MLB Betting

[gtrunner57]

My friend says that he has come up with a method to predict the outcome of MLB games. Are there winning sports bettors that have some kind of model for predicting baseball games? Is this even possible?

He says he has been using the Kelly betting system and has seen some nice % returns. I don't think he's been doing this for very long. So I guess I'm asking, how long would it be before you could tell if he was a winner, or what his edge is, if he has one? How many bets would he have to place for a large enough sample size? And are there any good statistical calculations to be made on results of sports betting?

[whipsaw]

If he still has a nice ROI after 1000 MLB bets, it's a pretty good chance he's a winner.

[New001]

Might as well throw this in here, what's the formula to get a confidence interval on these? On the season I'm at 1210 bets and 1.58% ROI on system plays and 1267 with a 3.44% ROI including regular handicapping (yay hot streaks).

Thanks.

[gtrunner57]

So I guess there are others out there with some sort of a system or method. I guess if you had a method or system, your method or system would put out percent chances to win the game, right? It seems to me that many winning (and losing, of course) sports bettors just put a small unit on each bet. Why does nobody use a Kelly betting method or a coefficient variation of a Kelly system?

[rjp]

[ QUOTE ]
Might as well throw this in here, what's the formula to get a confidence interval on these? On the season I'm at 1210 bets and 1.58% ROI on system plays and 1267 with a 3.44% ROI including regular handicapping (yay hot streaks).

Thanks.

[/ QUOTE ]

To do this test, I normalize your results and assume you win 639 bets and lose 1267-639 bets at +105 odds so that your ROI is 3.39%. Using binom.test in R, your 95% confidence interval is 47.64% - 53.22%. Your 99% confidence interval is 46.78% to 54.08%. I would conclude that you need a larger sample.

I assume most aren't familiar with R or other statistical software, so I created a Trend Confidence Calculator to help people perform these calculations.

[rjp]

[ QUOTE ]
Why does nobody use a Kelly betting method or a coefficient variation of a Kelly system?

[/ QUOTE ]

Many people use Kelly.

Most beginners have a hard time calculating their edge so they assume flat betting is correct. In reality this really isn't a bad strategy if you're laying 11 to win 10, as your edge is going to be pretty consistent (if you have an edge at all), so Kelly would dictate laying about the same amount on each bet.

The same can't be said for baseball, as having a 1% edge for a team expected to win 55% of the time is different from having a 1% edge for a team expected to win 45% of the time.

Anyway, yes, to properly use Kelly you need to be able to approximate your edge, so your "method" or "system" will need to calculate probabilities.

As for me personally, I use a half Kelly stake size.

[New001]

[ QUOTE ]
[ QUOTE ]
Might as well throw this in here, what's the formula to get a confidence interval on these? On the season I'm at 1210 bets and 1.58% ROI on system plays and 1267 with a 3.44% ROI including regular handicapping (yay hot streaks).

Thanks.

[/ QUOTE ]

To do this test, I normalize your results and assume you win 639 bets and lose 1267-639 bets at +105 odds so that your ROI is 3.39%. Using binom.test in R, your 95% confidence interval is 47.64% - 53.22%. Your 99% confidence interval is 46.78% to 54.08%. I would conclude that you need a larger sample.

I assume most aren't familiar with R or other statistical software, so I created a Trend Confidence Calculator to help people perform these calculations.

[/ QUOTE ]
Thanks, I'll play around with that a little bit. I've got a few questions to make sure I understand it though...

The real win/loss are 652-596-19 at -102 if that helps.

So with 95% confidence, my true winning percentage should be somewhere between the 47.64% - 52.22%. If that's true, then obviously I need a much larger sample size. Can you confirm or deny this? Thanks again.

[rjp]

New001, that is what I understand it to mean.

In one text I have in which the author introduces a confidence interval they state that a XX% confidence interval means that if you were to take the same sample size for the given event over and over again then the distribution of XX% of these samples would lie within that interval. I simply assume this is a fancy way of saying that you have XX% confidence that your true result lies within this interval.

[crockpot]

[ QUOTE ]
if you were to take the same sample size for the given event over and over again then the sample mean of XX% of these samples would lie within that interval.

[/ QUOTE ]

[rjp]

crockpot, where could one get further reading on this topic? If the statement is incorrect, then I'd like to know why the authors are incorrectly stating it this way.