Statistical Football Prediction Model 2008

[Rushian]

Following last year's attempt http://www.punterslounge.com/forum/f21/yet-named-ytbn-system-52802/ at creating a football prediction model and having made some necessary adjustments, I am prepared to give it another go. Information about the model itself is given in the link above, but to summarize it's based on modelling the actual scorelines of matches in the English Premiership, Championship, League 1 and 2 using some explanatory variables which deal with the standard of the competing teams, their current form, etc. The model produces estimated probabilities which are then translated into "fair" odds. The suggested picks represent value according to the model when compared to the available bookmaker prices. Let me repeat that the model is still under development so this is a paper trail. Level stakes will be used and to make things a bit easier to follow, I'll use only prices available from bookmakers rather than betting exchanges. The bookmakers considered will be the ones available through the football-data website. One last thing: the system attempts to find "value" bets. This means that the strike rate will be low (around 30%-35%) and I am expecting around 1000 bets over the whole season. Wish me luck! :hope Rushian P.S: Oh, and I've thought of a highly imaginative name for the system ... ;)

[aliando]

Re: Statistical Football Prediction Model 2008 good luck rushian! shouldnt the thread be called "the now named system"? ;)

[Rushian]

Re: Statistical Football Prediction Model 2008 Thanks for the wishes aliando :ok Good call on the name!! For the weekend mtches, I am currently waiting for the football-data website to be updated with the available odds, before posting the model's picks. Otherwise I'll do a scan for the odds and post the selections, sometime tomorrow.

[Rushian]

Re: Statistical Football Prediction Model 2008 Well , the football-data website does not seem to carry the latest odds, so I've manually put them in my spreadsheets, turned the virtual handle, and Excel spouted out the results. Hopefully, they data entry went ok but if you notice anyting outrageous let me know. A nice round 20 bets for the weekend:

	No	Div	Date	Home Team	Away Team	Bet	Stake	Bet Odds	Bookmaker	W/L/P	Profit/Loss
1	E1	9/8/2008	Birmingham	Sheffield United	A	10	4.20	Bet365	P	0.00
2	E1	9/8/2008	Blackpool	Bristol City	H	10	2.50	Bet365	P	0.00
3	E1	9/8/2008	Cardiff	Southampton	A	10	4.00	Stan James	P	0.00
4	E1	9/8/2008	Charlton	Swansea	H	10	1.95	Gamebookers	P	0.00
5	E1	9/8/2008	Coventry	Norwich	H	10	2.37	Bet365	P	0.00
6	E1	9/8/2008	Crystal Palace	Watford	H	10	2.10	Bwin	P	0.00
7	E1	9/8/2008	Plymouth	Wolves	D	10	3.30	Bet365	P	0.00
8	E1	9/8/2008	Sheffield Weds	Burnley	H	10	2.30	Bet365	P	0.00
9	E2	9/8/2008	Hartlepool	Colchester	A	10	2.90	VC Bet	P	0.00
10	E2	9/8/2008	Huddersfield	Stockport	H	10	1.80	Blue Square	P	0.00
11	E2	9/8/2008	Northampton	Cheltenham	H	10	1.90	VC Bet	P	0.00
12	E2	9/8/2008	Oldham	Millwall	A	10	3.75	Stan James	P	0.00
13	E2	9/8/2008	Scunthorpe	Leeds	H	10	3.00	William Hill	P	0.00
14	E3	9/8/2008	Bournemouth	Gillingham	H	10	2.20	Bet365	P	0.00
15	E3	9/8/2008	Bradford	Notts County	H	10	1.85	VC Bet	P	0.00
16	E3	9/8/2008	Bury	Brentford	A	10	2.87	Stan James	P	0.00
17	E3	9/8/2008	Grimsby	Rochdale	A	10	2.85	Bwin	P	0.00
18	E3	9/8/2008	Shrewsbury	Macclesfield	A	10	5.00	Stan James	P	0.00
19	E3	9/8/2008	Wycombe	Morecambe	A	10	5.00	Stan James	P	0.00
20	E1	10/8/2008	Nott'm Forest	Reading	A	10	2.55	Bwin	P	0.00

Here's hoping for a good start! :hope

[Kozzin Benson]

Re: Statistical Football Prediction Model 2008

Following last year's attempt http://www.punterslounge.com/forum/f21/yet-named-ytbn-system-52802/ at creating a football prediction model and having made some necessary adjustments, I am prepared to give it another go. Information about the model itself is given in the link above, but to summarize it's based on modelling the actual scorelines of matches in the English Premiership, Championship, League 1 and 2 using some explanatory variables which deal with the standard of the competing teams, their current form, etc. The model produces estimated probabilities which are then translated into "fair" odds. The suggested picks represent value according to the model when compared to the available bookmaker prices. Let me repeat that the model is still under development so this is a paper trail. Level stakes will be used and to make things a bit easier to follow, I'll use only prices available from bookmakers rather than betting exchanges. The bookmakers considered will be the ones available through the football-data website. One last thing: the system attempts to find "value" bets. This means that the strike rate will be low (around 30%-35%) and I am expecting around 1000 bets over the whole season. Wish me luck! :hope Rushian P.S: Oh, and I've thought of a highly imaginative name for the system ... ;)

Good Luck Rushian!! :ok

[Rushian]

Re: Statistical Football Prediction Model 2008 Thanks for the wishes Kozzin Benson :ok Good news and bad news: Good news is that data from the webpage have been updated, so in order for my spreadsheets to have the correct information I've updated the odds. Bad news is that some of the prices have slightly changed (both upwards and downwards) so I am re-attaching the selections with the correct prices:

No	Div	Date	Home Team	Away Team	Bet	Stake	Bet Odds	Bookmaker	W/L/P	Profit/Loss
1	E1	9/8/2008	Birmingham	Sheffield United	A	10	4.20	Bet365	P	0.00
2	E1	9/8/2008	Blackpool	Bristol City	H	10	2.50	Bet365	P	0.00
3	E1	9/8/2008	Cardiff	Southampton	A	10	4.00	Stan James	P	0.00
4	E1	9/8/2008	Charlton	Swansea	H	10	1.95	Gamebookers	P	0.00
5	E1	9/8/2008	Coventry	Norwich	H	10	2.30	Bet365	P	0.00
6	E1	9/8/2008	Crystal Palace	Watford	H	10	2.10	Bwin	P	0.00
7	E1	9/8/2008	Plymouth	Wolves	D	10	3.30	Bet365	P	0.00
8	E1	9/8/2008	Sheffield Weds	Burnley	H	10	2.30	Bet365	P	0.00
9	E2	9/8/2008	Hartlepool	Colchester	A	10	2.90	VC Bet	P	0.00
10	E2	9/8/2008	Huddersfield	Stockport	H	10	1.80	Bet365	P	0.00
11	E2	9/8/2008	Northampton	Cheltenham	H	10	1.90	VC Bet	P	0.00
12	E2	9/8/2008	Oldham	Millwall	A	10	3.75	Stan James	P	0.00
13	E2	9/8/2008	Scunthorpe	Leeds	H	10	3.00	William Hill	P	0.00
14	E3	9/8/2008	Bournemouth	Gillingham	H	10	2.38	Bet365	P	0.00
15	E3	9/8/2008	Bradford	Notts County	H	10	1.85	VC Bet	P	0.00
16	E3	9/8/2008	Bury	Brentford	A	10	2.88	Stan James	P	0.00
17	E3	9/8/2008	Grimsby	Rochdale	A	10	2.85	Bwin	P	0.00
18	E3	9/8/2008	Shrewsbury	Macclesfield	A	10	5.00	Stan James	P	0.00
19	E3	9/8/2008	Wycombe	Morecambe	A	10	5.00	Stan James	P	0.00
20	E1	10/8/2008	Nott'm Forest	Reading	A	10	2.55	Bwin	P	0.00

Coventry is now at 2.30, Bournemouth at 2.38, and Brentford at 2.88. Back tonight with an update... :)

[Rushian]

[Rushian]

[Rushian]

Re: Statistical Football Prediction Model 2008 I will be away for the next few days, so here are the selections for the weekend matches. Again, I've had to manually enter the odds so if something is slightly different (either upwards or downwards) to what football-data publish later, I reserve the right (in fact, the obligation!) to change the prices when I come back... :dude 27 picks for this weekend:

No	Div	Date	Home Team	Away Team	Bet	Stake	Bet Odds	Bookmaker	W/L/P	Profit/Loss
21	E0	16/8/2008	Arsenal	West Brom	A	10	15.00	Bet365	P	0.00
22	E0	16/8/2008	Bolton	Stoke	H	10	1.83	Bet365	P	0.00
23	E0	16/8/2008	Hull	Fulham	A	10	2.80	Bet365	P	0.00
24	E0	16/8/2008	West Ham	Wigan	A	10	4.20	Bet365	P	0.00
25	E1	16/8/2008	Barnsley	Coventry	H	10	2.40	Sportingbet	P	0.00
26	E1	16/8/2008	Bristol City	Derby	H	10	2.25	Bet365	P	0.00
27	E1	16/8/2008	Doncaster	Cardiff	D	10	3.25	Bet365	P	0.00
28	E1	16/8/2008	Preston	Crystal Palace	H	10	2.40	Bwin	P	0.00
29	E1	16/8/2008	Reading	Plymouth	H	10	1.66	Bet365	P	0.00
30	E1	16/8/2008	Southampton	Birmingham	H	10	3.00	Blue Square	P	0.00
31	E1	16/8/2008	Swansea	Nott'm Forest	H	10	2.20	Gamebookers	P	0.00
32	E1	16/8/2008	Wolves	Sheffield Weds	A	10	4.50	Bet365	P	0.00
33	E2	16/8/2008	Brighton	Bristol Rvs	A	10	4.33	Bet365	P	0.00
34	E2	16/8/2008	Carlisle	Crewe	A	10	5.00	Bet365	P	0.00
35	E2	16/8/2008	Colchester	Huddersfield	H	10	2.30	VC Bet	P	0.00
36	E2	16/8/2008	Milton Keynes Dons	Northampton	A	10	3.50	William Hill	P	0.00
37	E2	16/8/2008	Peterboro	Leyton Orient	H	10	1.90	Bet365	P	0.00
38	E2	16/8/2008	Stockport	Leicester	A	10	2.30	VC Bet	P	0.00
39	E3	16/8/2008	Brentford	Grimsby	H	10	2.00	William Hill	P	0.00
40	E3	16/8/2008	Chesterfield	Bury	H	10	1.80	Bet365	P	0.00
41	E3	16/8/2008	Lincoln	Dag and Red	H	10	2.00	Interwetten	P	0.00
42	E3	16/8/2008	Macclesfield	Bradford	H	10	3.25	Bet365	P	0.00
43	E3	16/8/2008	Morecambe	Rotherham	A	10	3.25	Bet365	P	0.00
44	E3	16/8/2008	Notts County	Darlington	D	10	3.25	Bet365	P	0.00
45	E3	16/8/2008	Port Vale	Accrington	H	10	1.67	Blue Square	P	0.00
46	E0	17/8/2008	Aston Villa	Man City	H	10	2.05	Gamebookers	P	0.00
47	E0	17/8/2008	Chelsea	Portsmouth	H	10	1.33	Bet365	P	0.00

Let's see whath this set brings about... Next update on Sunday/Monday. Rushian :hope P.S. By the way, feel free to add any comments/suggestions for the system. Maybe something positive will come out which could be helpfull for all of us.

[Rushian]

[Rushian]

Re: Statistical Football Prediction Model 2008 Quite a few longshots this week, with the average betting odds quite higher than the previous set of bets. Maybe it's time for a surprise result ... We shall wait and see :hope

No	Div	Date	Home Team	Away Team	Bet	Stake	Bet Odds	Bookmaker	W/L/P	Profit/Loss
54	E0	23/8/2008	Blackburn	Hull	H	10	1.62	Ladbrokes	P	0.00
55	E0	23/8/2008	Fulham	Arsenal	H	10	6.50	Bet365	P	0.00
56	E0	23/8/2008	Liverpool	Middlesbrough	A	10	10.00	Bet365	P	0.00
57	E0	23/8/2008	Tottenham	Sunderland	A	10	7.50	Bwin	P	0.00
58	E0	23/8/2008	West Brom	Everton	A	10	2.62	William Hill	P	0.00
59	E1	23/8/2008	Birmingham	Barnsley	A	10	8.00	Stan James	P	0.00
60	E1	23/8/2008	Cardiff	Norwich	A	10	4.50	Ladbrokes	P	0.00
61	E1	23/8/2008	Charlton	Reading	H	10	2.80	Ladbrokes	P	0.00
62	E1	23/8/2008	Coventry	Bristol City	A	10	3.50	Bet365	P	0.00
63	E1	23/8/2008	Crystal Palace	Burnley	H	10	1.95	Gamebookers	P	0.00
64	E1	23/8/2008	Plymouth	Swansea	H	10	2.50	Stan James	P	0.00
65	E2	23/8/2008	Bristol Rvs	Hereford	A	10	3.80	Bet365	P	0.00
66	E2	23/8/2008	Crewe	Walsall	H	10	2.40	Bet365	P	0.00
67	E2	23/8/2008	Hartlepool	Stockport	H	10	2.00	Bet365	P	0.00
68	E2	23/8/2008	Huddersfield	Milton Keynes Dons	A	10	3.80	Bet365	P	0.00
69	E2	23/8/2008	Leyton Orient	Carlisle	A	10	2.62	Bet365	P	0.00
70	E2	23/8/2008	Northampton	Millwall	D	10	3.30	Bet365	P	0.00
71	E2	23/8/2008	Swindon	Colchester	A	10	3.20	Bet365	P	0.00
72	E3	23/8/2008	Barnet	Brentford	A	10	2.62	Bet365	P	0.00
73	E3	23/8/2008	Bradford	Rochdale	A	10	3.40	Ladbrokes	P	0.00
74	E3	23/8/2008	Bury	Morecambe	A	10	4.00	Bet365	P	0.00
75	E3	23/8/2008	Dag and Red	Port Vale	A	10	3.20	Sportingbet	P	0.00
76	E3	23/8/2008	Darlington	Gillingham	A	10	3.80	Bet365	P	0.00
77	E3	23/8/2008	Grimsby	Chesterfield	A	10	2.60	Sportingbet	P	0.00
78	E3	23/8/2008	Rotherham	Chester	A	10	6.50	Bet365	P	0.00
79	E3	23/8/2008	Wycombe	Lincoln	A	10	4.50	Stan James	P	0.00
80	E0	24/8/2008	Man City	West Ham	H	10	2.30	Bet365	P	0.00
81	E0	25/8/2008	Portsmouth	Man United	A	10	1.80	William Hill	P	0.00

Remember, this is just a paper trail!

[Samba_SamPa]

Re: Statistical Football Prediction Model 2008 Just noticed this Rushian, best of luck

[Rushian]

Re: Statistical Football Prediction Model 2008

Just noticed this Rushian' date=' best of luck[/quote'] Thanks Samba! I'll definitely need it :lol

[Rushian]

[Rushian]

[Punto Banco]

Re: Statistical Football Prediction Model 2008 Just a quick one... nice start btw keep it up... ... personally i feel that for the most part 1x2 odds are for suckers (the % taken by the bookies is so much greater that AH)... have you thought about applying this to AH bets as your system on rarely picks draw I understand there may be reasons why you may not but with some bookmakers running lines of 1.96 especially on EPL it makes you wonder if 1x2 is almost useless...almost...:ok

[Rushian]

Re: Statistical Football Prediction Model 2008 Thanks for the comments MakDaddYSportsBET :ok

Just a quick one... nice start btw keep it up... ... personally i feel that for the most part 1x2 odds are for suckers (the % taken by the bookies is so much greater that AH)... have you thought about applying this to AH bets as your system on rarely picks draw I understand there may be reasons why you may not but with some bookmakers running lines of 1.96 especially on EPL it makes you wonder if 1x2 is almost useless...almost...:ok

I am a relative newbie in terms of betting and most importantly in terms of betting slang. So forgive me if I didn't catch your point: If I understand correctly, you are talking about the bookies profit margin (overround?) and how that's bigger in 1x2 bets. The thing is that I use 1x2 because that's what I am familiar with (don't know mach about AH, is it like double-chance 1X or X2?). Since this is just a paper trail, I guess I could try and see what the results would have been by using double-chance/AH, assuming the estimated probabilities are easily obtained from my current data setup (do I just need HDA probabilities or do I need to look back to the scorelines?). In simple terms what does AH mean? Once again, thanks for the contribution :ok

[Rushian]

Re: Statistical Football Prediction Model 2008 Forgot to add that from the data, the biggest overround for a single match was 8.5% whereas the lowest was just 0.8%. On average this figure was at 4.8% so far.

[Rushian]

[Rushian]

Re: Statistical Football Prediction Model 2008 The weekend is upon us with another batch of selections.

No	Div	Date	Home Team	Away Team	Bet	Stake	Bet Odds	Bookmaker	W/L/P	Profit/Loss
82	E0	30/8/2008	Arsenal	Newcastle	H	10	1.45	Gamebookers	P	0.00
83	E0	30/8/2008	Bolton	West Brom	H	10	1.91	William Hill	P	0.00
84	E0	30/8/2008	Everton	Portsmouth	H	10	2.10	Bet365	P	0.00
85	E0	30/8/2008	West Ham	Blackburn	A	10	3.00	Ladbrokes	P	0.00
86	E1	30/8/2008	Barnsley	Derby	H	10	2.55	Bwin	P	0.00
87	E1	30/8/2008	Burnley	Plymouth	D	10	3.40	Bet365	P	0.00
88	E1	30/8/2008	Doncaster	Coventry	H	10	2.38	Blue Square	P	0.00
89	E1	30/8/2008	Preston	Charlton	A	10	3.25	Bet365	P	0.00
90	E1	30/8/2008	Sheffield United	Cardiff	A	10	4.33	Bet365	P	0.00
91	E1	30/8/2008	Southampton	Blackpool	H	10	1.91	William Hill	P	0.00
92	E1	30/8/2008	Watford	Ipswich	H	10	2.38	William Hill	P	0.00
93	E1	30/8/2008	Wolves	Nott'm Forest	A	10	4.75	Bwin	P	0.00
94	E2	30/8/2008	Carlisle	Yeovil	A	10	5.50	Bet365	P	0.00
95	E2	30/8/2008	Cheltenham	Leicester	D	10	3.40	Bet365	P	0.00
96	E2	30/8/2008	Hereford	Crewe	H	10	2.40	Blue Square	P	0.00
97	E2	30/8/2008	Leeds	Bristol Rvs	H	10	1.61	Gamebookers	P	0.00
98	E2	30/8/2008	Millwall	Huddersfield	H	10	2.40	Blue Square	P	0.00
99	E3	30/8/2008	Brentford	Rotherham	H	10	2.30	Blue Square	P	0.00
100	E3	30/8/2008	Chester	Barnet	H	10	2.70	Bet365	P	0.00
101	E3	30/8/2008	Chesterfield	Wycombe	H	10	2.20	Sportingbet	P	0.00
102	E3	30/8/2008	Gillingham	Accrington	H	10	1.83	Bet365	P	0.00
103	E3	30/8/2008	Lincoln	Grimsby	H	10	2.10	Bwin	P	0.00
104	E3	30/8/2008	Macclesfield	Darlington	H	10	3.00	Bet365	P	0.00
105	E3	30/8/2008	Notts County	Shrewsbury	H	10	3.25	Bet365	P	0.00
106	E0	31/8/2008	Aston Villa	Liverpool	A	10	2.55	VC Bet	P	0.00
107	E0	31/8/2008	Chelsea	Tottenham	H	10	1.40	Bet365	P	0.00
108	E0	31/8/2008	Sunderland	Man City	A	10	3.40	Bet365	P	0.00

Fingers crossed :hope

[Rushian]

[Matthew]

Re: Statistical Football Prediction Model 2008 Hi Rushian, I've just read through your old thread and this one. I like what you have done. I'd be interested to know what you explanatory variables are, do you have a definitive list you would care to share? What range of scores do you predict, do you have a %age chane for 5-4 to the home team for instance or do you get to a point and do 1-(probs of all scores up to 4-4). Good luck with the system, I hope you get beyond Oct this time. I reckon you might get 3 out of 3 today. :ok

[Rushian]

Re: Statistical Football Prediction Model 2008

Hi Rushian, I've just read through your old thread and this one. I like what you have done. I'd be interested to know what you explanatory variables are, do you have a definitive list you would care to share? What range of scores do you predict, do you have a %age chane for 5-4 to the home team for instance or do you get to a point and do 1-(probs of all scores up to 4-4). Good luck with the system, I hope you get beyond Oct this time. I reckon you might get 3 out of 3 today. :ok

Matthew, Glad that some people are following the thread. Sometimes, the (forum) silence can be deafening! Regarding your questions, the few explanatory variables used (definitely not as many as you had on the premiership model!) cover the general strength of each team, based on past results, and the form of the teams, calculated by some statistics from recent matches. Nothing which cannot be found in, or calculated from, the football-data files. Please allow me not to divulge more specific information on the actual variables used, although what I can say is that they are not too dissimilar with the ones that you had used in your attempt ;) As I had written in http://www.punterslounge.com/forum/f21/yet-named-ytbn-system-52802/ thread, the work was probably done simultaneously, yet independently, to yours! Regarding the scorelines, the whole estimation procedure is written in a script so I predict up to something ridiculous as 9-9 and add everything up to get the probabilities of HW/D/AW. The sum of the three probabilities is then something like 0.99999 and the remaining is assigned proportionately. Thanks for the kind words. For better or for worse, I plan to follow it all the way to May this time, and see where it takes me :loon

[Rushian]

[Matthew]

Re: Statistical Football Prediction Model 2008 I would have thought you should check how the current results sit alongside your expectations from the development/validation and hold out sample you observed in the creation/building of this model. Without knowing what you did in the creation of this system I am unsure as to whether you have these. My personal preference is a system with a decent yield (which 6.9% definately is) that generates a large number of bets and builds slowly and steadily as opposed to relying on that long shot that came in once. Bottom line, if it is profitable then it's all good isn't it?

[Rushian]

Re: Statistical Football Prediction Model 2008

I would have thought you should check how the current results sit alongside your expectations from the development/validation and hold out sample you observed in the creation/building of this model. Without knowing what you did in the creation of this system I am unsure as to whether you have these.

In terms of expected results, the system is very similar to the validation period: 6.9% instead of 6.0% yield, 38.9% instead of 38.0% strike rate, average bet odds of 3.19 (instead of 3.17) and average strike odds of 2.75 instead of 2.79 in the testing period. My question refers to whether this is actually good or could it be just random. And following your remark:

... My personal preference is a system with a decent yield (which 6.9% definately is) that generates a large number of bets and builds slowly and steadily as opposed to relying on that long shot that came in once. Bottom line, if it is profitable then it's all good isn't it?

How do you define "slowly and steadily"? If I understand it correctly it means low odds/high strike rate. And you are probably right. But how do you go about quantifying it? If it helps, and if my calcuations are correct, based on the system's selections the 95% confidence interval for he true yield is between (-21.9% and 35.7%) which is not very helpful, is it? :wall Is there a better way to monitor its performance?

[Matthew]

Re: Statistical Football Prediction Model 2008 Well it sounds like you are bang on your expectations from your validation sets. So that is good. "Slow and steady" is probably best observed in a chart and the pitch of the trend of your bankroll. Rather than a yo-yo line, I'd personally be looking for a flatter gradual increase in your bank. But this is just my preference and it's maybe not what you would expect from your system. Again, a chart of the validation set will show you what to expect. A high strike rate isn't so important as you could think of your system in terms of days betting. As long as you observe some profit at the end of each day then your bank will slowly and steadily increase. You are right, the CI you have generated isn't too useful at the moment, but given the number of bets you make in a weekend I am sure this CI will tighten in next to no time. If I were you I would be pleased you are observing what you saw in your validaiton sample in a "live" environment so to speak. Maybe others will have a different take on things.

[Rushian]

Re: Statistical Football Prediction Model 2008

Well it sounds like you are bang on your expectations from your validation sets. So that is good. "Slow and steady" is probably best observed in a chart and the pitch of the trend of your bankroll. Rather than a yo-yo line, I'd personally be looking for a flatter gradual increase in your bank. But this is just my preference and it's maybe not what you would expect from your system. Again, a chart of the validation set will show you what to expect. A high strike rate isn't so important as you could think of your system in terms of days betting. As long as you observe some profit at the end of each day then your bank will slowly and steadily increase.

I now see your point Matthew. By increasing the frame (e.g. looking at the bank at the end of the day which may involve 20-30 bets) you smooth out the trend. I guess, this is where the strike odds play their part: for the trend to be smooth the data would need to be aggreagetd only every few bets when the odds are short, whereas as the odds lengthen, the number-of-bets-winfow needs to be larger for the trend to be smooth enough.... I think :eek

You are right, the CI you have generated isn't too useful at the moment, but given the number of bets you make in a weekend I am sure this CI will tighten in next to no time. If I were you I would be pleased you are observing what you saw in your validaiton sample in a "live" environment so to speak. Maybe others will have a different take on things.

Unfortunately, in my validation data, even after a total of 1084 picks, the confidence interval for the yield was still quite large (-3.0% to +14.9%) so I doubt that the CI will "tighten in next to no time" :sad Maybe I need to keep this running for a couple of seasons, and even include Conference data (which seem to perform similarly so far...) to get a big enough sample!!! Or maybe there is another way that I haven't though about. Does anyone have any views on this? Rushian

[WHUPUNTER]

Re: Statistical Football Prediction Model 2008 Good start with this Rushian, keep it up mate.

[werez]

Re: Statistical Football Prediction Model 2008 uhm interesting... "value betting" ok , but if u strike a 3.50 , and loose 3 bets , you just killed the profit . Maybe you should bet more on draws , cause AWs are sometimes rare . In Conference you get 30% away wins in a season , yes , but that doesnt mean stronger teams get the away wins ^^ . Oh and Chelsea vs Tottenham @1.40 isn`t value betting ... I wish you GL , keep it up , my sugestion is to chase more draws , cause your AW`s bets cause major damage to your winnings . You can actually increase your strike rate and the profit if you quit "random" betting and stick to the strong , "winning" picks . I dont know your sistem and your variables , but something is wrong when it comes to AW`s . Maybe go +10 on HW/DR . Here`s an example . I have my own betting/pred sistem , but i like to play only the strong winning hands . This is "value betting" in my opinion . I picked just one AW in 3 rnds , and still my profit is going sky high . This is just one league , only one example . I have a good strike rate cause i`m not betting "random" and hope i hit high odds. Loosing picks kill the profit anyway . You made 70 profit , but you made @ 90 bets . IF u bet HW/DR u can get more profit ... Oh and one more thing . Why do you people bet 10 , and dont try to increase stakes on safer bets . stake 15 for 1.40 maybe ?and 6 for a 3.10 ( for example ) . it will increase your winnings if you try to chase a constant 2xstake / winning . Boring theory ? yea maybe ... works for me btw.