Computing ratings recursively

[cosmic]

Have you tried this ?

Consider the following situation:
We are in the middle of the season and team A plays against team B, in the premier.
Team A has the rating 1800 and team B has the rating 1700.
Taking into account the home advantage this suggests the probabilities 60% (home) - 20% (draw) - 20% (away), or approximately so.
The corresponding fair prices would be 1.66 - 5.00 - 5.00 in decimal format.

But how were those ratings achieved ?
Maybe team A beat Liverpool sometime previously and was given a big bonus of rating points -say 10- because of that, but Liverpool is by now proven to be a weak force, not what they were expected to be.
So thosee +10 points were not deserved. Maybe 5.
Proceed in this way for all matches of both teams in the season and recalibrate.
Won't the new ratings prove more accurate ?

[Alex]

Our algorithm is desiged in approximately the same way. Newer game results are valued higher than the old ones. We do not work with each team individually. Magic job is made by linear algebra for up to 1024 teams (or players) simultaneously. Moreover, any new result affects ratings not only of just two participant teams, but of all the teams. You may check it yourselves: add new result, and click green arrow button to recalculate the ratings. All the ratings will be different, but the difference will not be visible for insignificant changes as the ratings are rounded to 0.001.