Total Bases Props

[Thremp]

Sample size is the biggest issue.

You could just do a BA/SLG adjusted for ABs and then adjust that per the pitchers peripherals (walks being more important than usual here).

Sample size makes adjusting per pitcher almost near impossible. The data is so polluted its virtually worthless.

[rjp]

[ QUOTE ]
Are you surprised/concerned that you found 8 bets against 5Dimes 30 cent lines? How much of an edge do you expect for your single unit bets?

[/ QUOTE ]

I've seen this movie before, and the ending isn't very good. [img]/images/graemlins/cool.gif[/img]

[gtrunner57]

If you've seen this movie before or if you think these bets are stupid, then of course by all means don't bet on them. Maybe you are right, but I like my methodology and will at least see where this goes for a while.

Thremp, I didn't quite catch what you meant there, please elaborate if you will.

Overall: -6.44u
Yesterday was pretty ugly.

Early today:
Ramirez pk over Dunn at -115 (2)

[rjp]

If you're finding so much value then you're likely overestimating your edges, which means you're placing negative EV bets and you're betting too much on the positive EV ones.

Just make sure you take a close look at the method(s) you're using to ensure you're not overestimating things based on a small sample.

[Thremp]

You should be using a BA/SLG model to determine the avg number of bases per at bat and then adjust for the # of at bats they'll have each game and then adjust for each pitcher. You can probably come up with a Std. Dev and punch it into a normal distribution or something of the sort.

For example a TTO hitter would be a significant dog over an Ichiro style hitter.

[gtrunner57]

I think I am doing something very similar to what you are saying. Just for clarification, the definition of slugging is the average number of bases per at bat, though I am not directly using either BA or SLG.

Take my Ramirez over Dunn pick. I'll go into what goes into the prediction specifically, though it will be somewhat lengthy.

I predict Ramirez, adjusted for playing in Chicago today and the HR allowed and non-HR hits allowed rates of Livingston and the CIN bullpen, to get 0.166 singles per PA, 0.071 doubles per PA, 0.007 triples per PA, and 0.055 HR per PA.

Based on his 4th spot in the Chicago lineup, and the WHIP of Livingston and the CIN bullpen, I give him 4.36 PA on average for the game.

So I say, the probability of zero TB's is (1-.166-.071-.007-.055)^4.36
The probability for one base is:
4.36*0.166*(1-.166-.071-.007-.055)^3.36

I've used whatever knowledge of combinations I have to extend this out to calculating the probabilities of getting exactly several TB's.

The final line looks like this:
TB's - 0 - 1 - 2 - 3 - 4 - 5 - 6 - 7 - 8++
Ramirez - .21 - .21 - .18 - .11 - .12 - .07 - .05 - .02 - .03
Dunn - .41 - .18 - .11 - .04 - .16 - .05 - .03 - .01 - .02

As you can see, there is definitely an Ichiro-type effect going on here with Ramirez being more likely to get any number of total bases over the all-or-nothing Dunn except zero and four.

From here it is simple math as to how I calculate my expectation on a given bet.

[rjp]

Are you taking into account prediction error?

That's typically the part that bites you (me) in the ass.

[gtrunner57]

As the nature of being predictions, I'm not going to pretend that they are perfect representations of what will happen. But no, I have not taken into account prediction error.

Actually, I don't even know how I would take into account prediction error. It would seem to me that it would be a fruitless, cyclical process. For instance, say I allow that I may have overestimated/underestimated the skill set of Ramirez. So I adjust accordingly. Well this doesn't make any sense because it seems equally likely that I have also overestimated/underestimated the skill set of Dunn.

Maybe allow for a margin of error that would in turn give you a margin of error of expected value, then only place a bet if the entire margin of error falls within +EV?

[rjp]

You're probably best of consulting a stats book for this, but essentially the more data you have the more confidence you can have in the skill set of both players. As such, you need to determine how you can account for this.

[Thremp]

rjp,

Umm.. Just using Pecota would probably be more than enough.