The Yet-To-Be-Named (YTBN) System - The Punters Lounge

Rushian · 06-08-2007, 06:08

Hello all!

I have been lurking around this interesting forum for quite some time, but have only contributed in very few threads. I am not a very experienced punter (so go easy on me!) and this will be my first attempt at producing a system of choosing bets on football matches in the English Premiership, Championship, League 1 and 2. The selections are based on a statistical model developed (by a friend and myself) for these leagues with data since the 2003/04 season. The model uses a small number of variables, a few of which are also included in plopplop’s “Predicting the EPL using stats” thread, although I have to clarify that the model was being developed simultaneously with, yet independently of, plopplop’s (exciting) work.

A few more things about the model (stats alert!): it attempts to model the bivariate distribution of football scores, by taking into consideration explanatory variables which account for the general strength of each team, the current form, the venue etc. The distribution can be used to estimate the probability of certain scores (e.g. 0-0, 2-1, 1-3 etc) and then the probabilities of a home win (i.e 1-0, 2-0, 2-1, etc), a draw or an away win are estimated by summing up the relevant scores. The probabilities are transformed to “fair” odds, and a bet is placed if the offered odds are higher than the estimated “fair” odds (adjusted by a “safety” factor) according to the model.

This method allows for the calculation of “fair” odds regarding exact scores as well, although the overround is much higher hence it would be more difficult to find value, I guess. A possible deviation/extension of this is to use it for over/under 2.5 goals bets, although I would intuitively think that a model of total goals (rather than of football scores) should have been used in this case.

The staking plan will be 10pts per bet, and the starting bank will be 1000 pts. I plan (or should I say “hope to be able”) to update the selections the day before matchday at the latest, and the results a couple of days after the last match of each group of selections. Odds will be taken from odds-comparison, and are valid at the time of posting, but I guess they may change later.

Let me make it clear that this is a paper trail. Moreover, since I don’t know exactly what the commission is for each bookmaker, I am omitting it from the results.

Wish me luck but be prepared to watch it fail spectacularly! I know I am!

Rushian

Rushian · 06-08-2007, 06:52

Ok fingers crossed

:

No	Div	Date	Home Team	Away Team	Bet	Stake	Bet Odds	Bookmaker	W/L/P	Profit/Loss
1	Div0	11/8/2007	Bolton	Newcastle	H	10	2.62	Betfair	P	0.00
2	Div0	11/8/2007	Everton	Wigan	A	10	6.20	Betfair	P	0.00
3	Div0	11/8/2007	West Ham	Man City	A	10	4.10	Betfair	P	0.00
4	Div1	11/8/2007	Ipswich	Sheffield Weds	H	10	2.25	Bluesq	P	0.00
5	Div1	11/8/2007	Preston	Norwich	H	10	2.12	Betfair	P	0.00
6	Div2	11/8/2007	Nott'm Forest	Bournemouth	A	10	6.50	Skybet	P	0.00
7	Div3	11/8/2007	Peterboro	Rochdale	A	10	5.05	Portalandbet	P	0.00
8	Div0	12/8/2007	Man United	Reading	A	10	17.00	Betdaq	P	0.00

Summary Results
Starting bank:		1000.0
Current bank:		1000.0
Total amount staked:		80
	Pending stake:	80
Total number of bets:		8
	Winning bets:	0
	Losing bets:	0
	Pending bets:	8
Strike rate:
Profit (or Loss):		0.0
Yield:

F. · 06-08-2007, 18:34

Good luck to you !!

Even though it's only a paper trail I kow from my own experience it is really exciting to you: after all that hard work on your model you finally decide to test it in public, looking for further improvement and of course to make a profit!!

Best wishes,

F.

AlanDB · 06-08-2007, 19:00

Good to have a relative newbie contribute a system Rushian and it looks like you've put a lot of work into it

. These kinda systems interest me greatly and I wish you all the best with it.

Matthew · 06-08-2007, 19:58

I echo the above......

Good luck with this.

Any stats on how it performed on the build data?
Did you test it on unseen data or is this a "live test"?

What are your predictors? I'd be interested in seeing what others deem "important".

Rushian · 07-08-2007, 06:23

Thanks for your good luck messages

Even if it is only a paper trail, I am nervous about this!

As for some extra information on the model, it performed ok over the test data (used data from 2000/01-2005/06 and tested it on 2006/07 data) with yield of around 4.5% over more than 800 "bets". So yes, the data was unseen.... However, I also calculated the probability of having a negative yield and this was a bit more than 20% which is too high for my liking... hence the paper trail!

Regarding variables, there are home/away strength parameters for each team, and some attacking/defensive form indicators based on each team's recent results. "Class is permanent, form is temporary" etc...

Rushian · 10-08-2007, 08:56

Since the last update some prices have changed, making additional bets appear as "value" according to the system. I am including these new bets with the current odds found on odds-comparison, but I think that it is only right to keep the original bets (already posted) at the initial prices even though some of them have gone up/down.

Fingers crossed again

:

No	Div	Date	Home Team	Away Team	Bet	Stake	Bet Odds	Bookmaker	W/L/P	Profit/Loss
1	Div0	11/8/2007	Bolton	Newcastle	H	10	2.62	Betfair	P	0.00
2	Div0	11/8/2007	Everton	Wigan	A	10	6.20	Betfair	P	0.00
3	Div0	11/8/2007	West Ham	Man City	A	10	4.10	Betfair	P	0.00
4	Div1	11/8/2007	Ipswich	Sheffield Weds	H	10	2.25	Bluesq	P	0.00
5	Div1	11/8/2007	Preston	Norwich	H	10	2.12	Betfair	P	0.00
6	Div1	11/8/2007	Sheffield United	Colchester	A	10	6.60	Betfair	P	0.00
7	Div2	11/8/2007	Crewe	Brighton	H	10	2.30	Betfair	P	0.00
8	Div2	11/8/2007	Nott'm Forest	Bournemouth	A	10	6.50	Skybet	P	0.00
9	Div2	11/8/2007	Port Vale	Bristol Rvs	H	10	2.32	Betfair	P	0.00
10	Div2	11/8/2007	Tranmere	Leeds	A	10	3.55	Betfair	P	0.00
11	Div3	11/8/2007	Grimsby	Notts County	H	10	2.48	Betfair	P	0.00
12	Div3	11/8/2007	Milton Keynes Dons	Bury	A	10	6.40	Betfair	P	0.00
13	Div3	11/8/2007	Peterboro	Rochdale	A	10	5.05	Portalandbet	P	0.00
14	Div0	12/8/2007	Man United	Reading	A	10	17.00	Betdaq	P	0.00

Summary Results
Starting bank:		1000.0
Current bank:		1000.0
Total amount staked:		140
	Pending stake:	140
Total number of bets:		14
	Winning bets:	0
	Losing bets:	0
	Pending bets:	14
Strike rate:
Profit (or Loss):		0.0
Yield:

Rushian · 11-08-2007, 20:02

An encouraging start although it's still eraly days...
(L = Lost, W = Won, P = Pending)

No

Div

Date

Home Team

Away Team

Bet

Stake

Bet Odds

Bookmaker

W/L/P

Profit/Loss

06-08-2007, 06:08	#1 (permalink)
Rushian Dedicated Punter Join Date: 12 Dec 2006 Posts: 239	The Yet-To-Be-Named (YTBN) System Hello all! I have been lurking around this interesting forum for quite some time, but have only contributed in very few threads. I am not a very experienced punter (so go easy on me!) and this will be my first attempt at producing a system of choosing bets on football matches in the English Premiership, Championship, League 1 and 2. The selections are based on a statistical model developed (by a friend and myself) for these leagues with data since the 2003/04 season. The model uses a small number of variables, a few of which are also included in plopplop’s “Predicting the EPL using stats” thread, although I have to clarify that the model was being developed simultaneously with, yet independently of, plopplop’s (exciting) work. A few more things about the model (stats alert!): it attempts to model the bivariate distribution of football scores, by taking into consideration explanatory variables which account for the general strength of each team, the current form, the venue etc. The distribution can be used to estimate the probability of certain scores (e.g. 0-0, 2-1, 1-3 etc) and then the probabilities of a home win (i.e 1-0, 2-0, 2-1, etc), a draw or an away win are estimated by summing up the relevant scores. The probabilities are transformed to “fair” odds, and a bet is placed if the offered odds are higher than the estimated “fair” odds (adjusted by a “safety” factor) according to the model. This method allows for the calculation of “fair” odds regarding exact scores as well, although the overround is much higher hence it would be more difficult to find value, I guess. A possible deviation/extension of this is to use it for over/under 2.5 goals bets, although I would intuitively think that a model of total goals (rather than of football scores) should have been used in this case. The staking plan will be 10pts per bet, and the starting bank will be 1000 pts. I plan (or should I say “hope to be able”) to update the selections the day before matchday at the latest, and the results a couple of days after the last match of each group of selections. Odds will be taken from odds-comparison, and are valid at the time of posting, but I guess they may change later. Let me make it clear that this is a paper trail. Moreover, since I don’t know exactly what the commission is for each bookmaker, I am omitting it from the results. Wish me luck but be prepared to watch it fail spectacularly! I know I am! Rushian

06-08-2007, 18:34	#3 (permalink)
F. Unbelievably lucky Join Date: 30 Jun 2006 Location: Mexico City, MX Age: 33 Posts: 198	Re: The Yet-To-Be-Named (YTBN) System Good luck to you !! Even though it's only a paper trail I kow from my own experience it is really exciting to you: after all that hard work on your model you finally decide to test it in public, looking for further improvement and of course to make a profit!! Best wishes, F.

06-08-2007, 19:00	#4 (permalink)
AlanDB Junior Punter Join Date: 04 May 2006 Posts: 50	Re: The Yet-To-Be-Named (YTBN) System Good to have a relative newbie contribute a system Rushian and it looks like you've put a lot of work into it . These kinda systems interest me greatly and I wish you all the best with it.

06-08-2007, 19:58	#5 (permalink)
Matthew aka plopplop Join Date: 08 Aug 2006 Location: Nottingham Age: 29 Posts: 618	Re: The Yet-To-Be-Named (YTBN) System I echo the above...... Good luck with this. Any stats on how it performed on the build data? Did you test it on unseen data or is this a "live test"? What are your predictors? I'd be interested in seeing what others deem "important".

07-08-2007, 06:23	#6 (permalink)
Rushian Dedicated Punter Join Date: 12 Dec 2006 Posts: 239	Re: The Yet-To-Be-Named (YTBN) System Thanks for your good luck messages Even if it is only a paper trail, I am nervous about this! As for some extra information on the model, it performed ok over the test data (used data from 2000/01-2005/06 and tested it on 2006/07 data) with yield of around 4.5% over more than 800 "bets". So yes, the data was unseen.... However, I also calculated the probability of having a negative yield and this was a bit more than 20% which is too high for my liking... hence the paper trail! Regarding variables, there are home/away strength parameters for each team, and some attacking/defensive form indicators based on each team's recent results. "Class is permanent, form is temporary" etc...