Botting the WNBA Season...

[Aicirt]

[ QUOTE ]
So how big of a point differential between predicted/spread do you think would be a 2 unit bet?

[/ QUOTE ]

This is basically a function of how much variance you want to see isnt it?

Assuming the bot is giving the correct percentages as to how often a team will cover and assuming an infinite bankroll, then anytime the bot reports that a team should win more than 52.4% of the time (assuming all bets are -110), we make the most money in the long run by making a max bet on all such bets.

But we dont have an infinite bankroll and we arent 100% sure that the estimated percentage to cover is accurate, so we size our bets down to be able to handle downswings.

So really, I think to solve the bet sizing problem you would have to estimate the error in the bot's percentage to cover estimate and then figure out how big of a downswing your BR can handle. Thus the bet size would be different for each bettor rather than the bot spitting out some optimal bet size.

[rjp]

Aicirt, it's called the Kelly criterion. [img]/images/graemlins/wink.gif[/img]

[Aicirt]

[ QUOTE ]
Aicirt, it's called the Kelly criterion. [img]/images/graemlins/wink.gif[/img]

[/ QUOTE ]

hah, thanks. Yeah I guess I didnt know there was a formal name for the concept.

So according to Kelly, for the CON game today (bot says 70% to cover), we should bet 37% of our bankroll on this game. Which would be 18.5 units assuming a unit is 2% of the bankroll.

So to answer the question of when should we be wagering 2 units instead of 1, Kelly says that we should bet 2 units when we are about 54% sure that we will win (again assuming 1 unit is 2% of the bankroll).

Im suprised that we should bet quite this much (even if CON truly will cover 70% of the time). Is there something else that is obvious that Im missing? [img]/images/graemlins/smile.gif[/img]

[ECB]

Watch out here - If Sales plays for Conn? - which she may not due to an ongoing foot injury. Thats 15 points and one of their best players. Conn stated that she would be playing one game on - one game off on this road trip. She played last night - so - this could be her night off. With her in they would cover - with her out it is going to be a tough cover due to her play and team fatigue.

[rjp]

If the true odds where indeed 70% and you were being offered -110 then yes you are correct, the full Kelly stake size is 70%.

At 70% the probability of covering is obviously overestimated, but if you still felt confident in this number then you want to use half Kelly (18.5%) or quarter Kelly (9.25%) to reduce volatility.

There is money to be made in the WNBA, but I've never run into a true 70% probability of covering. As such, it's time to figure out why such lofty percentages are being computed. [img]/images/graemlins/wink.gif[/img]

[Thremp]

[ QUOTE ]
As such, it's time to figure out why such lofty percentages are being computed. [img]/images/graemlins/wink.gif[/img]

[/ QUOTE ]

Let me translate: Your model clearly sucks.

[dankhank]

shrug, he bet it last night at +7 and it's down to +5.5 now

[rjp]

[ QUOTE ]
shrug, he bet it last night at +7 and it's down to +5.5 now

[/ QUOTE ]

For you to confidently say that CON has a 70% expectation at +7 they would have to win the game 60% of the time. If you're scratching your head at this point and wondering how in the hell +7 is only worth 10% then understand that the points only matter if CON loses, so when they lose 40% of the time each point is worth less than if CON lost a higher percentage of the time, and this is the percentage of the time they must lose to make the points probability and winning probability add up to 70%.

Anyway, if CON were to win 60% of the time then you bet the money line and call it a day. [img]/images/graemlins/wink.gif[/img]

[Thremp]

[ QUOTE ]
shrug, he bet it last night at +7 and it's down to +5.5 now

[/ QUOTE ]

I'm just randomly hating. It could very well just be mis-weighted so that factors that are undervalued are overvalued in his model. Indicating betable lines, but making the edge appear much larger than it should making it hard to bet size. Which is essentially what RJP was saying, something is inherently wrong with the model if its spitting out huge +EV plays on a regular basis.

It's like AceAce and his football models, where he is 4 pts off from the closing line. Yeah right.

[rjp]

Let me also add that I use conservative numbers when calculating the value of points to prevent overbetting as I use a half Kelly stake size, but lets say the market is correct and SAC wins 68.25% of the time. My conservative value for the points shows that CON +7 is a 48.74% bet. To break even at +7 SAC would have to win 63% of the time.