Betting system promotion

[Guest adamskitwo]

Hi, I wrote on this board a while back about a computer system I had almost developed, and decided that I would retire into the shadows until the system was fully developed....and now it is. This system is designed currently to predict the results of Premiership games although with more stats this could arguably be extended to cope with any game where the result could be a win, draw or a loss. The system takes in an array of past statistics and predicts the probability of a win, a draw and a loss occurring. This makes it useful to not only predict the favourite result, but also as a device to try and undercut odds. As an example, if I took the result predicted from the system to be the odds on result. I would have a result which I could call 'its prediction'. This can also be applied to the bookmakers odds also. I have found that comparing the predicted results of my system and the bookmakers prediction against the actual result, my system predicts correctly 50% of the time while bookmakers such as William Hill, Ladbrokes, SuperSoccer, Coral, etc tend to hover around 40% correct. As you can see this is a significant improvement, and seems on the surface to be very marketable. I figure that it is best to sell it to bookmakers given it has odds prediction included. From my research, the big boys, William Hill, etc, seem to predict their own results, and the small fish, (local bookies), use coupons from the big boys and dont want the risk of a computer system. It seems like to sell it, I would have to go to the big boys. So this is where it becomes a little sticky....so here are my questions... Extremely grateful for any help. 1. Do you think I am marketing the software at the right person? (the bookies). I figure it could make a killing and it would be wiser as usually selling s/w to the general public results in everyone pirating it in the end anyway. It doesn ideally need an administrator also. 2. How can I possibly go about marketing this software to the big bookmakers? 3. Does anyone have any contacts that I could use just to run the idea past them? 4. Does anyone else have any comments/suggestions/questions about the system? Thanks, Adam PS. It is a windows based neural network based system. The reason I dont use it to bet myself is that the risk is still there (especially with the bookies markup), and that I dont have the money to invest in casual gambling currently.

[Guest OddsAgainst]

Re: Betting system promotion Sorry to rain on your parade but I don't think bookmakers will be interested. They employ professional odds compilers who are very good at their job, otherwise they would not make any profits. Why not proof the results here for a season? If you can demonstrate that your system can indeed beat the odds compilers then either the bookmakers will buy you out to keep the system out of the punter's hands or we will buy it. But you need to prove it's worth over hundreds of live bets first.

[Joe]

Re: Betting system promotion

But you need to prove it's worth over hundreds of live bets first.

I think given the experience of bookmakers in some cases stretching over 100 years, this figure will be more like millions. William Hill, Ladbrokes and Corals will no doubt use fairly sophisticated algorithms in generating their predictions, although nothing so complex that a knowledgable punter could not follow. However, as Odds says, it is highly unlikely that they will wish to give these up in favour of something new produced by one person with an interest in beating them. I suppose it would be a bit like Galileo attempting to convince the church that in fact the world isn't flat and the Sun doesn't do round it. Not that I would imagine that William Hill would wish to burn you at the stake for heracy, of course. ;) To my mind, if you have a decent product, you'd be better off simply making money from it yourself, or perhaps setting up an advisory service.

[Guest OddsAgainst]

Re: Betting system promotion I meant prove its worth to us over hundreds of bets. ;)

[Guest adamskitwo]

Re: Betting system promotion Thanks a lot for your inputs so far. Unfortunately, I think you are probably right, and so from this day on, I will try to publish all Premiership predictions by the system, including odds on this thread. The reason for my confidence (and impatience) ;-) however is because when everything is said and done, it does seem to have predicted a lot better (10%) than the bookies over the previous 9 years (all tested data into the system). Not quite millions, but about 4,200 results to my knowledge!!! Also, I think that the whole point of developing a system using neural networks is that it does cut out the bias that maybe even professionals encounter. So anyway, we will see; the first man vs machine battle... Keep all suggestions coming though...all are welcome. P.S. Believe me, I felt a bit like Galileo explaining the system on the phone to someone the other day ;-)

[Guest OddsAgainst]

Re: Betting system promotion Glad to hear you are going to proof them, there will be a lot of interest. Do you intend to publish your own probability estimates and then back those outcomes where the bookie's odds are higher or is there more to it, eg. home wins only or favs only?

[Guest madmick]

Re: Betting system promotion Adamskitwo Some comments: I have looked at this area myself and believe it might be possible to achieve something here, so maybe that is some reassurance - or maybe not ;) I agree with the lads, the bookmakers are likely to be pretty sceptical about it Proofing the system here is a pretty good way to go if you intend to set up a tipping service or to sell a package - personally I would recommend the tipster service route. 9 seasons data sounds very like what Joe has on his site ;) Is this the data you used? Did you pre-process it in any significant ways? I presume I don't need to ask you whether or not your test data was completely seperate from your training data, and that your net isn't too big so that it has actually 'memorised' the training data? If I understand you correctly you are saying that you are predicting at about 50% accuracy whereas the bookies odds would suggest 40%. This would be with overround removed, so that you are in fact achieving an edge of 25% (i.e. 50/40)? If you can achieve this kind of level, you don't need to sell anything, you can make a fortune from betting your own money, probably doubling your bank every 3-4 weeks or so

[Guest adamskitwo]

Re: Betting system promotion Hi, Firstly, I will intend to back all odds which are lower than the bookies because regardless of home/away etc, the system has been setup to take that into account and so any probablilities should be good for usage. Whether that results in any bets at all, Im not sure!! As for the data, I bought a Virgin premiership guide to get all archive data for the premiership. Painstakingly slow to enter though... For processing, the system starts with a blank DB, resets all weightings, and then reads in result by result predicting each outcome until it gets to the end of the data. It is obviously starting from scratch as it starts off pretty badly before it acquires its intelligence. Lastly, the bookies overround shouldnt affect any prediction accuracy total. This is because I used their odds on prediction to be their 'prediction'. This shouldnt be affected by any overround. eg. If the real odds are 3/1, 5/1, 2/1, and the bookies list them as 2/1, 4/1, 3/2 then the 'prediction' should remain the same (Away win). So its still means 50% vs 40%. Adam

[Guest OddsAgainst]

Re: Betting system promotion Adam, just a comment on the expected number of bets. If we assume that the bookies odds are indeed very accurate (the fact they remain in business supports this) then we should expect any system to produce few bets as we would not expect the bookies to make many mistakes. Do the results of your testing agree with this?

[Guest madmick]

Re: Betting system promotion adam I am slightly concerned by your statement that the system reads in result by result predicting each outcome until it gets to the end of the data. It is obviously starting from scratch as it starts off pretty badly before it acquires its intelligence unless you are using some sort of massively parallel learning through time (i.e. recurrent network) structure, I don't know why this should be. I presume you are using some sort of feed forward MLP network trained with back-propagation (the most common type of ANN), or something similar (e.g. radial basis or suchlike). If so, then your network wouldn't be aware of a time structure on the data - you present it with the training data and then test it with the test data. You seem to have something different, unless I have got this wrong? Maybe you could tell us a little bit about what type of network you are using (no need to reveal your exact inputs ;) )

[Guest adamskitwo]

Re: Betting system promotion As relatively untested as the system is, it does seem to show that the bookies usually offer the lowest odds to make their money. As such, any personal use would be strictly to try and catch the bookmakers out as often as possible. Sorry about any confusion btw. The system uses a Backpropagation network. So you feed all inputs in for example, it then feeds it through the network to give outputs, (its predicted result for the match at that given time), and then compares with the actual output. It then calculates error and adjusts the weights. So I put each result chronologically into the system one by one until the end, at which point it should have learned as best it can. This is the point I was making earlier when I said it predicts badly at first; as it has had limited error correction at this point. Adam

[Guest adamskitwo]

Re: Betting system promotion Here are the system predictions for this w/e. I put in the learning rate wrong so take these with a pinch of salt. The system is tehnically running at 47% accuracy instead of 50% so I dont know what impact thatll have..... H D L 1/1 Tottenham 3.2/1 Man City 4.8/1 2.4/1 Aston Villa 3.2/1 Chelsea 1.1/1 1.8/1 Bolton 2.5/1 West Ham 1.7/1 1.0/1.1 Charlton 7.3/1 Birmingham 1.8/1 1.4/1 Everton 2.4/1 Liverpool 2.3/1 1.2/1 Fulham 4.9/1 Newcastle 1.5/1 1/2.2 Man United 3.3/1 Blackburn 11/1 2.6/1 Middlesborough 2.9/1 Arsenal 1.1/1.0 1.6/1 Southampton 12/1 W.B.A 1/1 Apologies for confusing you for decimal places etc ;-)

[Guest madmick]

Re: Betting system promotion adam I hope I am being excessively pessimistic, but it sounds to me like you are doing the following: training the network with N matches, try to predict the result for match N+1 re-train the net with N+1 matches, try to predict match N+2 etc I think you'll find, if this is what you have done, that the network has in fact overtrained and just learned the pattern presented, i.e. it will generalise poorly. The acid test will be real-time testing, but I suspect you will disappointed. This is not meant to dissuade you - I think NN's could be a very useful approach to the area, just be careful - there is a lot of skill required to get useful results, and they are not a panacea. What NN package are you using out of interest? PS. If you prefer to discuss this offline, send me a message ;)