11 Comments
User's avatar
Nolan's avatar

I've seen for other sports a lot of these bookies give a ton of props for players. I wonder how successful you could be building a model that took those as conditional probabilities (e.g. The over/under is 10 but it has 5 players with 2.5 RBI over/under so therefore either the player over/unders are wrong or the game over/under is wrong). Since there are lot of different props it is harder from them to all be consistent with the game lines.

Expand full comment
Zvi Mowshowitz's avatar

In general the play pile be to bet the props. With notably rare exceptions the high limit lines are to be trusted more, and props are badly priced. There are exceptions, but I think the best place to get prop like info you can trust is fantasy. When everyone is starting a player there is a reason...

Expand full comment
Nolan's avatar

Where do you pull the pitcher info from? Also, do you just scrape the odds from pinnacle? Also, doesn't take US players, how do you place your bets?

Expand full comment
Zvi Mowshowitz's avatar

I get the pitcher and odds info from a secondary third party website that has it in nice form. Well, not nice, but it stores histories.

As for bets, I used to have workarounds but I am not wagering real money right now. I may start doing so again at some point, if things go well.

Expand full comment
krixusthegaul's avatar

I think this is a very useful post that was missing last year when you were posting lines. So thanks for that!

What factors does Aikido consider in the first place? I'm still a little unclear. (obviously you don't have to share if you think this is what gives you your edge)

Small typo: *brier score

Expand full comment
Zvi Mowshowitz's avatar

Aikido considers teams, starting pitchers, which pitchers are left handed, and parks. It uses all information from the past games but ignores same day games.

It would like to include more info but so far attempts to do so didn't improve results. Also if you try to do that it makes manual adjustments harder. I do have some good ideas for improvements if I had coding time.

Expand full comment
krixusthegaul's avatar

By teams, do you mean team roster or team identity? i.e. "which players are in lineup" vs "which team name"

Expand full comment
Zvi Mowshowitz's avatar

I mean team name, it doesn't try and evaluate players. If you're going to do that you need to do it right or all it does is make adjustments impossible to calibrate.

Expand full comment
Andrew Hunter's avatar

Other than generic issues of edge/Kelly why is -110 insufficient? Are there hidden fees on Pinnacle that everyone knows (but I don't, not doing much active sports book work?)

Expand full comment
Zvi Mowshowitz's avatar

Not about hidden fees. You do have deposit/withdraw costs to worry about but they get amortized easily.

The danger is, your model does not know everything, and one must show market some respect. You don't want to have to always be right, and you know that small things are often present you don't track. If you chase a tiny edge you risk paying a lot of vig for ephemeral gains.

Also, if you bet tiny edges, you introduce a lot if variance, which you don't want, and you give up the opportunity to wait for a better spot.

Expand full comment
Andrew Hunter's avatar

Sure, all of that makes sense as per normal--something about your language made it sound like you were making some particular claim about those lines. Just checking.

Expand full comment