I do not understand the necessaritiy of that. You are reducing the points within the system. For what exact reason? Why not keeping the rating as it is? And force the player to play games to rank again?
That's actually the same "accuracy" argument at one hand, and a practical argument at the other hand:
* accuracy: after a long absence, the ELO rating when the player left is less relevant as his/her skill might have changed. Additionally, even if the skill of the player didn't change, the skill of others might have (e.g. all players may have become better), meaning that the rating does not reflect the correct skill anymore. (EDIT: plus what Master Farah explains, more or less members can also impact ratings.) If variance estimates were used, these could be increased artificially to reflect this this new uncertainty (this is what RTL should do, but doesn't). However, in the absence of variance, the next best thing is to (very slowly) move the rating back to the most neutral point, 1500.
* practical: with games themselves not expiring, people who did bad and left consequently, can stil get a new chance with less baggage similar to game expiration if they leave long enough. Meanwhile, people who did really well, can't just come back and claim a high spot with limited games, they'd have to prove themselves again. Both these effects are desired.
I think this concept itself is very good, but the parameters may be up for discussion. Is a 50 days wait to start this process to slow or too fast? Is the point per day too fast or too slow? These numbers were made up on the spot (50 days = max vacation for non-members, point per day well, it's easy?) and seem reasonable to me, but one could argue about that I guess.
With these numbers, after 5 months (150 days), you are 100 points closer to 1500. Meanwhile, very high ranked players have a cushion of more than a year before they have to start anew completely, which at that point is fair I think.
Edited 7/31/2017 20:22:42