03 March 2010

Projected Season Wins: Vegas vs Diamond Mind Projections . . . Part II

Last week I compared actual season wins against projected seasons wins using projections systems and the Vegas Line. In general, what we found was that each system was pretty accurate. We also found that teams do not overperform considerably (exceeding one standard deviation) if either system predicts the team to win more than 90 games, but that there are several instances of underperformance for these teams (particularly with the projection-based system). Two interesting and most likely non-applicable observations were that Vegas lines tended to under estimate teams they predict to win 85-90 games while the Projection systems tend to underestimate teams in the 73-78 win range. I do not know why in either instance and think it may just be a statistical abnormality.

Today, after the jump, we will be comparing the Vegas line to the Projection system. Again, I have little to no clue about gambling, so I am not sure what it all means related to that . . . if someone wishes to guest blog something about that would be swell. For me, it has more to do with differences between the wisdom of the crowd and the wisdom of a carefully optimized set of algorithms as well as how that might apply to betting . . . which again is something I do not well understand. In this set of analysis, I will be looking for instances where the crowd wisdom would prevail in an over/under bet against the projection system.

The way I chose to compare the Vegas line against the Projection system is to look at instances where they differed against each other by a standard deviation (~8 wins) and then by an arbitrary number that sounds nice (5 wins).

There are 11 instances in which the two system disagreed by 8 or more wins. In three of these instances, siding in agreement with the projection system would win up with losses against the crowd line. I cannot find much in agreement with these three data points other than all three have Projection estimates of under .500 winning percentage. Limiting it to that criteria, we would get a 4-3 betting scenario. The data here is limited, but that might be a decent working hypothesis: trust the projection systems when there is an 8 game spread between the two and the projection system guesses a winning season. Another working hypothesis is to go with the projection system when the spread is 8 games and the projection system predicts fewer wins (5-1) as opposed to more wins (3-2) than the Vegas line. Still, the amount of data is not very great.

There are 27 instances in which the two systems disagreed by 5 or more wins and again the rate at which the Vegas line wins is about 25%. Using the working hypothesis generated above with 5 game or greater spread and a winning Projection-based record, 9-2 is slightly better than the entire data set. The second working hypothesis (5 or more spread with projection predicting less) winds up with a 10-2 result, which is better than the alternative 11-5). Again, the power of this study is pretty limited, but it is something that we will tease apart more so in the future.

When all of the projections come in, we'll comment on who fits into these classifications and follow how well the working hypotheses work.

No comments: