[Gradius]

I feel like I have hit a wall, and I would like to improve myself. I feel like the math that I use for my models isn't as sophisticated as it could be, and I would like a list of math and statistics subjects that the sophisticated professionals use to help better their models. You can make the list as long as you want, as long as it is relevant. Thank you.

[matthew919]

Cross validation is the most sophisticated topic you'll ever need to learn, and it's relatively straightforward.

http://en.wikipedia.org/wiki/Cross-v...n_(statistics)

[Gradius]

Thanks for the link, Matthew919. By the way, can you suggest some advanced math courses that would help me? And is there a specific statistics book that I should read, or would any statistic book do?

I'm also interested in learning how to scrape a database, but I think I can find that info on my own.

[747planes]

if math and statistics can mac good betting models, all the matheticians will make the casions and sportsbooks go broke. math/statistics can be used to compliment your handicapping but you cannot rely on it completely to win consistently. all your models that you can think of are almost always implemented in the line already. if you suck at handicapping, use a pro service like bettingresource.com. thank me later. 608 units of profit for 2013! Had some big swings but if you had applied their 3% max bet money management, you would be up big safely without getting wiped out from swings.

[Gradius]

I agree, that you can't purely use a math model and ride it to victory; even Haralabos, and Dr. Bob, have said that there are situational factors that they look at that aren't quite in the math models. But any type of edge for myself wouldn't hurt, and it would also broaden my current capabilities.

And I'm not a big fan of tailing players. What if the player either completely stops, or declines from previous years and doesn't adapt? I would rather rely on myself, rather than hope someone else continues to keep riding good years and doesn't stop.

[matthew919]

Read up on data mining. Get a good book on classification and regression methods- just skim the equations and pay more attention to concepts. You'll find that most if not all of the statistical packages you will need are already out there- the more important factors are: 1) identifying a good set of predictor variables, and 2) knowing how to apply/test your models correctly. Two different algorithms can take the same input and give VERY different predictions, so experiment to identify strengths and weaknesses. This concept is completely lost on most people (see post #4). Be prepared to spend some time iterating through these two steps as you refine your methods.

Having a good fundamental understanding of a sport is also very helpful, but if you're good with numbers you shouldn't focus on that too much at the start. I pay attention to key injuries and that sort of thing, but 90%+ of what I bet is strictly based on model output, and involves no subjective/situational analysis at all. GL.

[rwtig]

Personally logistic regression is the technique I use most when handicapping. Other than that, a good knowledge of matrix algebra is important. The best thing to do is familiarise yourself with the concepts and if you feel there is an opportunity to use them, someone will have already written the code for them so you just need to plug it into your program.

[theclutch7]

wELL TONIGHT EVERYONE HAD bAMA! i TOOK okl. BECAUSE THE LINE WENT FROM 15 TO 17, TAKE THE DOG IF IT MOVES 2 POINTS UP.
cause the fav has to deal with a higher game from the dog whom was louzy last weeek, but because the FAV plays better it raises the play level of the dog. the fav need another score to cover. BAMA needed 3 from kick off!

[theclutch7]

Stick with three teams all season keep doubling your bet untill they win, then start with original bet.

[peacebyinches]

Personally logistic regression is the technique I use most when handicapping. Other than that, a good knowledge of matrix algebra is important. The best thing to do is familiarise yourself with the concepts and if you feel there is an opportunity to use them, someone will have already written the code for them so you just need to plug it into your program.

Sharpest first post EVER. Gradius, its great that you want to better your understanding of advanced modeling techniques. Try reading up a little on simple linear regression then maybe generalized linear models which you can find tons of information on the internet without shelling out money for overpriced Statistics textbooks. There are a range of statistical methods out there that range from straight-forward to mind-bogglingly complex, some of which are useful and some which are not going to be useful for your needs. I'd give you some more explicit examples but the best IMO to start with is just understanding concepts related to regression analysis such as ordinary least squares and residuals.

[matekus]

Gradius,

I would strongly recommend Harvard Statistics 110 (34 lectures on YouTube).

matekus

[a4u2fear]

I created a regression equation based on 1505 games from 2003 to 2011 (week 7 to 17) using my estimated/calculated totals for a game as the input, and the outcome total as the output.

I then tested the equation on 2012-2013 (week 7 to 17) and found that when there is a difference between the output (estimated total for the game using the regression equation) and the vegas line of 3 points, I correctly guess whether the game would go over or under 57.3% of the time.

I used weeks 7-17 because there is too much variation with team performance in the first 6 or so weeks.

Any thoughts?