Fast Forecast Reconciliation Using Linear Models

Forecasting hierarchical or grouped time series usually involves two steps: computing base forecasts and reconciling the forecasts. Base can be computed by popular forecasting methods such as Exponential Smoothing (ETS) Autoregressive Integrated Moving Average (ARIMA) models. The reconciliation step is a linear process that adjusts to ensure they are coherent. However using ETS ARIMA for computationally challenging when there large number of forecast, each model must numerically optimized series. We propose avoids this computational problem handles in single step. proposed method very flexible incorporating external data, handling missing values selection. illustrate our approach datasets: monthly Australian domestic tourism daily Wikipedia pageviews. compare ARIMA, show much faster while providing similar levels forecast accuracy.