1. #1
    Miz
    Miz's Avatar Become A Pro!
    Join Date: 08-30-09
    Posts: 693
    Betpoints: 3132

    Probability that luck created these results

    Thanks to all who post here. I have learned a lot by reading topics in the think tank. I finally decided to join and post a question...

    Of particular interest to me is the idea of a statistically significant sample size. I read through the idea of z-scores etc.... looked at the binomial calculator as well.

    I bet NBA sides and MLB totals - that's it; I don't really bet other sports, I am disciplined, I don't have cable, I don't watch the games I bet on.

    I flat bet, no stars or multi-unit games. I simply bet 1 unit per game.

    I keep records in an excel file every year for each sport.

    I am trying to determine whether or not I believe that I am good enough to increase my bet size next season in NBA sides - whether I believe my winning percentage is a result of handicapping ability or that luck created (some or all of) the results.

    For 2007-2008 NBA season I was 234-161 ATS (I only bet sides). I immediately thought this result had to be helped out by some lucky breaks because of how high the winning percentage was at around 59%.

    As I suspected, the next year it came down to earth a little bit; but there was a slight change in approach.

    In 2008-2009 I began using matchbook and selling half points on nearly every game I bet. I used the half point calculator to find +EV spots, and I also did it in break even spots because I hate laying -juice. So, when you start selling half points your raw percentage is going to drop but your units should increase or stay the same, which is all that matters.

    In 2008-2009 my raw record with selling points was 220-197 = 52.76% but my average vig was probably about -101 or -102 instead of the previous season laying -110 and getting better lines. I ended up +19.7 units that season, flat betting 1 unit per game. Had I not sold points my record would have been about 54.6% which would give the same amount of units at -110 over the same number of games-- a record of about 228-189 over the same number of games at -110 ~+20 units.

    So over the course of two seasons I am either a raw 454-358, or probably a more accurate 462-350.

    So the bottom line is, do you think this is a large enough sample size of games?

    Using the binomial calculator I get

    0.0004 if I use n=812, x=454, p=0.5 (at least 454)
    0.00005 with n=812, x=462, p=0.5 (at least 462)

    I am tempted to believe that the results are not solely that of luck, but I was curious to see what others use to quantify their results.

    Am I missing anything? I am either 55.91% with the raw numbers or 56.9% with the adjust numbers from 2008-2009. This seems to be a real result with a realistic winning percentage over a large number of games.

    I am starting to believe in my ability but is there anything I am missing? I just don't want to fool myself into drinking my own kool-aid too soon. I thought 2 full seasons would be enough but ....

    Should I be looking at the numbers differently? Should I be using a different tool than a binomial calculator?

    Thanks to all who read and/or respond
    Last edited by Miz; 08-30-09 at 10:04 AM.

  2. #2
    Justin7
    Justin7's Avatar Become A Pro!
    Join Date: 07-31-06
    Posts: 8,577
    Betpoints: 1506

    Your raw numbers look good.

    Have you compared how the line moved on your plays? Line movement is a bettor predictor of future wins than past winning percentage.

  3. #3
    Wrecktangle
    Wrecktangle's Avatar Become A Pro!
    Join Date: 03-01-09
    Posts: 1,524
    Betpoints: 3209

    Careful...

    In sports the leagues *change.* I cannot tell you how many great static models I've built over the years that are now worthless. And most of the time you cannot put your finger on what changed it.

    If you are keeping a record of your picks, you should be able to pull a Prior on the level of "goodness" your model is predicting. i.e. if your model predicts a group of teams to be 3-5 points better than the other team, does it do better, worse, or roughly equal to those it says are 5-7 pt dif? how about 8-10? ...and so forth. If you do not see a progression in your win% as your model says it likes a team better in terms of difference, you are much better off flat betting. If your win% falls off with higher diffs, i.e. a "sweet sport," you have a poor fitting algorithm and it is heading for failure, probably in this upcoming season.

    BTW, read up on Bayesian Priors. It is the single most important concept in multi-unit betting and few people understand it to the extent they should, if at all. It is critically important if you plan on using any sort of Kelly Criterion betting.

  4. #4
    Miz
    Miz's Avatar Become A Pro!
    Join Date: 08-30-09
    Posts: 693
    Betpoints: 3132

    Thanks for the reply Justin; I read a lot of your posts.

    I would say that for NBA, generally the line gets worse for my plays, closer to game time (usually a good sign). I haven't tracked all of the games, just sort of noticed it. Sometimes it stays the same and occasionally I get a worse line by betting early.

    It would probably be good for me to go back and look at each game manually and come up with real data for all 812 games.

    The MLB totals is far more correlated to market movement so I put those bets in the night before, typically.

    The MLB totals model only has 115 games under its belt and is 64-44-7. I tend to believe its effectiveness but need a much larger sample size. It is based on xERA, SLG, OBP, and I think it is fundamentally sound. It generates on average 1 to 2 plays per day.

    With matchbook there is a balance between minimizing vig (waiting until closer to game time), and getting a good early line. I usually jump on the lines that people post using the market maker tool. They sometimes misjudge the line and I pounce.

    Thanks for chiming in!

  5. #5
    Miz
    Miz's Avatar Become A Pro!
    Join Date: 08-30-09
    Posts: 693
    Betpoints: 3132

    Thanks Wreck, I appreciate the reply. I will read more about Bayesian priors. I looked at average margin of victory for wins and average margin for losses and the average is higher for the wins. The average ATS win is a cover margins of 5 while the average loss is a lack of cover by 3 pts or so. That was encouraging. I prefer flat betting to avoid hitting my profits too stiffly if my predicted team has an off night. I just prefer to keep it straight forward by flat betting.

    I haven't tracked it in different bins, but will probably take a closer look at that now that you've mentioned it.

    Thanks

  6. #6
    Wrecktangle
    Wrecktangle's Avatar Become A Pro!
    Join Date: 03-01-09
    Posts: 1,524
    Betpoints: 3209

    Miz, below I've reproduced my current WNBA results. Notice how dramatically the higher win% bins are doing in terms of units won. You must know how your model is performing in regards to diffs against the line otherwise you are leaving a lot of cash on the table. You can only do this by strict record keeping both at the time of your betting and at the closing number. BTW, this is using a generic money management system. No Kelly.

    WNBA sides only
    overall: 49 34 59.0% 23.7
    53%: 10 12 45.5% -1.6 (dif: 2-3)
    54%: 25 18 58.1% 5.2 (dif 3.5-5)
    55%: 12 3 80.0% 17.4 (dif 5.5-7)
    56%: 2 1 66.7% 2.7 (dif 7.5+)
    Units: 23.70
    ROI: 24.6%

  7. #7
    Justin7
    Justin7's Avatar Become A Pro!
    Join Date: 07-31-06
    Posts: 8,577
    Betpoints: 1506

    Miz,

    I'd suggest you look at everyone one of your plays in the last 2 years, and compare it to the closing price. Be thorough and systematic. If you are beating the closer by an average of 3/4ths of a point or more, PM me - I'll give you a job.

  8. #8
    Miz
    Miz's Avatar Become A Pro!
    Join Date: 08-30-09
    Posts: 693
    Betpoints: 3132

    Wreck, Thanks, I see your point and can understand how weighted betting would be advantageous the way your projections have fared.

    Justin, I will look back and see how it has done versus line movement. I will record the open and closing pinnacle line and the line I ended up playing. Sometimes I would wait to place the bet until the market was more efficient because the vig was more like 1 to 2 cents, but I will also include the line I played in the spreadsheet. It may take me a couple of days to compile the data. I will start with the most recent season first...

    Thanks for the feedback fellas.

  9. #9
    roasthawg
    roasthawg's Avatar Become A Pro!
    Join Date: 11-09-07
    Posts: 2,990

    Quote Originally Posted by Justin7 View Post
    Your raw numbers look good.

    Have you compared how the line moved on your plays? Line movement is a bettor predictor of future wins than past winning percentage.
    Wouldn't it be just as easily to get lucky and beat the closing number as to get lucky and win the bet? Luck is luck... whether he's beating the closing number 55% of the time or he's winning the bet 55% of the time.

  10. #10
    Miz
    Miz's Avatar Become A Pro!
    Join Date: 08-30-09
    Posts: 693
    Betpoints: 3132

    Roast,

    I am assuming he is placing importance on the idea that betting markets become more efficient the closer they get to game time, so by beating the closing number on a consistent basis, you are ensuring that you have a positive EV.

    Also, beating the closing number on a consistent basis allows people with large bankrolls to execute arbitrage and opens up middling opportunities; assuring themselves income.


  11. #11
    3put
    3put's Avatar Become A Pro!
    Join Date: 01-10-07
    Posts: 31

    Quote Originally Posted by Wrecktangle View Post
    Miz, below I've reproduced my current WNBA results. Notice how dramatically the higher win% bins are doing in terms of units won. You must know how your model is performing in regards to diffs against the line otherwise you are leaving a lot of cash on the table. You can only do this by strict record keeping both at the time of your betting and at the closing number. BTW, this is using a generic money management system. No Kelly.

    WNBA sides only
    overall: 49 34 59.0% 23.7
    53%: 10 12 45.5% -1.6 (dif: 2-3)
    54%: 25 18 58.1% 5.2 (dif 3.5-5)
    55%: 12 3 80.0% 17.4 (dif 5.5-7)
    56%: 2 1 66.7% 2.7 (dif 7.5+)
    Units: 23.70
    ROI: 24.6%
    I have trouble understanding this table.

    Take this:

    54%: 25 18 58.1% 5.2 (dif 3.5-5)

    Please explain what dif 3.5-5 and 54% means.

  12. #12
    Justin7
    Justin7's Avatar Become A Pro!
    Join Date: 07-31-06
    Posts: 8,577
    Betpoints: 1506

    Quote Originally Posted by roasthawg View Post
    Wouldn't it be just as easily to get lucky and beat the closing number as to get lucky and win the bet? Luck is luck... whether he's beating the closing number 55% of the time or he's winning the bet 55% of the time.
    No.

    A winning player might beat the closing number 60-70%. It's much harder for a coin-flipper to do that, than hit 55%.

  13. #13
    Wrecktangle
    Wrecktangle's Avatar Become A Pro!
    Join Date: 03-01-09
    Posts: 1,524
    Betpoints: 3209

    Quote Originally Posted by 3put View Post
    I have trouble understanding this table.

    Take this:

    54%: 25 18 58.1% 5.2 (dif 3.5-5)

    Please explain what dif 3.5-5 and 54% means.
    54% is the expected win% bin (1 z score / standard deviation down from shown in test)
    25 is actual wins this year in 54% bin
    18 is actual losses this years in 54% bin
    58.1% is win% of bin
    5.2 is units won in bin
    dif 3.5 -5 is the "average" difference against the line of the 54% bin, i.e I need to see between a 3.5 to 5 pt difference for a given game to be in this bin.

  14. #14
    ICE-BLOOD
    ICE-BLOOD's Avatar SBR PRO
    Join Date: 07-21-08
    Posts: 1,004
    Betpoints: 7296

    Miz, nice post, got me thinking about your approach

    looks like you should increase unit size, with a bankroll increase

    why only NBA sides and MLB totals? did you find these as your strengths compared to other types of plays?

  15. #15
    Miz
    Miz's Avatar Become A Pro!
    Join Date: 08-30-09
    Posts: 693
    Betpoints: 3132

    Ice, yeah, I tend to do better in those two sports.

Top