Originally posted by heinzkat
I think the rating maths would start going wrong if it worked like this suggestion.
It will go wrong everywhere, but here is an somewhat unrealistic example to illustrate the point:
Imagine a 2000 rated player, and a 0 rated player.
They start 100 games, the 2000 rated player resigns them all:
This counts as 100 wins against a 2000 rated player for the 0 rated player, his rating climbs to (say) 3000, meanwhile the 2000 rated player's rating will go down to about 0
Now the players swap and repeat and their ratings inexorably climb without them playing a single full game. The reason this happens is, that if ratings are adjusted with respect to historical rating data then "conservation of total rating" is lost, which is an important property of the rating maths. It's a bit like getting interest on the maximum of this and last years bank balance. You can give all your money to someone else so you both get interest on the same money and the rhp bank goes bust.
With the current system if the players play the same trick then the player that started on 0 will be rated about 2000 by the end of the first 100 games, and the 2000 player will be rated 0 so nothing will be gained overall.