Like the idea around LOL betting based on draft results but think we need to rely on data & game knowledge rather than chat's emotions. Game logs for pro leagues were easy to find fortunately (
https://oracleselixir.com/match-data/) though I don't have historical odds.
Maybe we start by predicting kill total based on team comp? Starting simple and looking at distribution of kill total, I notice that the median is ~1 kill lower than the avg but book lines seem to be set based on the average. For some reason the skew is larger in the LPL and LEC (25 v. 26.7 and 24 v. 26) than LCK/LCS where it's only 0.5 kill difference.
For FPX/EDG tonight the team average kills are a bit below the kill total (25 avg vs. 25.5 o/u) and the median being even lower would make me think u25.5 is a good play, but I'm a novice to the eSport. Maybe outcomes where FPX (favorite) wins have a higher kill total which the market is factoring in?
Could the logistic regression method sigs posted earlier work here for LOL team comp? The kill distribution is tri-modal with a long tail which makes sense given match length isn't fixed by time or kill total. Two variables make up kill total, game length and kills/min (engagement rate). Games that either end quick or have low engagement will go under, while 30+min games with lead changes will go over. It makes sense that team comp could predict how likely each branch is but the champion pool is deep enough that I'm not sure I could catch up on direct game knowledge. I think we need to come up with some additional attributes we can add to the game log so we can see what types of team comps lead to faster vs. slower kill pace. Thinking stuff like AP vs. AD champion, ability scaling, maybe an early/mid/late game rating.
A dog taking an early-game comp vs. a favorite with a late-game comp should surely move the odds but instead the books are relying on their live feed from riot being 5mins faster than the live stream the bettors are watching.