Posted On:2/13/2009 2:30pm
Also, based on the above logic and not strictly following the system, Kos at -450 is probably an ok pick, because Thiago has fought a WEAK ASS schedule, .400-&-something SOS ans SOSO.
SOS=Strength of Schedule. Win % of last 10 opponents
SOSO= Strength of Schedule's Opposition. Win % of the last 10 people your opponents fought, as of those fights.
Last edited by J_Treez; 2/13/2009 2:49pm at .
NOTE TO SELF - MOAR GRAPPLE - GET A NORMAL HAIR CUT - REPEAT
Posted On:2/13/2009 3:10pm
Style: Submission Grappling
Originally Posted by J_Treez
Yeah, using my prototype odds conversion chart the system put Diaz as a -200 (2 to 1) based on their power rankings of 6.75 vs. Guida's 6.25 or so. Clay had lost 5 of his last 10 fights, but to more difficult competition (.770) that had fought pretty tough schedules (.604). Diaz, on the other hand, was 9-1 against a .645 strength of schedule that had fought a weak .481 schedule themselves.
A similar scenario was the last Matt Brown fight, which he won with a very Guida-like record against a guy who was kicking the **** out of cans who'd fought cans. I want to examine scenarios like this further, and come up with some sort of modifier to reflect accurately on the situation.
-200 = bet $200 to win $100
+200 = bet $100 to win $200
I know nothing about gambling, but I know a fair amount about probability, and the scientific method. let me make the following observation:
A method, even an optimal method, for predicting outcomes based on past observations will select the most rational choice based on the data available. This can be at great odds with what is actually observed when prediction is checked against reality, and still be no fault of the method employed. The fault will lie with the data and its inability to support the "correct" conclusion using sound methods.
For example if I ask you to predict the next symbol in the sequence "1,1,1,1,1" you would be correct (ie, reasonable) in predicting "1" as the next symbol in both the case where the correct sequence is 100 "1"s followed by 100 "0"s, and in the case where the sequence is 5 "1"s followed by 5 "0"s. Even though the prediction would be wrong in the second case, it is still the most reasonable choice.
It is very risky to bet money based on the type of modeling you are doing.
That said, here is a suggestion:
If this were a scientific problem, I would add two features to your model.
(1) Remove outliers. Freak knockouts, injuries, or any case where a rational person who follows the sport would shout "WTF!" at the result, should be removed form the power rating calculation.
(2) Watch the trends. Fighters rise and fall. Plot their power rating over time, fit a polynomial trendline to the graph (PM me if you don't know how do trendline in excel) and project it forward by one match. THAT is the power rating you should be working with. This way you are estimating a fighter's PR at where it is likely to be in the future, rather than on the mean of where it has been recently.
Now darkness comes; you don't know if the whales are coming. - Royce Gracie
KosherKickboxer has t3h r34l chi sao
In De Janerio, in blackest night,
Luta Livre flees the fight,
Behold Maeda's sacred tights;
Beware my power... Blue Lantern's light!
Posted On:2/13/2009 5:52pm
(1) I can see the benefits of doing this when possible. On many fights, though, you really don't know much more than the result, time, and method. Like small shows, early fights, etc.
I like the idea, though, and will look to incorporate it.
(2) Should I still use my "last 10 fights" sample size when delving into plotting trends, or would doing spreadsheets for a fighter's entire career better suit this type of analysis? I get your basic idea and really dig it. I think this is a big part of what's missing from my football handicapping as well. I do ok (54% when 52% is break even, 55-60% is considered the maximum sustainable winrate ), but not as well as I'd like. I'm going to read up a little so I can discuss this with you more intelligently. Expect that PM when I do.
Given your grasp of probability, here's all you need to know about gambling other than money management: (1) Determine what you think the odds of a given contest should be. (2) Find lines that you think are incorrect. (3) Look at a few sportsbooks to find the best price on the contest you've analysed. This is a BIG deal. Some people don't even handicap, just go around looking for the best odds. (4) Bet.
Of course, this whole thread is about doing the best job we can on step one.
Articles and Reviews
Tools and Info