Lunatism’s Statistical Selections (lss)

[lunatism]

LUNATISM’S STATISTICAL SELECTIONS (LSS)

About Me

I’m NOT new to this forum; I’ve been around for what seems like ages (since middle 2006….something like that). Last year I was consistently posting selections last season for a number of leagues – primarily UK Championship, League One & Two, and then I branched out to world fixtures too (Russian, Scandinavian, etc). Now THIS!!

System Background

All the selections I have EVER posted on PL have always been based on statistics, but I have always chosen to go down the round of explaining my selections in terms of team news (that I’ve compiled from all over the web), that way took a lot of time and effort (all that research after my spreadsheets had already picked out the teams). But I wanted to be sure about my system before I started posting selections as system selections (I’m fussy like that). So for the last 6 months or so I’ve pretty much been on an extended break as I attempted to cover the kinks on my system before I started a thread about it all.

Function

I created this system because I felt I needed something long lasting that could separate the good from the bad, to filter those “likely†winners who consistently seem to let you down from the unconsidered outsiders who seem to put in good odds (and performances) every now and then. In the creation of this system, I hope to finally do that…although we all hope that about our creations, it’s the results that tell the story in the end.

Leagues

The system currently looks at over 295 leagues, and I’ve made a text file that lists all these leagues and you can have these emailed to you on request. Any requests for additional leagues will be seriously considered…although they won’t be many unless we have a new country forming! There aren’t any from Latin America, nor is the United States included, apart from that – take a guess, it’s probably included. That is, unless there are a ridiculously short number of teams in that particular league….then I would have avoided it. So don’t be surprised to find picks from the Algerian Second Division alongside the more familiar Championship selections!

LSS

As you might have gathered, my system is based on statistics. SOLELY on stats that can be gathered from the net. At present, ALL of my data uses the league tables, form tables, head to head and fixture lists from soccerway. I’ve built my system around the stats they provide, but these stats can obviously be obtained from elsewhere too.

After grabbing data tables and fixtures into excel, this information is then translated into something useful, a rating. Individual teams are not rated in my system, for example, I couldn’t right now (using all those tables I have) pick out a general rating for Wolverhampton Wanderers, rather a rating is calculated only when two teams are presented in the fixture format.

It’s my opinion, the most important aspects of analysing any league fixture (since I avoid the cups – too unpredictable) are form, league and head 2 head – in THAT order. With form tables, the stats I primarily use are those that relate to what I guess I’ll have to call location performance. Basically, the home team’s performance at home, and the away team’s performance away, the overall performance IS considered though, but only when looking at league data.

Using form & league data primarily, I’m able to create the ratings for each fixture, and then the fixtures that get a rating of 26 or higher (for home sides), or -26 or lower (for away sides) will be selections – as long as the predicted score is by 2 goals or more (although this particular criteria may change in the future after further testing). This brings me onto score prediction!

Using the location performance for each team, the most likely score is created using a personal variation of the “Poisson Distributionâ€. With matches rated above 20 (and below -20), this has been proving quite successful, being spot on with much frequency, while being one goal out (still winning….just not by two goals, etc) the majority of times. But I still prefer to stick to >26, <-26 at the moment.

After a match has got through “that stage†of things, it then goes through further checks on head to head between the two teams. If the head to head, score prediction and rating all correlate, then that fixture becomes a final selection! Simple, huh?

I WILL be posting the selections that only miss out on the “final cut†due to a predicted 1 nil victory or head 2 head issues (because the ratings are generally right on their own), but these selections I will not be backing (or including in the main stats)…..just putting them up for reference sake. I’ll keep a few limited stats on how these perform of course, before I make a decision whether to drop the 1 nil criteria or not. Just seems to me that if 1 nil is predicted that it WILL be a close game, no matter what the rating is – the match will be prone to the flukes and other variables that are less easy to calculate.

Testing

From the extensive tests I’ve been conducting (well…..not as extensive as I would like, but from 295 leagues – around 1000 fixtures), the average odds have been 1.30; the lowest odds I’d taken have been 1.15, while there is no higher limit. On average there are 5 – 10 selections a week, each one has score prediction as well as rating, and that is why an Asian Handicap can be used – with or as opposed to fixed odds.

Selections

Although most weeks can average between 5 and 10 selections, this is nothing certain, I’m sure that some weeks they may be nothing (although this is yet to happen), while other weeks will see perhaps 20! The majority of selections will come at weekends (due to most countries playing over the weekends), and there obviously be a reduction during major international competitions.

I’ll always try and post selections at least a full day before start time to give you a bit of time to get a look at it and do any additional research (if you were going to place bets). The format will always be something similar to this:

DATE

LEAGUE

FIXTURE

RATING

LEAGUE

PREDICTION

SELECTION

ODDS

16/01/2008

Belgian 3e Klasse - Group A

RACS Couillet

v

RAA Louvieroise

32

15 v 5

0 - 2

A

1.45

17/01/2008

Algerian Division 2

Raed Club de Kouba

v

JSM Cheraga

27

6 v 18

2 - 0

H

1.30

Staking

I will be using a version of the Kelly Criterion to calculate stakes, for those of you who aren’t aware of it, it simply decides on stakes based on odds and the accuracy of selection (a percentage based on what strength you believe the selection to be). I calculate the percentage by looking at the strike rate of the rating number against the strike rate of that score against the strike rate of those odds against the strike rate of location performance. For example, for the RACS game (from above):

Location (H) Strike Rate: 80%

Rating (32) Strike Rate: 90%

Odds (1.45) Strike Rate: 75%

Score (0 – 2) Strike Rate: 20% (NOTE: this is not the CORRECT SCORE strike rate, just the win rate of matches with THIS score as a prediction)

Now we can calculate the match probability – AVERAGE(80%*90%*75%*20%). This returns the value of 66.25%, which can then be used with the Kelly Criterion.

I will post these probability percentages along with the selections….I guess it would help! Now because I’m essentially starting from scratch when I starting posting my selections, these averages will NOT be as accurate as they could be. But over time, they will become extremely accurate. Because there will be no initial strike rates to base the stakes on, I will use a different method for staking until I’ve got through at least 100 selections. It runs along the same lines as above, but instead of a strike rate based on past performance, it’s a ranking, my own opinion of how important each factor is in numerical form (between 1 and 10). There is a lookup table with each rating given such a number, each score, etc. So for the same match:

Location (H) Strike Rate: 3 (playing away from home factors to be considered here)

Rating (32) Strike Rate: 9 (rarely lose matches rated over 30)

Odds (1.45) Strike Rate: 7

Score (0 – 2) Strike Rate: 7 (it’s much more difficult to be predicted a 0-2 victory, so this rating is HIGHER than for the same score at home)

These average out at a 6 (6.33), so a stake of 6.33 (from a starting bank of 100) would be recommended.

If I cannot find odds for the match, it’ll still be a selection, but just one that I cannot bet one – maybe YOU know of where I could bet on this team, and I would welcome such information. The information from these games will still count towards the stats, UNLESS A MODERATOR DOESN'T AGREE – of course no profit can be added for this games, but the strike rate will rise (if they were to win!).

Stats

Each time I post my selections, I’ll also include the latest stats to go along with it (if available), this will include – average odds, strike rates for: correct score, correct winning margin, winning at HT, losing at HT, home win strike rate and away win strike rate. Anything else you think might be interested in, just say, I’ll be happy to consider including it. At the end of each day of selections, i'll post updated stats and put down how the results go.

Archive

I keep full copies (that’s league/form tables and fixtures) of active leagues (by active, I mean, within a 7 days period there are fixtures) and store these in Excel binary workbooks. If anybody is interested in using these for back testing, just let me know the date and I’ll send you a copy – I can also convert it to *.xls if necessary.

Last year I extracted ALL results from soccerway (from their earliest records which date back to Bundesliga ‘93/94 season). The last time I updated this was November, but I will soon be using it again as part of this system, so it will be updated once a month. This includes fixture date, fixture league, home, away and score – that is all, no odds, half times, corners, scorers, cards, etc.

The workbook also includes a handy (but simple) VLOOKUP function that will produce head 2 head data for any two teams’ represented. There are currently over 444400 fixtures on record. Hopefully there will be somebody out there who could help me to come up with a fast solution to incorporating these fixtures into an Excel/Access solution.

Input from YOU

I’m open to any suggestions from you guys that will help to improve this system, it’s still an infant, I’m still learning too, and any input will be considered – I’ll try out most reasonable ideas, and then report back on their results.

THIS SYSTEM WILL START OUT ON HERE PAPER TESTING, AND THEN AFTER A MONTH OR SO I WILL SHIFT IT TO REAL AND I’LL MAKE SURE IT’S CLEAR TOO. Didn't mean to turn this into an essay, but hey - i want you guys to know what's going on :tongue2

[lunatism]

Re: Lunatism’s Statistical Selections (lss) CAN A MOD PLEASE "FIX" MY TABLE ABOVE PLEASE?!! (AND THEN DELETE THIS MESSAGE TOO OF COURSE :)) THANK YOU IT "works" HALF-OK ON FIREFOX.....BUT WHEN I OPEN IT IN IE IT LOOKS A MESS