[Panekkkk]

Hey all,

I've created a model that helps me pick favorites. That said, this isn't based on a regression so I don't implicitly forecast a line (or win probability) for each pick. I don't want to get into too much detail so I'll get to the question:

Currently I'm betting it flatly. That is, I will be betting it flatly for this upcoming season. I have about a year and a half of data, and from that I can derive the average win probability for my picks. I can calculate my edge by comparing the average historical win probability to the win probability of the offered line for each game. Therefore I can apply a fractional Kelly. I will be testing to see whether the average win probability holds up over this year and I can prospectively compare the Kelly to the flat betting strategy. And for next year will use a fractional Kelly if it comes out on top.

Can someone help me understand the potential pitfalls of using an average win percentage for Kelly? Again, there is no way to forecast the win probability of each individual game. Any thoughts?

I can at least stratify the historical average win probability by the offered line (e.g., from -120 to -130, from -131 to -140, from -141 to -150 and so on) such that I am using different average historical win probabilities on the basis of the offered line to calculate my edge.

[Panekkkk]

Should mention this is for ML's on MLB.

[WendysRox]

I am by no means a math guy, so take this with a grain of salt. But, I've often used line as a criteria for determining the win probability of my predictions. I've noticed peculiar phenomena surrounding certain lines, such as: NFL Road Dog of 3 - my database hits at like 86%, whereas it will hit a 7pt Road Dog at 56%. I'm not saying that line is ALL I use, but it certainly does factor into how confident I am about a game.

And, I would argue that it is entirely acceptable to believe that a database/model can be better at hitting certain lines over others. It just makes sense that given a finite amount of data, certain situations would be more predictable than others. I've noticed in NBA that my database is far more accurate at predicting dogs than favs. And, it's even better at picking dogs of 1-4 points than dogs with higher spreads.

Good luck with your research.