A0205
Title: Alternative percentage error measures for forecasting intermittent and lumpy time series
Authors: Peter Julian Cayton - University of the Philippines (Philippines) [presenting]
Abstract: Intermittent and lumpy time series data are kinds of time series in which zero values are frequently observed due to the nature of the observed phenomenon which generated the series. These may be observed from environmental, logistical, and epidemiological processes when the area or scope of the data is small or narrow. Forecast evaluation with intermittent and lumpy time series using percentage errors is complicated as they are difficult to compute given that zero may be a denominator. The research work surveys alternative formulas for percentage error measures in the case of intermittent and lumpy time series and proposes other alternatives for investigation. These measures are assessed using publicly available data on rainfall in key areas within the Philippines, demand logistics data from the R software and other open sources, and COVID-19 reported cases of local government units in the Philippines.
