Making a model

[rfr3sh]

So im somewhat new to betting online..and ive been reading about mathematical models that some claim are the best way to go about betting. I also read that you do not need that much knowledge in mathematics but more of a mathematical mind which I feel like i have..currently in math and computer applications at university also took a few years of computer science in university so I'm not computer illiterate either. Anyways to get to the point:


1) What is a model?
2) How do you go about making a model?
3) What kind of stuff would I need to collect?


PS no haters please .. dont waste your time coming in here talking shit or whatever makes you happy cuz its just a waste of time

[Justin7]

1) What is a model?

A program or spreadsheet you design to project scores of matchups in a sport

2) How do you go about making a model?

There are many approaches. I like analyzing 20 years of data to identify factors that predict future performance.

3) What kind of stuff would I need to collect?

All data you think might be useful to predict a team's future performance.

[rfr3sh]

alright thanks for the response..
next question what kind of calculations do you use to predict..is it like turning the data into a line equation and estimating where the next point is?

[Masu485]

i dont think i'd have the resources to make a model, but i am also curious into the whole process. i will try to find a step by step guide online or something, because it does seem more complicated than it looks. I dont even know what programs to use or anything.

[rfr3sh]

exactly im starting at ground zero...but i do beleive with the right knowledge on the subject I am more then capable of making one and I know there are enough people on here who can help so hopefully they read this

[brettd]

Hey rfr3sh, I was asking the same questions only recently. I started a couple of threads about certain aspects of the modeling process. I hope you find these useful:

http://forum.sbrforum.com/handicapper-think-tank/333720-what-level-statistics-do-i-need-create-my-first-model.html

http://forum.sbrforum.com/handicapper-think-tank/331808-best-statistical-software-handicapping-modelling.html

[rfr3sh]

alright i got RExcel will look at it thanks
i will look over how it works

[rfr3sh]

so now what should i start doing start collecting data..or is there a bigger problem to look at first

[dodger33]

before you start collecting data you want to narrow down exactly what you want to put into your model, so you don't end up with way too much information and get overwhelmed.

[rfr3sh]

Yeah thats what I was thinking..so I need to locate variables that I feel are the most important right I'll start with that then thanks

[rfr3sh]

Also what is better to compare..A teams last 10 games lets say, or compare them statisically agaisnt the team the are playing

[Justin7]

Also what is better to compare..A teams last 10 games lets say, or compare them statisically agaisnt the team the are playing

Try them both, see which is more accurate.

[C.S.]

There is a very good Google Tech Talk by David Mease called Statistical Aspects of Data Mining. It's on youtube and based off the course he taught at Stanford the same year. It has 13 or 14 hour lessons. It uses excel and "R" for software. Could be helpful if you're not quite sure where to start.
Here's day 1

http://www.youtube.com/user/GoogleTechTalks#p/search/10/zRsMEl6PHhM

[rfr3sh]

thanks guys I should probably look at this more seriously after my midterm lol

[Bsims]

Here is a link to a paper someone posted about analyzing past soccer scores with an eye toward predicting outcomes.

http://www.football-data.co.uk/ratings.pdf

It's a pretty simplistic approach, but the process might be instructional for relative beginners.

[rfr3sh]

very nice article..I Beleive it could work similar to NBA if you made a model to "rank" a team then subtract the rank of each team to get a predicted spread?

[xraygord]

Cool thread.

[20Four7]

I develop my own rating system and go from there. Taking into account injuries, lost games (Indi/jets in NFL last week of reg season for example) and work out where I believe the team should be. I do some team analysis to figure out how many free throws, how many 2 pointers, how many 3 pointers (I also use this to bet props) and try to work out an approximate score. If manning starts (for example) how did the team do? If he didn't start (for whatever reason) how did the team do? I think justin's approach of 20 years data is too much I rarely use more than the seasons data for calculating things, but a good data base is essential for back testing your approach.

GL.

[rfr3sh]

I think older seasons data is irrelevant because of different players am I right on that one? maybe you could test your model in a particular season though? And today just quickly only looking at points I came up with:

For last 10 ppg


PTS FROM 3's = 3PM * 3
PTS FROM 2's = (3PM - FGM) * 2
PTS FROM FT'S = FTM


TO PREDICT:

Number of 3S ATTEMPTED * % (OVER LAST 10)
Number of 2's ATTEMPTED = (FGA - 3PTA) * % (OVER LAST 10)
Number of FT ATTEMPTED = FTA * % (OVER LAST 10)




PTS = (((3PTA * 3PT%) *3) + (((3PTA * 3PT%) *3)-((FGA-3PTA) * FG%) * 2) + (FTA * FT%)


so here is my question:

By using previous season stats would you be able to use linear regression to predict how many attempted 3s, fgs and free throws a team will get? And for the % would you just use the current % over the season or again the percentage over the last 10 games or whatever I choose to use?

[Justin7]

Teams change a lot from year to year, and even game to game with player changes.

You also need to factor in the opposing team's defense.

[rfr3sh]

any links on how to scrape data as I obviously am not going to enter it manually

I would like to get

Field Goals Attempted/Made
3pts Attempted/Made
Turnovers
Offensive Rebounds/Defensive Rebounds
Free Throws Attempted/Made

For both sides

[Wrecktangle]

Here's a post from a while back that may help:

http://forum.sbrforum.com/handicapper-think-tank/178356-wrecks-commandments-new-sports-modeler.html

[Wrecktangle]

Well, that didn't work:

Here's the cut and paste:

-------------------------------------------------------

I’m sure most of you think I’m pretty twisted; i.e. a modeling fanatic who rarely ventures out in the light of day.... <blink, blink>

…this thread will convince you…

If you are a newbie to sports modeling, these are Wreck’s Commandments to live by when you build a model:

(imagine James Earl Jones reading this to you)

1. Thou shalt (TS): Pick one sport to model. Actually one variant of a sport like NBA sides. Pick a sport you like and want to research because you will be spending a lot of time on it if you are to do even moderately well. Until you can do one sport variant well, you should not do others, as you’ll simply spread yourself out too thinly and accomplish little.

2. (TS): Examine the data available to pick which type of modeling to use. In the NFL, with Gamebooks, you can build Monte Carlo Simulations (Sims) to do both Expected Value (EV) modeling and simulations. NBA is too difficult to model play-by-play so most mortals do EV, as the data to do play-by-play doesn’t exist online.

3. (TS): Divide your data set into two sections, one to build your model from, the second to test against. This helps in preventing overfitting, but is no guarantee. I have a woeful story about this in regards to the NFL and the old USFL.

4. (TS): Spend time and money, if you have to, getting quality data. Most of the data in on-line dbs are full of errors. Sharing data with other modelers works well, but most modelers are like herding cats, and won’t trade their precious data. Not only that, but a great portion of them think they are God’s gift to the modeling world so why should they help anyone else out? …come to think of it, my sh*t doesn’t stink either…

5. (TS): Paper trade your system until you get it working, and use a very small bankroll (BR) in your first season when you fist make wagers. If you exhaust your BR, don’t refill it, work on the model to prepare for next season.

6. (TS): Keep strict records of how your model is doing against the line (ATS) and straight up (SU). Annotate any changes in modeling against your win% both ATS and SU.

7. (TS): Not use Kelly Criterion to bet your first year. You might want to seriously consider NOT using Kelly in other years too. But if you do, plan on spending some quality time on Bayesian Priors. BTW if you use it, plan on Kelly kicking your ass…a lot…

8. (TS): Plan on spending a lot of time rebuilding your model during the season, especially when your model breaks down, i.e. can’t hold a winning percentage over at least 30 games.

9. (TS): Plan on your model breaking down, because it will. When it does, stop everything (yes folks, this does mean: no betting) and examine everything. Look at your data set first because: garbage in with get you sewage out.

10. (BTW): This list doesn’t cover good betting practices (don’t chase, don’t bet more than you can lose, yadda, yadda). Plenty of advice on this elsewhere.

When you conquer a single sport variant, PM me with a copy of your code, and go back to step 1 for the next variant.

Have fun boys and girls…

…and girls, PM me anyway, cause any chick who can code has got to be another alien like me…

[roasthawg]

^Great post wreck.

[rfr3sh]

ya thanks for sharing that...I've decided to build a soccer model for the EPL first since thats my favourite sport to watch makes sense right..plus it seems easiest to start with..Im collecting data right now but its damn time consuming

[durito]

starting with something with 20k limits and 5 cent lines at pinnacle isn't really a good idea, assuming you want to make money.

[rfr3sh]

I dont use pinny for soccer, but also 20 k limits are no problem because I am a student and I think if i could afford to wager 20k on a game I wouldnt be in Uni still..thanks for looking out though

[brettd]

What he's saying refresh is that you shouldn't enter such a highly liquid market to start off with. The higher the limits, the sharper the lines, thus the harder it is to show a profit.

[rfr3sh]

ahh makes sense, I guess I was too confused with NBA and thought soccer would be the easiest route to take maybe totals only or something