Tennis Betting from PlopPlop

[Matthew]

Hello all, I've been beavering away this past few months with all the data from tennis-data to create a statistical model - much like the one I described in Predict the EPL thread (see sig). I have calculated the following for each and every ATP match between 01-01-2001 --> 31/12/2006. This was some 16000 matches. Current stats<?xml:namespace prefix = o ns = "urn:schemas-microsoft-com:office:office" />

Current world rank

Centrebet Odds

Change in rank (from one month ago to now)

Change in rank (from two months ago to now)

Change in rank (from six months ago to now)

Change in rank (from one year ago to now)

Playing in country of birth (Yes or No)

Age on match day (in days)

H2H stats - all time

Number of H2H matches

Number of 3set H2H matches

Number of 5set H2H matches

Days since last H2H match

W & L %ages

W & L %ages for 3 set matches

W & L %ages for 5 set matches

avg sets won per match

avg games won per match

avg games won per set

avg sets won per match in 3 set matches

avg games won per match in 3 set matches

avg games won per set in 3 set matches

avg sets won per match in 5 set matches

avg games won per match in 5 set matches

avg games won per set in 5 set matches

%age tie breaks won

%age tie breaks won in 3 set matches

%age tie breaks won in 5 set matches

- on surface of upcoming match

Number of H2H matches

Number of 3set H2H matches

Number of 5set H2H matches

Days since last H2H match

W & L %ages

W & L %ages for 3 set matches

W & L %ages for 5 set matches

avg sets won per match

avg games won per match

avg games won per set

avg sets won per match in 3 set matches

avg games won per match in 3 set matches

avg games won per set in 3 set matches

avg sets won per match in 5 set matches

avg games won per match in 5 set matches

avg games won per set in 5 set matches

%age tie breaks won

%age tie breaks won in 3 set matches

%age tie breaks won in 5 set matches

Using KXEN I randomly split this file of 16000 matches up into 3 datasets. 35% to build the model 35% to train the model 30% to test the model KXEN uses the build set to learn what drives a winner, it then tests this on the validation set - repeatedly arguing with itself until it is happy that the resultant statistical model is both explanatory of what drives a winner and is robust enough to use in the real world. The model then simply scores the test set to allow you to see how it would have got on had you used it - it does not use the test data for any other reason. In order to see how viable it is - I scored the test set with the model and assessed the financials to £1 level stakes. Which were:

Staked:	£4,195.00
Returned:	£4,141.74
P+L:	-£53.26
Strike Rate:	68.1%
Yield:	-1.3%
Avg Odds:	1.50
Min Odds:	1.011
Max Odds:	3.46

Slightly disappointing - however in the above scenario I am merely betting £1 level stakes on the person the model gives a more than 50% chance of winning to. For example - if A. Murray has a 50.01% chance of winning and T. Henman has a 49.99% chance - I would back A. Murray. This is very sub optimal in my eyes so I set about optimising where I should set my cutoff as 50% clearly isn't the place. Using VBA in excel I slowly cranked up my cutoff from 50% --> 100% in 0.01% increments - some 500 possibilites. I discovered that the optimum cutoff for me is 63.9% which produced financials of, again at £1 level stake:

Staked:	£2,978.00
Returned:	£3,460.35
P+L:	£482.34
Strike Rate:	83.3%
Yield:	16.2%
Avg Odds:	1.39
Min Odds:	1.011
Max Odds:	3.46
Avg Win Odds:	1.40

This is a reduction in bets placed of 29% a reduction in return of 16.5% an 22.3% increase in strike rate from 68.1% --> 83.3%. P&L and Yield have flipped the right side of the minus sign!!! I reflected on this for a while and decided that a true test would be data that wasn't even in existence yet. So back I went to tennis-data and got the 2007 ATP results up till the end of March 2007. 764 matches. The same stats were calculated and upon scoring the data with the model, I set about discovering the financials - which looked like this (again £1 level stakes, using the 63.9% cutoff):

Staked:	£569.00
Returned:	£660.14
P+L:	£91.14
Strike Rate:	83.5%
Yield:	16.0%
Avg Odds:	1.39
Min Odds:	1.01
Max Odds:	6.75
Avg Win Odds:	1.39

Much relief from my end as the model has held its own on data that wasn't even in existence when it was built. All metrics are almost identical to the results from the test dataset. I spent some time reflecting again and did a bit of research of staking methods and the one I thought suited my model best would be %age of current bank staking. My aim is to increase P&L. Using 1% of a 100 unit initial bank I observe:

Staked:	£905.10
Returned:	£1,049.98
P+L:	£144.88
Strike Rate:	83.5%
Yield:	16.0%
Avg Odds:	1.39
Min Odds:	1.01
Max Odds:	6.75
Avg Win Odds:	1.39

(level of stake is changed after each days betting, not each bet - as in tennis multiple bets go off at the same time) Obviously my strike rate doesn't change but my yield does drop slightly - although in the 3-4 decimal places area - certainly not a material difference. I'm planning to start posting picks on Monday 9th April as that is when the ATP tour kicks off again after a break. I have a few questions for the folks out there: Is the staking plan right - should I be more aggressive given my strike rate? Where are draws for the ATP available from in advance? - I can find nothing for this coming Monday from ATPtennis.com nor the actual websites for the tournaments. Which bookmakers usually offer the best odds for tennis? I am currently registered at: Bet365 BetDirect Betfair Betfred Blue Square Centrebet Gamebookers Interwetten Ladbrokes Mansion Pinnacle Premierbet Skybet SportingOdds Stan James Totesport Victor Chandler (.com) Willhill am i missing a big player in the tennis market? I think thats it - wish me luck in the coming weeks and here's to making some money.:cheers :cheers Matt

[Youssef]

Re: Tennis Betting from PlopPlop I'm not "absolutely" sure, but I "think" bet365 are one of the most decent bookies in the tennis market. However, I'm only comparing them to BetInternet, GameBookers & BWin (personal experience) yet a couple of my friends whom actually bet more on tennis say the same. Also, BetFair (or any other Betting Exchange) are not included, simply because the price changes too quickly. I do use BetFair for live betting on tennis though. I'm gonna keep an eye on this thread as well. I wish you all the luck (as I'm following you, and that means luck for myself as well!) *LOL*