I'm using autoTS R package to fit automatic time series estimation and prediction to a very very large number of time series.
One of the outputs of the prediction function (my.predictions()) is a 95% prediction interval. But sometimes the lower and upper bounds are missing. Does anyone know the reason(s) for this? I've inspected the R codes of the involved functions but didn't get anywhere...
Thanks in advance!
Not all methods of producing point forecasts are probabilistic in nature, in which case there may really be no prediction interval to report. This package is focused on point forecasts and picks amongst different models based on point forecast performance alone.
As far as I can tell, autoTS::my.shortterm is the only method available in the package which is not based on a probabilistic model and it accordingly doesn't report prediction intervals. It's not completely clear to me what this does from a quick glance; this is not a standard, commonly used model name, and it is not particularly well-documented. But, it does not produce prediction intervals.
If you are using autoTS::getBestModel, you may or may not have prediction intervals returned in the end, depending on if it picks that model or not. If you must have prediction intervals, you may want to choose amongst the other models only.
