Concept Check: How do you use stats to predict individual game win%?

ICE-BLOOD · 10-20-08, 05:36 PM

lots of good points you made

assuming your referral to win% for 1 game is moneyline, not the more popular pointspread in football and basketball.

i agree that in 1 game anything can happen to throw off the result compared to the reasoning behind the odds that are set and what the result is. for example, the moneyline on a favorite for 1st half or 1st quarter is less than the game moneyline, meaning a shorter term contest has a lesser chance that the favorite will win as the odds indicate the favorite should win for a longer term. this correlates to 1 game having a less chance the favorite will win compared to if the game is played 10 times. this also indicates data to arrive at a result for 1 contest is less meaningful.
hope that made some sense, had a hard time wording it.

as for the data being used to come up with a prediction, straight averages arent good enough. better to put more weight in certain categories. also better to put more weight for more recent data. again hope that made sense.

the bookies model compared to mine.
his is better, he has more information. if i set the opening lines, i wouldn't last long.
but here is where the edge is.
bookies use data to set lines. as stated the bookies margin of error is large. but he sets the lines. i get to choose a side where i see the line is off. i also get to choose which games i want, not all of them, among the large amount of choices.

final thought- not worried over short term results such as 1 game. i can use data to get a long term edge, over the course of many events, 1 event at a time

roasthawg · 10-20-08, 05:45 PM

Great answer Ice...agree 100%.

MonkeyF0cker · 10-20-08, 11:15 PM

You have to remember that books need to account for balanced action. They don't want the number to move. That said, there is no model that is 100% accurate in determining each and every game's line to efficiency. In fact, it almost never happens where a line is stagnant from open to close. Short term results are essentially meaningless to advantage gamblers so long as their models are accurately forecasting probabilities. Also, the books use models for their futures probabilities as well.

Peep · 10-21-08, 03:15 PM

I agree most everything, including the kitchen sink, is built into the final number.

So you need something that is outside the box, that bookies and/or other players, are not currently including in their calculations.

Ganchrow · 10-22-08, 12:27 AM

Originally posted by picoman

The data you use are long term averages. All the model predictions are going to work probably for the long term as well. Where i see models can more accurately improve your bets are the season total future wagers.

My argument is that your win% for individual games might be useless in the super short term (one game). Reason is, bookies use some kind of model to come up with the lines. Bookie model has a margin off error that is fairly large. I bet your model isn't that great either. When you combine the two, that will just cancel out any +EV win% advantage that you think is out there. So using data to predict the outcome of individual games might be much less useful than you think.

Any counter-points here?

I'm not sure I'm really understanding your point.

You're asserting that because forecasts are noisy they're useless? That seems a rather silly point to me.

And what exactly do you mean when you write. "The data you use are long term averages"? Different modelers use different units of datum, and in turn use them in different ways. I'm not sure I really get that at which you're driving.