[PuffPaffy]

Hi guys,

(This is my first post and I'm very glad to have found a forum/community with so much quality information and responsens)

So here's the deal:

I want to test a model based on a match-index system which contains several factors like form, home/away power etc. However, the only input-data I have is based on the end of the season-stats. So the stats I have are 'too' accurate cos off a bigger sample size (end-season).

I could (but this will be a LOT of work) genarate the stats I need for every gameround individualy...

Is there an easier way to back-test this model on accuracy?

Cheers!

[Justin7]

You need to get stats for each game. Compute your power rankings by date... so if you were analyzing a game played on June 5th, you would only use games played on or before June 4th.

Realistically though, a pure numbers based soccer model has virtually no chance of succeeding. The changes in lineups (due to injury, discipline and coaching decisions) will change the fair price of the model too much to be overcome by a pure stats model. Believe me, I have tried

[easyliving]

I have somewhat used a similar thing for a short time as well. Looking at previous games, looking at previous games between the 2 teams etc. It doesn't work as Justin said.Too many factors that can change and due change.

[PuffPaffy]

Thanks guys. I'm sure Justin is right about the fact that a purely (historical) stats driven model isn't gonna work. For future purposes it wil be updated with real-time information (injuries, derby, importancy of match etc). But I figured these things aren't neccesary for back-testing. Because the sample I want to test it on is big enough to even out these 'micro'-stats.
I expect the model to be slighty -EV after testing, but want to see if additional (future) information could make up fot that.

Btw, I found a way to get these game-day-specific stats. Haven't figured out how to use them as ez as possible tho.

@easyliving: I think history between 2 teams is vastly over-rated.

[easyliving]

\
Btw, I found a way to get these game-day-specific stats. Haven't figured out how to use them as ez as possible tho.

would you mind sharing how I can get these stats. I will look further into it as well

[PuffPaffy]

http://www.standbundesliga.nl/

Here you can select stats per game-round for 4 euro-leagues. Have no idea how to link/use these this - I almost think it's easier to make a spreadsheet to calculate these things yourself because you can use more accessable info.

[chunk]

You need to get stats for each game. Compute your power rankings by date... so if you were analyzing a game played on June 5th, you would only use games played on or before June 4th.

Realistically though, a pure numbers based soccer model has virtually no chance of succeeding. The changes in lineups (due to injury, discipline and coaching decisions) will change the fair price of the model too much to be overcome by a pure stats model. Believe me, I have tried

This advice is accurate until the distinction for soccer only, this would hold true for all sports.

[Justin7]

This advice is accurate until the distinction for soccer only, this would hold true for all sports.

Some smaller sports can be beaten while ignoring injury information. If max limits are under $1k, you can probably beat it just looking at a league summary page.

[Juret]

Some smaller sports can be beaten while ignoring injury information. If max limits are under $1k, you can probably beat it just looking at a league summary page.

With a simple regression model for these markets, what z-score over one or two seasons would make you confident having an edge?

[Justin7]

With a simple regression model for these markets, what z-score over one or two seasons would make you confident having an edge?

Even in the smaller markets, I'd rather see consistent line movement in the correct direction.