve forecasting accuracy of demand in supply chains is the use of central information for joint forecasting. However, empirical evidence regarding the benefits of this approach is to be presented. As such, the present study is the first to capture seasonality, calendar effects and cross-variable dynamics in daily cash demand series in cash supply chains.

Following the rationale outlined above, this study investigates forecasting accuracy and the potential of joint forecasting in cash supply chains by contrasting a vector time series model and a seasonal ARIMA model. Calendar effects are modeled using exogenous variables. The impact of co-variability is illustrated for a network of ATMs. Forecasting accuracy is evaluated using daily cash dispense records.