[austintx05]

I have been trying to devise a mathematical way to have my own set of power ratings based on outcomes, scheduling, opponents outcomes....etc.

By use of matrix analyzation, how can I link every team into every team? My thought on this was that it would be more precise if I could achieve the desired result...in short I am looking make sense of the relationship of every team and their outcomes...

If anyone is good at this stuff...let me know....

[dogman]

Austin, I don't have the answer but King Yao has written a power ratings article that is excellant at his blog, weighing the odds at poker and at the two plus two internet magazine site(vol.3, no.9). I think it maybe will be up only until the end of Sept. so if you want to check it do so in the next couple of days.

[bookie]

Here's the link to the article dog mentioned.

http://www.twoplustwo.com/magazine/current/yao0907.html

[Ganchrow]

Originally Posted by austintx05

I have been trying to devise a mathematical way to have my own set of power ratings based on outcomes, scheduling, opponents outcomes....etc.

By use of matrix analyzation, how can I link every team into every team? My thought on this was that it would be more precise if I could achieve the desired result...in short I am looking make sense of the relationship of every team and their outcomes...

If anyone is good at this stuff...let me know....

I'd certainly consider myself confident in my calculus competence.

But how exactly the above relates to calculus, I have no idea.

[Arilou]

I know calculus well, but I've never found a way to relate it to sports handicapping or any form of power rankings. In general, I find that whenever systems use mathematically complicated methods they are occasionally very strong but if you're not very careful you end up with gibberish.

[jon13009]

I would start with the following site:

http://www.phys.utk.edu/sorensen/ran...steamsv01.html

Other sites with possible ideas to help:

http://www.nutshellsports.com/wilson/nfl.html

This site has a number of links to various statistical ranking sites:

http://www.thepredictiontracker.com//prednfl.html

I have used excel to generate power rankings:

http://home.comcast.net/~nflstats/phil.htm

The idea is simple: Rank the teams according to a statistical category, assign points to the various categories and sum up the points. Then scale the points that each team generates.

Trying to compare the team using more advanced statisitcial methods (such as ANOVA, t-test, chi squared analysis....) is something too difficult and I have yet to even attempt it because there are too many variables involved with some that are not quantative (ie injuries, motivation, rivalries....

[jon13009]

I believe there were a number of mathematicians using differential equations to determine a means of predicting the stock market. Perhaps using a Fourier series to generate a series equation that resembled the Dow Jones average to predict where it was going? Anyway, that would be similar to using ATS trend data to predict game outcomes, and I think trends are a problem because there is an underlying reason why they exist that is independent of how good the teams really are. I hate to use trends in predicting game outcomes because they are going to end at some point.

[jjgold]

Lines are too sharp, your not going to outsmart books

Good Luck trying though, better than not trying

[Ganchrow]

Originally Posted by jon13009

I believe there were a number of mathematicians using differential equations to determine a means of predicting the stock market. Perhaps using a Fourier series to generate a series equation that resembled the Dow Jones average to predict where it was going? Anyway, that would be similar to using ATS trend data to predict game outcomes, and I think trends are a problem because there is an underlying reason why they exist that is independent of how good the teams really are.

I worked in Statistical Arbitrage for many years, developing quantitative (and proprietary) models for predicting certain aspects of US equity and equity derivatives price movements. I personally see no meaningful similarities between what I was doing and "using ATS trend data to predict game outcomes".

[Ganchrow]

Originally Posted by Arilou

I've never found a way to relate [calculus] to sports handicapping or any form of power rankings.

I'll just point out that were it not for calculus my Kelly calculator wouldn't exist.

[goldengoat]

i am good at calculus but am retired from math atm

[jon13009]

Originally Posted by Ganchrow

I worked in Statistical Arbitrage for many years, developing quantitative (and proprietary) models for predicting certain aspects of US equity and equity derivatives price movements. I personally see no meaningful similarities between what I was doing and "using ATS trend data to predict game outcomes".

Perhaps that is ignorance on my part, and I am not saying what you do is not significant and as petty as looking at ATS trends.

My Apologies.

I was making a silly (and rather crude attempt - in light of your profession) to find a means of using calculus to predict NFL games. As you know, the basic concept of using a fourier series is to use a series of periodic functions to describe a seemingly random function. I assumed that the ATS could act as a random function and looked for a means of predicting NFL games using a fourier series to predict such a function. I believe at at some point (long long ago) in a differential equations class my professor indicated that there were MIT scientists employed in the 1950's trying to predict the stock market using various methods and tools such as complex fourier analysis (I think he said using such analysis was not successful).

Are calculus and statistics similar? Perhaps you can make that distinction for me (you being a mathematician). I can also assume the first post was asking for an expert in Linear Algebra, or statistics - not calculus.

To me calculus was primarily a mathematical tool devised by newton to describe physical events in a mathematical language. Clearly concepts such as motion needed a means or tool for describing the actions of physical bodies and calculus is that tool (or language) used by physicists, engineers... How to apply calculus to a complex body of motion (eg. the NFL), is an interesting question. Not being a mathematician, I would have a hard time untying the differences between calculus and statistics. Clearly, the NFL (and the stock market) are better predicted using statistical methodology not calculus.

Again, my apologies.

[Arilou]

Ganchrow, no question it can find its uses in bankroll management. I've never personally felt the need to go that deep into MM questions because I find that emotional impact and variable limits / odds available and other practicalities make them impractical to worry about.

[SPECULATOR 13]

The idea is simple: Rank the teams according to a statistical category, assign points to the various categories and sum up the points. Then scale the points that each team generates.

[/quote]

Dear John13009,this is great advise,but there is one little detail i hope you can help me with:
when it comes to the statistical category that you mention above the problem i have always had with the theory of ratings is how much point to assign to each category.What value Do you give to each category? Is it 0.5 to 5 pts depending of the category or is it lower than that?or is the process of assigning value more involve than that?I would truly appreciate your input on this one

[jon13009]

That is a good question, and I have yet to find the answer. I will usually allocate a number of statistical categories (YPP, O rank, D rank, penalties, W/L, HTA......) a variety of points. What category is more imoportant than the other? I thought the best way to do that is to take the game data and optimize the stat cat values according to the actual game results. The problem is that I never found any optimal values for allocating the point values per statistical category. As the season progresses, the value of one statistic changes over the other. An example was if I allocated turnover rankings at 40% of the total points value, I went from 6-8 to 11-3 one week - the next week would result in a 3-11 result. All in all, it seems that all the stats (and power rankings) do is give you a general feel for what each team's potential is for that week.

To answer you question, I usually allocate 20 pts for YPP, 15 pts for O rank, 15 points for D rank, 25 pts for W/L,........... - this is not what I really do, this is just to give you a general feel that I try to do. After summing the total statistical points for each team, I scale the total points for each team from a maximum vaule of 65 (best team) to a minimum of around 20-30 points (poorest team). Once again I am experimenting with changing the point ranges (for example a maximum of 80, 75.....) In effect, I think it is up to you to determine which statistical category is more important for the various teams during that particular week.

The problem with using too many stats is that the error goes up with each stat implemented. My sheet is loaded with errors, and I am working to find something that is reasonable. I usually end up with a decent power ranking sheet in the end.

However, I never rely on my power rankings alone to determine how to play a game. You need to look at injuries, motivation, inter team rivalries....... This is why sports betting is so interesting, and you should be careful about relying on stats alone to make any decision on what teams to bet for that week.

Hope this helps.