Top-down forecasting for high dimensional currency circulation data of Bank Indonesia

This paper provides a solution for forecasting high dimensional time series, in the case of currency circulation in Indonesia . Currency circulation data are divided to currency inflow and outflow. Each of them are treated as hierarchical time series separately. The top-down method is applied based on historical proportion, thus only the total series of inflow and outflow need to be modeled. We have compared the implementation of some time series models in top-down forecasting, including Naïve, decomposition, Winters', ARIMA, and two levels ARIMAX with Eid al-Fitr effect. Each model was specified with varying type of proportion and historical period for calculating the proportions. The results showed that the best method is top-down method with historical proportions type 2 that use the forecast of Naïve method. The proportions are best calculated by using historical data from the last 12 months.