Forecasting Daily Demand in Cash Supply Chains
Read this article. It is important because seasonal demand is addressed as the authors attempt to successfully predict demand. In your experience, what are some seasonal products or services that you purchase?
DISCUSSION
In this study, a framework for investigating the overall forecasting accuracy of cash demand and the potential of joint forecasting in cash supply chains is presented. The approach is motivated by advances in supply chain management. The applied methodology considers the potentially interrelated nature of individual daily cash demand series combined with seasonality and calendar effects. In particular, a vector time series model and a seasonal ARIMA model were compared using naive methods as benchmark.
The vector time series model captures the dynamics and co-variability that characterizes a multistage supply chain. From a theoretical perspective, such a model is expected to yield more accurate forecasts, thereby enabling integration of upstream and downstream relationships and improving the efficiency within the cash supply chain. The approach is illustrated for a network of ATMs.
Forecasting accuracy was measured and compared using MAPE. Both, the SARIMA model and the vector time series model resulted in lower MAPEs than the respective benchmark models. Results showed that overall forecasting accuracy is high with a mean absolute percentage error of approximately 20 percent. Particularly, cumulative forecasts for periods of 14 days are very robust with mean percentage errors of approximately two percent. Specification of models revealed strong and consistent seasonal patterns that led to the SARIMA (0,0,0)×(0,1,1)7 model being the single best performing SARIMA (p,d,q)×(P,D,Q)s model. The vector time series model resulted in smaller MAPEs than the SARIMA model during the in-sample period. However, the joint forecasting approach did not yield a higher accuracy than the independent forecasting approach for the holdout sample. Exogenous variables capturing calendar effects accounted for part of the variability in daily cash demand series. Results suggest further that other factors such as weather and local events affect the demand for cash and may even dominate calendar effects.
Practical implications of this study concern the ability to forecast currency needs in order to manage cash supply chains more efficiently. More accurate forecasts enable cost savings by reducing excess cash holdings as well as by cutting the number of emergency replenishments needed to prevent cash outs. For example, a EUR 10000 reduction of the average stock held by a bank branch or ATM results in annual cost savings of EUR 400, given the cost of capital is four percent per annum. The potential economic impact is large considering the entire euro area with 190886 commercial bank branches and 249705 ATMs or the MasterCard network with more than 1 million ATMs worldwide.