Say you have a model that describes a league of teams by rating with significant degree of confidence and can predict match probability.
Say you have team A with a rating of 4.0
Say you have team B with a rating of 3.9
Team A @ team B, in a league where home factor is worth 0.2
Thus, it would imply the rating would be:
Team A 4.0 vs Team B 4.1 (3.9 + 0.2 home)
Thus team B should be slightly favored.
However, team A has won the past 10 matchups vs team B including several this season. Team A is favored over team B while the model states this should not be. The public will likely hammer Team A.
Assuming no injuries, how do we determine a correlation between past results of specific matchups to incorporate in a rating or is this mathmatically not recommended, as this would be recency bias and this value is already in the individual team rating?
I would argue specific matchups will be correlated in some ways. Some will have a correlation near 0 while others seem to have trends. From a macro perspective the overall rankings per team using a model explain the value of the team vs another random team on a neutral venue but on a micro scale, vs specific teams, a smaller correlation should apply.
Thoughts on how to determine this? Do linesmakers ever put up bad lines knowing when the public will be all over them and refuse to move those lines even when sharps max bet it?