Forecasting Daily Demand in Cash Supply Chains

Abstract

Problem statement: Previous studies focused on explaining the long run determinants of currency demand offering limited insight into the short-run determinants and co-variability of daily demand in cash supply chains. Approach: This study contrasted competing techniques of forecasting daily demand in cash supply chains in order to determine the overall performance and the potential of joint forecasting for integrated planning. A joint forecasting approach was compared with wellestablished causal forecasting techniques, namely, a vector time series model and a seasonal ARIMA model using simple methods as benchmarks. Evaluation was based on multiple time series obtained from mid-size European bank with forecasting horizons of up to 28 days. Forecasting accuracy was measured using the mean absolute percentage error. Results: The seasonal ARIMA model resulted in a higher forecasting accuracy compared to the vector time series model. Variability in demand was mainly attributed to the day-of-the-week effect. Co-variability is captured by seasonality and calendar effects limiting the potential of joint forecasting. Cumulative forecasts for periods of 14 days are very robust with mean percentage errors of approximately two percent. Conclusion: The results confirmed the benefit of advanced forecasting techniques for daily forecasts. However, the study suggested that the role of information sharing is limited to coordination of replenishments across the cash supply chain and does not yield more accurate forecasts based on joint forecasting.