Mr Intensity Posted August 10, 2008 Share Posted August 10, 2008 Re: Tennis rating/info/assessment system I was thinking of some categories and ended up with these.... Big server Problem server Aggressive baseliner Grinder Serve/volley most Serve/volley some Strong mentally Weak mentally Obviously if you're neither a big server or problem server then you're an average serve so probably don't need that (same with aggressive/grinder and serve/volley). Quote Link to comment Share on other sites More sharing options...
Datapunter Posted August 10, 2008 Author Share Posted August 10, 2008 Re: Tennis rating/info/assessment system Ok, imagine you have those categories on each player, then what ? Quote Link to comment Share on other sites More sharing options...
Matthew Posted August 10, 2008 Share Posted August 10, 2008 Re: Tennis rating/info/assessment system Ok' date=' imagine you have those categories on each player, then what ?[/quote'] Once we understand what type of player they are then we can understand what impact that has on their performance against other types of players. It's all very well knowing that player A has beaten player B in the past but if we know that in the last 6 matches against a gritster they have been hopeless then that's worth knowing isn't it. Without these categories we couldn't do that. It's just another set of metrics for which to assess a players chance against another. Quote Link to comment Share on other sites More sharing options...
Datapunter Posted August 10, 2008 Author Share Posted August 10, 2008 Re: Tennis rating/info/assessment system It's all very well knowing that player A has beaten player B in the past but if we know that in the last 6 matches against a gritster they have been hopeless then that's worth knowing isn't it. Without these categories we couldn't do that. With you so far except for one thing, knowing that doesn't neccesarily lead to profit does it ? If in those matches they where hopeless fine, but if they where priced up at 20.00 each time still not much use. So whatever metrics used firstly you still need to check whether there's a point to the metric and secondly what its contribution could be towards profiting from betting ? So basically shouldn't the test for validity be included in the definition of the metric ? Quote Link to comment Share on other sites More sharing options...
Matthew Posted August 11, 2008 Share Posted August 11, 2008 Re: Tennis rating/info/assessment system With you so far except for one thing, knowing that doesn't neccesarily lead to profit does it ? If in those matches they where hopeless fine, but if they where priced up at 20.00 each time still not much use. So whatever metrics used firstly you still need to check whether there's a point to the metric and secondly what its contribution could be towards profiting from betting ? So basically shouldn't the test for validity be included in the definition of the metric ? I like to think of it like this. Lets say there is a certain amount of variation that explains a tennis match. The aim of a statistical model is to explain as much of the variation as possible. The perfect model would explain 100% of the variation in a tennis match. Simplistically speaking you could argue that, of the 100% of the variation in the results:- 40% is explained by the current form of the players 20% is explained by the h2h record between the two players 20% is explained by the players records against their opponent hand players 20% is explained by the players records againsts their opponents type of players (big serving etc) Where I am coming from is an attempt to come up with as many different variables that we know at the start of a given match to explain what the blummin hell is gonna happen in the upcoming match. Whether it is ten variables or 910 variables, the stats software will work out what is most important variables in predicting the upcoming match. They will be assigned a weighting. We are all familiar with Paul Steele's book with his power ratings. Well they are very simplistic stats models. His model is 5*this + 4*that + 9*other-thing How he arrived at these numbers I have no idea but I suspect they are far from optimal. Were he to pump them through a stats package he would have got very close to the optimum answer. Which I suspect would have raised the predictive power of what he was trying to do. Do that make sense? Quote Link to comment Share on other sites More sharing options...
Datapunter Posted August 11, 2008 Author Share Posted August 11, 2008 Re: Tennis rating/info/assessment system Yes, but, imagine you can get to a 100% model, and on a match that model says each player has 50% chance of winning, and the available odds are 1.95 on each, not much use is it ? Or in other words, for any metric to have any value in our context, it needs to be tested against the world of odds. Without that test how do you know a metric has any value ? A big server has an advantage over an average server, don't need any testing to see that, common sense will do, but the odds compilers also know this so the advantage will show up in the odds of the big server being lower than the other player. Maybe i missed something in the term "statistical model", but i'm missing the relationship metric vs betting. Maybe we should take a couple of metrics like "big server" and dig in to the bottom, make it more specific ? Quote Link to comment Share on other sites More sharing options...
Smokin Joe Posted August 11, 2008 Share Posted August 11, 2008 Re: Tennis rating/info/assessment system Yes, but, imagine you can get to a 100% model, and on a match that model says each player has 50% chance of winning, and the available odds are 1.95 on each, not much use is it ? DP, I haven't been involved in this thread so far, so you can tell me to sling my hook...... I would say that the model has been of some use in your example. The model has told you that the fair price of either competitor is 2.0, the fact that you can only get 1.95 about either of them means that if you were to back either competitor you would lose (in the long term). Therefore the model has saved you losing 5% of your stake. No model is going to advise a bet in every game, as just occasionally the bookies do price up event correctly :) Quote Link to comment Share on other sites More sharing options...
Datapunter Posted August 11, 2008 Author Share Posted August 11, 2008 Re: Tennis rating/info/assessment system It's a public thread Smoking, the actual database and programming will remain with those that put the work in, but anyone is welcome to join in the conversations. And i'm with you completely, the perfect model will bring the bookies errors to the surface, or in other words will show where the general public, whom to a large extent dictate the odds by "weight of money", is betting too much on the wrong side. And considering an average overround of 10% value will only show up in a minority of cases. But a statistical model is built up of a series of metrics. For each individual metric, I'm just saying, or asking really, how do you know any metric has value unless you somehow test it against the world of odds ? Quote Link to comment Share on other sites More sharing options...
Matthew Posted August 11, 2008 Share Posted August 11, 2008 Re: Tennis rating/info/assessment system You don't know that a given metric has any importance until you have tested it. For all we know performance against the type of player someone is to face might be completely irrelevant, I was merely thinking of something I hadn't encompassed in my previous efforts. The stats software will sort out what is important and what is not. But we need to feed the software with as many metrics as possible, despite how pointless we may think they are. Common sense will dictate where effort is directed in terms of calculating these metrics. I was just tryng to get a wishlist together and working through that...........I do however think Mr Intensity has a wealth of knowledge and if he thinks it's an important factor in deciding the result in a tennis match then it's at least worth throwing in. Possibly we are agreeing in what we are trying to say - it's just getting a little lost in translation. Smokin Joe is right - if we had a perfect model and it said the match was 50/50 but we could only get 1.95 at best then it's a no bet surely?!?! I think startegy and money management is a long way away - we need to get together as much data as possible with as many "pre match" facts as possible to derive our own % chance. Back testing will help with our decision about hunting value or not. My previous experience suggests that finding value was tough - but then my previous effort included bookies odds, which was the major factor and as such the outputs were very close to the bookies odds, making value hard to find. Quote Link to comment Share on other sites More sharing options...
Mr Intensity Posted August 12, 2008 Share Posted August 12, 2008 Re: Tennis rating/info/assessment system Right I'll think of some categories and probably get it done Monday/Tuesday. Might take a smidge longer as I've got a lot on this week:$:$ Quote Link to comment Share on other sites More sharing options...
Datapunter Posted August 12, 2008 Author Share Posted August 12, 2008 Re: Tennis rating/info/assessment system Yeah, we're going towards the same goal, just taking a slightly different path. Should generate some interesting conversations. No reason we can't explore multiple ways of doing things. No worries Mr I , this will be a long term thing anyhow. Quote Link to comment Share on other sites More sharing options...
Matthew Posted August 12, 2008 Share Posted August 12, 2008 Re: Tennis rating/info/assessment system I'd be up for a series of different approaches. I could certainly benefit from thinking about things in a different way. I'm mindful that not too many people would attack the problem the way I would.....should be fun. I've been playing around with my data that I have at the mo and I think it's fairly encouraging that we do not need to turn over a massive deficit to see profit from a tennis system. I suppose it depends on your approach but consider the simple strategy of backing the highest ranked player. From the 17,793 matches I have, were you to back the highest rank player then you would have the following system stats at 1pt level stakes. No Bets: 17,793 Stake: £17,793 Return: £17,204.69 Profit: -£588.31 Yield: -3.3% Now lets take it a step further and consider the ranking of the players involved, banded into groups. If both players are in the top 10 and you blindly backed the highest ranked. No Bets: 266 Stake: £266 Return: £258.50 Profit: -£7.50 Yield: -2.8% If one player is in the top 10 and the other ranked 11-20 then: Again backing the highest ranked player....... No Bets: 370 Stake: £370 Return: £367.72 Profit: -£2.28 Yield: -0.6% If one player is in the top 10 and the other ranked 21-30 then: No Bets: 443 Stake: £443 Return: £447.96 Profit: £4.96 Yield: 1.1% If both players are ranked 11-20 then: No Bets: 113 Stake: £113 Return: £108.73 Profit: -£4.27 Yield: -3.8% If one player is ranked 11-20 and the other ranked 21-30 then: No Bets: 315 Stake: £315 Return: £313.66 Profit: -£1.34 Yield: -0.4% I dunno about anyone else but this doesn't seem like anything to hard to overcome. It appears that using a rudamentry system of banking the highest ranked player where one is in the top 10 and the other ranked between 21-30 returns a profit over 440 bets. I'm sure we can come up with something better. Quote Link to comment Share on other sites More sharing options...
Mr Intensity Posted August 12, 2008 Share Posted August 12, 2008 Re: Tennis rating/info/assessment system If I was trying to improve on that I'd say... Exclude the first match in the tournament of the higher ranked player Exclude when the higher ranked player isn't favourite Exclude the higher ranked player when playing a qualifier Just out of interest Matthew, how would my system do backing top 10 and top 20 against the rest? Quote Link to comment Share on other sites More sharing options...
Matthew Posted August 12, 2008 Share Posted August 12, 2008 Re: Tennis rating/info/assessment system I can have a look at that later on today if I have time. Do you use the exceptions you bulleted above? Quote Link to comment Share on other sites More sharing options...
Mr Intensity Posted August 12, 2008 Share Posted August 12, 2008 Re: Tennis rating/info/assessment system I always like opposing a few big names in the first round, especially in lower valued tournaments. Very simple you don't know if the big names are motivated that week, where a lesser guy is more likely to be up for it and has the chance to get a big scalp. Someone told me about a system where you just back the lower ranked favourite. I liked the idea behind it but only tracked it for a couple of months (didn't do very well in that time). In 2006 (I think) I ran a system called 'The vlaue in qualifiers' which was very simply backing any qualifier in the 1st and 2nd round using fixed profit staking. Over about 400 bets it had around a 16% yield and looking at the numbers afterwards I could have over 20% yield backing in 1st round to fixed stake. I haven't really followed it since then but when I'm looking at a first round match I do normally look at the qualifier. Quote Link to comment Share on other sites More sharing options...
Datapunter Posted August 13, 2008 Author Share Posted August 13, 2008 Re: Tennis rating/info/assessment system Did a bit of work on those qualifiers but after backtesting 1,5 seasons it was showing a loss, so gave that up. We can look at it again with several seasons data. I'm intrigued by something: Mr I what would be your definition of "big server" based on stats info only ? Quote Link to comment Share on other sites More sharing options...
Mr Intensity Posted August 13, 2008 Share Posted August 13, 2008 Re: Tennis rating/info/assessment system I'm intrigued by something: Mr I what would be your definition of "big server" based on stats info only ? Based on stats I've no idea as I don't really follow the %'s:loon:loon. For me it would just be from watching them play - not just the mph or aces but guys who can rely on a big serve when they need it. Obviously guys like Karlovic but I'd also include guys like Djokovic. Just looking at the stats isn't enough IMO. Just having a quick look at 1st serve % won, Mahut 72%, Seppi 71% and Verdasco 69%. Mahut is a big server but playing badly, Seppi serves like a girl and Verdasco can be a big server but often chooses not to. Can't read much into those numbers alone. Quote Link to comment Share on other sites More sharing options...
Datapunter Posted August 13, 2008 Author Share Posted August 13, 2008 Re: Tennis rating/info/assessment system from watching them play But how do you "measure" a players performance in this area ? Firstly i assume you use your brain to determine who's a "big server", fair enough, but then secondly do you actually keep notes or do you rely on memory ? In both cases the human brain is a pretty bad measuring instrument, very unreliable, biassed, prone to biological fluxtuations, etc... :lol I'm not dismissing using common sense at all, i'm just looking for a starting point based on numbers and then apply common sense on top of that. What intrigues me with a "big server" is this: you can measure a player serve ability based on 1st serve % , return % and maybe a few more factors but this is not enough is it. Imagine a player who gets 95% first serves in but then they all get retured and he loses most of them. Compared to a player who gets "only" 70% first serves in but actually wins most 1st serve points. Which one is the "big server" ? Just thinking out loud here. This is why i'm looking for a definition of a "big server" and with that definition comes some means of measuring serve ability. Then on top of that you watch a player to see if he can live up to his ability in all situations. Hypothetical: For example it is possible for a payer to have a good serve but only when he's up against a lesser or equal opponant. When faced by a much better player there's something which lowers his actual serve performance. This would be something very difficult to measure based on numbers only and here's where watching matches comes in. "Player X has great serve ability but gets intimidated by top players." Quote Link to comment Share on other sites More sharing options...
Mr Intensity Posted August 13, 2008 Share Posted August 13, 2008 Re: Tennis rating/info/assessment system Good points. What intrigues me with a "big server" is this: you can measure a player serve ability based on 1st serve % , return % and maybe a few more factors but this is not enough is it. Imagine a player who gets 95% first serves in but then they all get retured and he loses most of them. Compared to a player who gets "only" 70% first serves in but actually wins most 1st serve points. Which one is the "big server" ? I think it's obvious the 70% who wins more points is the big server. The 95% is a clay courter who can barely server:unsure I do it all on memory - if I can't remember a player I'll try to find some clips of them on youtube;). Without seeing the stats, I think the best stat for determining a big serve would be % of first serves that are unreturned. But not sure if that data is available:unsure Hypothetical: For example it is possible for a payer to have a good serve but only when he's up against a lesser or equal opponant. When faced by a much better player there's something which lowers his actual serve performance. This would be something very difficult to measure based on numbers only and here's where watching matches comes in. "Player X has great serve ability but gets intimidated by top players." That would definately be a mental thing. Just thinking - Murray tends to serve more flat serves against better opponents IMO. So his % may go down but a lot less come back or come back with a weak return. Also think surface has a bit to do with it. Guy I mentioned before Verdasco bugs the hell out of me, got a massive serve but on clay just kicks it in pretty much all the time. Quote Link to comment Share on other sites More sharing options...
Sgt.Sunshine Posted August 13, 2008 Share Posted August 13, 2008 Re: Tennis rating/info/assessment system Doh, been away for a while and hadn't checked the thread until today. I have been staring at my racing database all day and can't take any more computer time. Will catch up asap, but looks interesting so far....:ok Quote Link to comment Share on other sites More sharing options...
Datapunter Posted August 13, 2008 Author Share Posted August 13, 2008 Re: Tennis rating/info/assessment system I do it all on memory Gonna have to do something about that :lol We should have a very simple webpage where people can enter notes related to a player and match. Obviously going to be filled with very subjective info. But in itself should be interesting to see if different people "see" different things in a match. a big serve would be % of first serves that are unreturned Agree, that would be a basic measure i'll be looking for. "Big server" is the one that can consistently turn a serve into a win point. As to how to measure that... ermmm... to be continued :lol Quote Link to comment Share on other sites More sharing options...
Matthew Posted August 13, 2008 Share Posted August 13, 2008 Re: Tennis rating/info/assessment system Would "Total Service Points Won" suffice? In conjunction with # Aces perhaps. If a player wins 70% of service points and makes, on average, 10 aces a match - are they a "big server"? Quote Link to comment Share on other sites More sharing options...
Matthew Posted August 13, 2008 Share Posted August 13, 2008 Re: Tennis rating/info/assessment system Just out of interest Matthew' date=' how would my system do backing top 10 and top 20 against the rest?[/quote'] Based on 1pt level staking. If there is a match featuring at least one top 10 player and you blindly backed the highest ranked. No Bets: 3,763 Stake: £3,763 Return: £3,733.78 Profit: -£29.22 Yield: -0.8% (max odds) -4.3% (avg odds) If there is a match featuring at least one top 20 player and you blindly backed the highest ranked. No Bets: 6,564 Stake: £6,564 Return: £6,413.83 Profit: -£150.17 Yield: -2.3% (max odds) -5.8% (avg odds) If there is a match featuring one top 10 player against a player outside the top 10 and you blindly backed the highest ranked. No Bets: 3,497 Stake: £3,496 Return: £3,475.27 Profit: -£21.73 Yield: -0.6% (max odds) -4.1% (avg odds) If there is a match featuring one top 20 player against a player outside the top 20 and you blindly backed the highest ranked. No Bets: 5,815 Stake: £5,815 Return: £5,678.88 Profit: -£136.12 Yield: -2.3% (max odds) -5.8% (avg odds) Not too healthy but better than blindly backing the highest ranked. Quote Link to comment Share on other sites More sharing options...
Matthew Posted August 13, 2008 Share Posted August 13, 2008 Re: Tennis rating/info/assessment system Someone told me about a system where you just back the lower ranked favourite. I liked the idea behind it but only tracked it for a couple of months (didn't do very well in that time). Using £1 level stakes. Were you to back the favourites on occasions when the favourite had a lower rank. No Bets: 3,110 Stake: £3,110 Return: £3,237.08 Profit: £127.08 Yield: 4.1% (max odds) 0.0% (avg odds, 42p profit!!!) Seems like there might be something in this as a simple system. Not sure how we would encompass in to anything we do but worth knowing. Mr I, has that someone given you an update on how they have been getting on with it? Quote Link to comment Share on other sites More sharing options...
Sgt.Sunshine Posted August 13, 2008 Share Posted August 13, 2008 Re: Tennis rating/info/assessment system Just a quick one: Matthew, are you using the odds from Tennis-Data.co.uk? If you are (and this maybe also applies if you have another source), do you plot profit against time, or just look at the final yield, because many of these "well known" "systems" yielded highly at the beginning and tail off for the last few years. Another word of caution if you use the tennis-data dataset - the bookies used changed around 2006 I think, most importantly with the introduction of Pinnacle odds. This upsets the "equilibrium" of testing systems. This tailing off is a very common problem now unfortunately, with more and more people analysing the same datasets (unless it's just the systems I test:lol). Quote Link to comment Share on other sites More sharing options...
Sgt.Sunshine Posted August 13, 2008 Share Posted August 13, 2008 Re: Tennis rating/info/assessment system The way I looked at the "big server" problem was to normalise all the variables to the mean (although not really necessaty I guess), and call the top x% big servers (using aces and first serve pts won%). I got as far as listing the players and thinking that it looked about right. I did the same for "good returners" (Andy Murray came out on top). Never took it any further, but considered looking at how certain players played against big servers compared to their normal performance. Quote Link to comment Share on other sites More sharing options...
Sgt.Sunshine Posted August 13, 2008 Share Posted August 13, 2008 Re: Tennis rating/info/assessment system Mr I, don't know if you've seen this: http://tt.tennis-warehouse.com/showthread.php?t=58284 I looked it up a few years back when I was trying to get my head round the problem. Quote Link to comment Share on other sites More sharing options...
Mr Intensity Posted August 13, 2008 Share Posted August 13, 2008 Re: Tennis rating/info/assessment system Mr I, don't know if you've seen this: http://tt.tennis-warehouse.com/showthread.php?t=58284 I looked it up a few years back when I was trying to get my head round the problem. Yeah those are probably better categories than mine. Quote Link to comment Share on other sites More sharing options...
Mr Intensity Posted August 13, 2008 Share Posted August 13, 2008 Re: Tennis rating/info/assessment system Using £1 level stakes. Were you to back the favourites on occasions when the favourite had a lower rank. No Bets: 3,110 Stake: £3,110 Return: £3,237.08 Profit: £127.08 Yield: 4.1% (max odds) 0.0% (avg odds, 42p profit!!!) Seems like there might be something in this as a simple system. Not sure how we would encompass in to anything we do but worth knowing. Mr I, has that someone given you an update on how they have been getting on with it? Yeah that's not too bad. No updates, think he stopped doing it quite a while back to move onto something else. Quote Link to comment Share on other sites More sharing options...
Mr Intensity Posted August 13, 2008 Share Posted August 13, 2008 Re: Tennis rating/info/assessment system This tailing off is a very common problem now unfortunately, with more and more people analysing the same datasets (unless it's just the systems I test:lol). I think this happened to the qualifiers - quite a few times this year I've looked at a qualifier in the 1st round and been surprised at how short they are:sad Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.