Unfortunately the cited paper is terrible science which was written in ignorance of how common Bitcoin software works.
They basically declared any coin "horded" if it was currently assigned to an address which had never spent. But the default behavior of the reference client is to always send your change to a never used address, which will only ever be spent from once. Additional, the common and strongly recommended pro-privacy behavior is to use a new address per transaction, and businesses need to do that to sort out which payment is which.
So it was actually surprising that the figure was as low as it was... but what it wasn't measuring was "hording".
They basically declared any coin "horded" if it was currently assigned to an address which had never spent. But the default behavior of the reference client is to always send your change to a never used address, which will only ever be spent from once. Additional, the common and strongly recommended pro-privacy behavior is to use a new address per transaction, and businesses need to do that to sort out which payment is which.
So it was actually surprising that the figure was as low as it was... but what it wasn't measuring was "hording".