CODEWITHSUNDEEP
rtime-seriesforecastinghierarchical
martka
martka

Reputation: 11

What is proper way of forecasting grouped time series specified via hts-package in R?

I'm trying to understand accurate way of forecasting grouped time series specified as in example posted here. I wanted to use all hierarchical forecasting methods available in hts package with base model - ARIMA. I get an error with some of them (methods: 'mo', 'tdfp', 'tdgsa', 'tdgsf') and as I found on GitHub these methods are not available for gts class object.

I've huge problem with understanding why. As I know both of grouped time series are like 2 hierarchies, so my thought was that we can use for e.g. top-down methods for each hierarchy independently (and the same with middle-out method).

Maybe somebody more familiar with construction of these methods could describe the problem of usage these methods on grouped time series?

Any kind of help would be appreciated.

[ Edit ]

My goal is to analyse forecast accuracy of grouped time series.

To be more detailed I've 2 variables used for disaggregation: car owner (A, B) and car type(X, Y). And as I understood from this paper this data can be describe as groped time series. Am I wrong?

It's really hard for me to find clear definition of grouped time series and examples of forecasting such data. So I would be happy to get links/reference to further literature.

Many thanks for replay!

Upvotes: 1

Views: 1364

Answers (1)

Rob Hyndman
Rob Hyndman

Reputation: 31820

Two hierarchies is not a hierarchy. If you want to use methods designed for purely hierarchical data, then specify the hierarchy. Do it for the two hierarchies and average the results.

Upvotes: 0

Related Questions