I am attempting to do a grouped time series forecast in R using an ARIMA method at the base nodes. However at such a granular level, a few of nodes do not have enough data and so the auto.arima method is returning an error. This is occurring when one brand sells one or two units in a store only every few months or so causing lots of missing data. What is the correct approach for dealing with these cases? Should they simply be excluded entirely from the data or is there another method to handle them? Or should the forecast only begin at an aggregation level which does not include these errors?