Testing soccer model/results

[PuffPaffy]

Hi guys,


I am building a soccer model wich i believe has some potential, its far from finished though.
What i did to run some tests is the folowing:
- Use data from 3 seasons, 8 different competitions, 07-08 till 10-11
- convert that data to predict odds i need for the folowing season: 11-12
- check in those 8 competitions the first +/- 100 games at start of the season 11-12 if i can find +EV bets according to my model
I used the pinny opening lines, and i strictly bet 1x2


Total stats
Games : 995
Bets : 626
%bets : 62.9
win% : 42.5
Profit(U): 24.39
ROI : 3.90
AVG odd: 2.88


Since i am new to SB but i find it very facinating and interesting i would like to know a couple of things:
- What is a decent sample size with those stats?
- I read alot about that backtesting is often useless and flawed, however since i am not using any data
that was available during that particular season, or any other future events, how should i see this?


Any pointers or advice would be highly appriciated! I might have made some mistakes in my post so feel free to ask
if you want me to clarify certain things.

Ps: reason i only used about 100 games from a season is because after that period i would like to adjust my model.

[Justin7]

That's an interesting approach. I am a bit surprised that it works, because there is usually a fair amount of changes from year. If you are betting into one of the smaller markets though (versus openers, as you identified), you may have the best of it.

If you made those same bets against closers, how would you have done?

Perhaps a better way to backtest it is to also include partial current-season data (only include games played before the match you are evaluating).

[PuffPaffy]

Perhaps a better way to backtest it is to also include partial current-season data (only include games played before the match you are evaluating).

Hey Justin, thanks for replying.

I am currently working on that, using partial current season data. However, i havent got that quitte finished but the reults i got sofar look that it might have some potential.

I would love to get my hands on the pinny CL and see how i would do then. I havent found a (free) website though were they are available. This is partially true because i could get my hands on them at oddsportal but i have to select each game one by one, wich is a shitload of work . If thats the only/best way to get better information about my model, no pb, geuss no sleepy time for me tonight.
If there is any other way wich would be less time consuming, please enlighten me.

One last question, wich is maybe a quitte complicated one: if i could only bet the closing line my picks might be different( and less volume) from the opening ones. Do you mean if i would still have a profit left after placing those same picks/bets against the CL?

Thanks in advance!

[Justin7]

One last question, wich is maybe a quitte complicated one: if i could only bet the closing line my picks might be different( and less volume) from the opening ones. Do you mean if i would still have a profit left after placing those same picks/bets against the CL?

Thanks in advance!

How would you do, betting the same plays, but versus closing prices? If your model is good, your results will get worse (meaning the market agrees with your picks).

[PuffPaffy]

Ok great, thanks for your help!

Anyone knows were i can find the pinny CL for soccer, without having the trouble to do every match manually ( like on oddsportal) ?

[HUY]

Hi guys,


I am building a soccer model wich i believe has some potential, its far from finished though.
What i did to run some tests is the folowing:
- Use data from 3 seasons, 8 different competitions, 07-08 till 10-11
- convert that data to predict odds i need for the folowing season: 11-12
- check in those 8 competitions the first +/- 100 games at start of the season 11-12 if i can find +EV bets according to my model
I used the pinny opening lines, and i strictly bet 1x2


Total stats
Games : 995
Bets : 626
%bets : 62.9
win% : 42.5
Profit(U): 24.39
ROI : 3.90
AVG odd: 2.88


Since i am new to SB but i find it very facinating and interesting i would like to know a couple of things:
- What is a decent sample size with those stats?
- I read alot about that backtesting is often useless and flawed, however since i am not using any data
that was available during that particular season, or any other future events, how should i see this?


Any pointers or advice would be highly appriciated! I might have made some mistakes in my post so feel free to ask
if you want me to clarify certain things.

Ps: reason i only used about 100 games from a season is because after that period i would like to adjust my model.

100 bets are on the low side... To find out how you are doing you can calculate your z-score. Read up on the binomial distribution and how it is approximated by the normal distribution. It's stats 101 and they will come in handy pretty much every day.

[PuffPaffy]

Thanks for the input HUY!

Each soccer season has between 220 en 380 matches, i choose the first half of the season, wich is 995 games like i posted, so not a 100. Reason i didnt choose the whole season is because i wanna be able to adapt to how teams develop during season. Makes perfect sense to me, if im wrong please enlighten me.

Think you misunderstood about the 100bets, i found good bets (according to my model) in 626 matches out of 995.

[HUY]

Thanks for the input HUY!

Each soccer season has between 220 en 380 matches, i choose the first half of the season, wich is 995 games like i posted, so not a 100. Reason i didnt choose the whole season is because i wanna be able to adapt to how teams develop during season. Makes perfect sense to me, if im wrong please enlighten me.

Think you misunderstood about the 100bets, i found good bets (according to my model) in 626 matches out of 995.

Ah, sorry then. 600 bets sounds great. I think you should give it a go, but first read up on the statistics concepts I mentioned before.

[PuffPaffy]

OK, I calcalated my z-score (4+ for 2 datasets I use). I want this number to high right? I read another discussion about this http://forum.sbrforum.com/players-ta...r-success.html and I'm not really sure if a high or low z-score is preferred. The way I see it is that a high z-score (over 800+ sample) says that there's a very small chance you're just lucky, amirite?

[hutennis]

What leagues are you looking at?

[PuffPaffy]

euro-soccer

[hutennis]

Premier, bundesliga and such?

[PuffPaffy]

mostly softer leagues

[hutennis]

like what?

[PuffPaffy]

Basically anything but Premier league, Bundes Liga, Serie A, French and Spanish League. Can anybody help me with those z-scores?

[Juret]

Higher z-score is better. In Excel you can put 1-NORMSDIST(z) to get the probability for a monkey to get the same result. Isn't that explained as well in Ganchrow's post?

[tukkk]

i didnt know that monkeys can break even with the no-vig line

[hutennis]

Basically anything but Premier league, Bundes Liga, Serie A, French and Spanish League. Can anybody help me with those z-scores?

I see. Somehow I don't think so, but that does not matter.

Even without those monsters, you are talking some of the most mature markets in all SB (even "obscured" comparing to Premier leagues have high enough limits to attract very sophisticated players) that are constantly watched by millions and are being analyzed from any angle imaginable by some of the most advanced statistical methods known to mankind.

With that in mind, can you answer following 2 questions please:

1. What are the chances that gigantic edges and opportunities have been missed by everyone in a world only to be discovered by someone who does not have enough knowledge on a subject to be able to interpret z-score on his own?

2. What are the chances that "edges and opportunities" is an illusion manufactured by erroneous and biased approach to analyzing past data which is all too common among those who don't have enough knowledge on a subject even to be able to interpret z-score on their own?

Here is my estimation.

1. 0%
2. 100%

Let's compare notes.

P.S.

In a process of discovery it is a great idea to look for and embrace every possible reason on why you can be wrong.
Beating those reasons one by one is how you get ahead to your goal.
Only looking for conformation why you are right will lead you to a dead end.
It is counter intuitive and certainly not the way most people feel, but it is certainly a correct approach.

[hutennis]

i didnt know that monkeys can break even with the no-vig line

What would stop them?

[PuffPaffy]

Thanks for the reply Hutennis, but I think you being all cynical about me not being able to interpret z-scores is a bit too easy. If my opinion, 'The way I see it is that a high z-score (over 800+ sample) says that there's a very small chance you're just lucky.' is wrong you can just say it.

After extensive tests I just want to reassure myself I'm on the right track - thinking about reasons why it can be flawed. Now I wanted to see what my z-score was - but I mean there's loads of different scores/numbers which tell you exactly the same. I'll just post the raw data (this is from the soft leagues) and you can correct me if I'm wrong (or tell me if I'm a very lucky monkey^^):

Games: 735
Bets: 409
Win%: 43.52
AVG odds: 2.88
Profit: 58.93 Units
ROI: 14.41 %
Z-score: 5.2

Ofcourse I would like to know what things all I might have done wrong. They way I see it, the following could be the case:
- The way I tested is wrong (I betted all OL's which gave me a minimal 10% edge on my models prediction up to 4.00 odds, above 4.00 I looked for much bigger edges - and every bet was 1 unit)
- The samplesize is too small
- I lie about the stats that came out


Option 1 and 2 could be right, but since im new to sb like I said I could easily have overlooked things.
Option 3, I would not joke myself, otherwise, whats the point off putting al this effort in it.

[tukkk]

how about all three

- The way I tested is wrong

most likely, probably made at least one mistake calculating variance
if you even calculated it, probably just took sqrt(n) or smth

- The samplesize is too small

mos def

- I lie about the stats that came out

i guess technically it doesnt count as a lie, when you dont acknowledge that you have tweaked and fitted your formulas several (hundred) times thinking that everything is logical and should be that way

[PuffPaffy]

For z-score (and variance) I used:

Z = 43.52-34.7/sqrt(34.7*34.7/409)
34,7 = 1/2.88

Is this wrong?

And regarding my 'tweaked' formula - it has been tweaked twice because I had a ratio a bit off. It's not like I made/used a random formula and tweaked it untill I had max profits as you are implying.

What would you consider a decent sample for calculating a z-score?

[hutennis]

Thanks for the reply Hutennis, but I think you being all cynical about me not being able to interpret z-scores is a bit too easy. If my opinion, 'The way I see it is that a high z-score (over 800+ sample) says that there's a very small chance you're just lucky.' is wrong you can just say it.

OK I'll just say it.

It is wrong in a way that you, for whatever the reason (most likely not enough knowledge on a subject), completely disregard the fact that
z-score is very far from being an universal tool.
It is not applicable to great many things in life and what (and how) you are trying to apply it to is one of them.

This (ty tukkk)

i guess technically it doesnt count as a lie, when you dont acknowledge that you have tweaked and fitted your formulas several (hundred) times thinking that everything is logical and should be that way

is overwhelmingly most probable reason why we even have this conversation.

P.S.

Would be nice to compare notes on those 2 questions I asked.
Do you think 1. 0% and 2. 100% are reasonable numbers?

Why is it important for you?
B/C unless you beat those arguments silly, there is no point going any farther.

[PuffPaffy]

Yo Hutennis, you don't HAVE to reply if it's not worth your time. Therefore I have no idea why you want to get in comparing-notes argument. But yes, I don't think you need vast knowledge about z-scores to be able to derive reasonable outcomes for football matches.

[hutennis]

After extensive tests I just want to reassure myself I'm on the right track - thinking about reasons why it can be flawed.

That's great, but for that exercise to be a useful one you have to make sure that in your thinking you cover all the reasons.

There are dozens and dozens of them to start with.

How many have you been thinking about?

[hutennis]

Yo Hutennis, you don't HAVE to reply if it's not worth your time. Therefore I have no idea why you want to get in comparing-notes argument.

I reply b/c I'm positive that my replies are in fact the best advise i can give you.
I'm sorry that it is not what to want to hear.

But yes, I don't think you need vast knowledge about z-scores to be able to derive reasonable outcomes for football matches.

You are wrong.

[PuffPaffy]

No offence but you dont give any advice whatso ever. What you do is telling me that i must do something wrong but dont quantify what. Mostly without any reason behind it.
Im here to learn and i can handle all the critisism even if it would mean that my current thoughtprocess is wrong, i just want to know: what,were,how why etc.

[PuffPaffy]

1) i did not tweak my model to get positive results
2) i did not use any data from my testingperiod in my model
3) i have 1750 matches, were i found more then 1000 good bets
4) according to à variancecalculator its almost impossible that im just lucky
5) i dont think those numbers are correct ,0 and 100%, i dont think its that black and whit

[hutennis]

Here is what you are doing wrong.

You are doing something that has 100% chance of being done before by someone else who did it, with 100% probability, better than you ever be able to do it.

That makes all your efforts completely useless. Even if your math and approach is correct it's still does not matter.
All your efforts are completely negated by a simple fact that all of it has been done before and on a far more sophisticated level.
Jugging from the questions you post, there is 100% chance that you are as unable to compete with what already been done as you are unable to compete in 100m Olympic race. Maybe someone else in a world can, but you, personally, can not. It is way out of your league.

Not understanding such a simple and rational concept IS your biggest mistake.
I'm being your best friend and adviser right now by pointing it to you, so you can save a lot of time and money.
You can continue to do what you are doing if you find it to be highly entertaining hobby, but that would be the only reason I can think of.

If you don't understand what I just said - God help you!

[PuffPaffy]

Now we are getting somewhere!

I understand your thoughtprocess completely, however, i think its wrong. Now im obviously gonna tell you why, cus if i wouldnot, that wouldnt be really constructive.....

you say the following:
"That makes all your efforts completely useless. Even if your math and approach is correct it's still does not matter.
All your efforts are completely negated by a simple fact that all of it has been done before and on a far more sophisticated level."

Meaning, like someone already did/does it better then you then there is no point in trying it. I might be missing something, but that sounds very wrong to me.
1) Phil Ivey is a great pokerplayer, im never gonna be as good as he is, so there is no point in trying making a living/profit out of poker.
2) Macdonalds has a very good chunck of the fastfoodindustry, im never gonna be able to do a better job the MD, so its useless to start a fastfoodrestaurant.

"Not understanding such a simple and rational concept IS your biggest mistake."

whut?
You are never gonna be as great as the greatest man ever lived, why dont you just kill yourself?

It really sounds to me your logic is flawed and not mine.

If you have proof of Sportbetting not being beatable, please enlighten me.
Really, i hope you can proof me wrong, but the answer you gave me doesnt change my mind/opinion atall.

[Justin7]

Hutennis has recieved yet another ban for trolling the HTT. He may come back in two weeks.

In small markets, all kinds of approaches can be EV. Even unscientific ones. If you want to attack NFL or another major market, that is different.

[CHUBNUT]

back testing is a waste, it can only tell you if your model is good or bad and if its bad your distraught and if its good you will be over confident going forward. the biggest problem with launching any model induced betting strategy is confidence, and thats down to stakes. it doesnt matter what the stats and dry runs have given up, a sudden expected on paper bad run can devastate everything and destroy moral. i find 50% or even 25% of normal stakes are the best for these type of circumstances, if they start losing badly you can console yourself that by being sharp and only losing 25 % of your normal stakes you live to fight another day while winning will fortify you with increased stakes.

You can go down the mathematical road too far as is the case of justin7 and not see the reality of sports betting. as for betting these situations i can assure you that outside the main leagues you have no chance, either you lose or get banned. people in the industry are telling me that any winning accounts betting on anything other than the premiership are either closed or significantly lowered stakes, so sadly most of the talk on these forums is hypothetical. personally outside pinnacle and maybe SBOBET on the day, I find myself lost for realising my expected wager so one has to question when you read posts assuming massive bets at premium odds, the reality is it doesnt happen.

[Justin7]

You can pretty much ignore all of Chubnut's trolltalk. He'll be gone awhile also.

[indio]

If your system is working, and producing profits, stick with it. Do annual reports and continue to do what's profitable until it stops being profitable. People beat markets all the time, so if your being smart by attacking with same risk amount on each play and keeping risk amounts in proper proportion to compensate variance, you'll do fine. Since your playing Pinnacle, you don't have to worry about getting limited or booted, so keep with it, and good luck.

[cyberbabble]

You've done about all you can as far as backtesting. You need to do realtime testing now.
Put a $100 in an account and start making $1 bets to see how it works. It is important to actually "pull the trigger" by making a real bet.

There will be ambiguous or close decision cases that will turn up in real betting that didn't show up while backtesting. Real bets allow you to decide how to handle these cases. Your bets will usually not be at the open/closing lines but somewhere between. A few cents difference on moneyline bets can make the difference between winning and losing.

Record the open/close lines in your betting results spreadsheet. BTCL - beat the closing line - provides a measure of the performance of your system. Even if your W/L record or money won results aren't real strong, if you BTCL then you may have a good system. You just happened to start with a losing streak. Most of the BTCL discussion on here is worthless, but the concept is good.

The reason it is tough to win is that you must ultimately be better than the other players. They've been doing this for a long time. They have more experience, knowledge, and resources than you and they are not easy to beat. This does not mean you can't beat them. It means that it is not easy.

It's like going from college athletics to the pro leagues. The college guys are great athletes but the pro guys are almost superhuman.

Good Luck.