I came across this site: http://www.prosoccer.gr/
Anyone any thoughts on the effectiveness of neural networks in finding the true odds of football matches and other sporting events?
Jeff
Neural networks
Since the start of the year I have been deep into some very detailed work on market pricing and volatilty. I am messing around with some neural net stuff, but I don't rate it at the moment as it's throwing off some odd results. I already understand market models very well and the neural net stuff doesn't really seem to be adding much value.
Could easily be a case of GIGO.
Could easily be a case of GIGO.
-
rubysglory
- Posts: 309
- Joined: Thu Nov 04, 2010 7:02 am
I have never fully understood use of the term "Neural Network" from a form asessment perspective. Pre the computer age, a lot of form and price assessors used index cards.This enabled future assessment that often used a complex penalty and bonus system. This process was called form or price assessing. Am I wrong to think that a "Nueral Network" is just a new age computurised term for an index card ?
rg
rg
How to use a neural network to create a trading algorithm: http://blog.danmachine.com/2012/01/neur ... rithm.html
The author is Betfair's Principal Research Engineer: https://docs.google.com/a/danmachine.co ... WQ5Njc0OTM
Jeff
The author is Betfair's Principal Research Engineer: https://docs.google.com/a/danmachine.co ... WQ5Njc0OTM
Jeff
He is obviously a very switched on guy..but it just seems over complex to me..
Dont you think that sometimes you can overcook something; whereas "Good Enough" is good enough? By the time he has worked out what he is doing he will have missed the boat!
I wonder what you would find if you were to analyse the top 500 traders in the country and the methods/strategies they employed? I can't see many going to the lengths used that that guy?
Regards
Peter
PS - I would expect a fair chunk of those 500 to be trading on insider or privileged knowledge...and maybe others using court side tactics!
PPS: I remember studying the maclaurin series as part of a subject at university..I was trying to work through a difficult example at my office and a work colleague (Phd) said let me show you...he rattled the answer off in no time...
The next day I saw him sat at his desk scratching his head..When I asked him what was troubling him.. he told me he was trying to build a small fence at the front of his house and he was trying to calculate how deep he should dig the hole for the posts/concrete..I said 2 foot should do it. He replied by saying how do you know the wind won't blow it over?
He really couldn't get his head around the fact that it would be OK and adequate. It was good enough..
Thats when I remember a saying my mum told me " Common sense isn't that common"
Dont you think that sometimes you can overcook something; whereas "Good Enough" is good enough? By the time he has worked out what he is doing he will have missed the boat!
I wonder what you would find if you were to analyse the top 500 traders in the country and the methods/strategies they employed? I can't see many going to the lengths used that that guy?
Regards
Peter
PS - I would expect a fair chunk of those 500 to be trading on insider or privileged knowledge...and maybe others using court side tactics!
PPS: I remember studying the maclaurin series as part of a subject at university..I was trying to work through a difficult example at my office and a work colleague (Phd) said let me show you...he rattled the answer off in no time...
The next day I saw him sat at his desk scratching his head..When I asked him what was troubling him.. he told me he was trying to build a small fence at the front of his house and he was trying to calculate how deep he should dig the hole for the posts/concrete..I said 2 foot should do it. He replied by saying how do you know the wind won't blow it over?
He really couldn't get his head around the fact that it would be OK and adequate. It was good enough..
Thats when I remember a saying my mum told me " Common sense isn't that common"
Possibly, but there's something to be said for having a bot that can make you regular profits without any intervention on your part.PeterLe wrote:He is obviously a very switched on guy..but it just seems over complex to me..
I believe Steve Jobs once said that 'Good enough is not good enough', and in a sense I think he was right - one should always strive to improve.PeterLe wrote:Dont you think that sometimes you can overcook something; whereas "Good Enough" is good enough?
But 'good enough' is good enough in the sense that it's better to have an imperfect approach that makes you money than a perfect one that you're still formulating...
Jeff
This is an interesting blog which I found quite inspiring a few years back, not that I do anything with neural networks at the moment. Not sure why it stopped. I did email the author a while ago but got no response. Wonder if they read this forum...
http://dynamicnotions.blogspot.com
http://dynamicnotions.blogspot.com
-
rubysglory
- Posts: 309
- Joined: Thu Nov 04, 2010 7:02 am
Spot on Peter. To me these 'Neural Networks' are an over complication of a process that was once done manually. They are computerised index cards. I am however open minded enough to be convinced otherwise. Don't get me wrong, computers have simplified modelling and the amount of data that can be processed, but I wonder some times whether people get caught up with all the bells and whistles and become removed from the basics. Like Euler said - GIGO !PeterLe wrote:He is obviously a very switched on guy..but it just seems over complex to me..
rg
It's hard to make a living in this game and personally i think because of this people tend to think the answer lies in extremely complex methods.
Im not saying that it doesn't work, just that there are much more simple ways of achiving it.
It really isn't rocket science and i think alot of newcomers are looking for a secret that doesn't really exist.
Im not saying that it doesn't work, just that there are much more simple ways of achiving it.
It really isn't rocket science and i think alot of newcomers are looking for a secret that doesn't really exist.
-
superfrank
- Posts: 2762
- Joined: Fri Aug 14, 2009 8:28 pm
very true. i think it's because some people treat trading as an intellectual challenge rather than the mindset challenge that it really is.freddy wrote:It's hard to make a living in this game and personally i think because of this people tend to think the answer lies in extremely complex methods.
you only have to figure out a little bit about markets, and then apply the edges in a disciplined way, in order to be profitable.
What about full-blown artificial intelligence? Human intelligence would surely be no match for super-human intelligence in the markets?
Trading probably doesn't need this level of complication, but what about handicapping (pricing the true odds), I think that's more complex isn't it? Could there be a role for AIs there?
Alan Woods (the world's greatest horse handicapper), used a combination of subjective expertise and a computer model developed over several years to rack in his huge profits on the Hong Kong races:
http://www.tonywilson.com.au/writing/alanwoods.php
"Alan's team employs a dozen or so staff to review, analyse and compile data for every horse running in Hong Kong, every time it runs. They are a disparate bunch, scattered across Asia and Australasia, coming together only in the cyber-confines of an email inbox. The data is then plugged into a computer program, and using a formula based on past results that has been refined over many years, a probability for every horse in every race is calculated."
"Producing the formula is the tough bit. It has been mathematically chiselled out of all the factors that Alan and his team have determined decide races. Each factor is a coefficient in the formula. Some factors - gender, track, distance, weight, last-start result - are objective and can be collected from the humble form guide. The key is getting the weighting of these objective factors right."
"Other factors are more subjective, which is why Alan's team employs expert analysts to watch every horse in every race. 'We had a factor called bad rides. We had a factor called not trying. If a couple of horses disputed the lead together, the guys would give it numbers for that. Premature speed - if the horse went too fast too early. Late speed - if it came home very fast in the last 400.' It's the kind of stuff any experienced punter might consider. What the computer teams do is systemise the process, eliminating sentiment and superstition and minimising human error."
Trading probably doesn't need this level of complication, but what about handicapping (pricing the true odds), I think that's more complex isn't it? Could there be a role for AIs there?
Alan Woods (the world's greatest horse handicapper), used a combination of subjective expertise and a computer model developed over several years to rack in his huge profits on the Hong Kong races:
http://www.tonywilson.com.au/writing/alanwoods.php
"Alan's team employs a dozen or so staff to review, analyse and compile data for every horse running in Hong Kong, every time it runs. They are a disparate bunch, scattered across Asia and Australasia, coming together only in the cyber-confines of an email inbox. The data is then plugged into a computer program, and using a formula based on past results that has been refined over many years, a probability for every horse in every race is calculated."
"Producing the formula is the tough bit. It has been mathematically chiselled out of all the factors that Alan and his team have determined decide races. Each factor is a coefficient in the formula. Some factors - gender, track, distance, weight, last-start result - are objective and can be collected from the humble form guide. The key is getting the weighting of these objective factors right."
"Other factors are more subjective, which is why Alan's team employs expert analysts to watch every horse in every race. 'We had a factor called bad rides. We had a factor called not trying. If a couple of horses disputed the lead together, the guys would give it numbers for that. Premature speed - if the horse went too fast too early. Late speed - if it came home very fast in the last 400.' It's the kind of stuff any experienced punter might consider. What the computer teams do is systemise the process, eliminating sentiment and superstition and minimising human error."