Predicting Starting Prices using data analytics and Machine Learning

Vikas     09 Jul 14 17:49 
Not sure if this is the right place for this topic or not but have to start from somewhere, What are chances that Machine Learning and Mathematical modelling can be used to predict starting prices of a horse in a race using past data available publicly?

www.betegy.com and www.smartodds.co.uk do same for football! I somehow think Horse Racing seems more suited for


........... but starting prices seems to be always based on past performances and current conditions

Obviously objective is to predict with enough accuracy that money can be made out of it, I know there are loads of tipsters but are there any companies working on something like this?

All thoughts are welcome!

Your statement in bold, would relate to sp forecast.

Actual starting price is determined by weight of money. Hth

I think this is an interesting topic. You could perhaps feed a set of best morning prices and SP's (although R'4 would need accounting for unless you simplifying by training only on races with no NR's) into a net along with a host of factors you feel may be relevant to price movement and then see what the NN could come up with. This would be a situation where an understanding of the masses as well as form factors would be useful.

predicting the market in the last 15 minutes is difficult, its about how the horses look, travel to post and the insider angle on what the jockeys orders are

.......and the money

wisdom of the crowd

Jul 9, 2014 -- 11:58AM, Facts wrote:

Vikas     09 Jul 14 17:49  Not sure if this is the right place for this topic or not but have to start from somewhere, What are chances that Machine Learning and Mathematical modelling can be used to predict starting prices of a horse in a race using past data available publicly?www.betegy.com and www.smartodds.co.uk do same for football! I somehow think Horse Racing seems more suited for ........... but starting prices seems to be always based on past performances and current conditionsObviously objective is to predict with enough accuracy that money can be made out of it, I know there are loads of tipsters but are there any companies working on something like this?All thoughts are welcome!Your statement in bold, would relate to sp forecast. Actual starting price is determined by weight of money. Hth

Hi Facts, Thanks for commenting, I agree mathematically SPs are decided by weight of money but volume of that money which is coming from different punters etc on different runners is again based on past performances and current conditions?

As far as SP forecast is concerned, they are just indicative and mostly real SPs are very out of the way.

Just a point about SP and "weight of money":

Imo, this is not sorted out on the course, but comes from "elsewhere".

One day at Wolver, I was stood alone by the three bookies who had turned up. We were inside the hall because it was raining.
An overhead tv was chuntering on about "massive gamble on x at Wolverhampton". At once, all three boards flicked over in unison with shortened prices on x.
Nobody had come near us to place a bet.
The so-called gamble actually veered off after the chute and came no where.

Years of asking anybody and everybody about sp have led me to believe that, sadly, it is "Machine"- driven nowadays.
My guess is that the betting shops and the exchanges control so-caled "SP". This may well be described as "weight of money", but, for sure, 'tis NOT money on course.

of course it isnt...you've got hundreds of thousands traded on here in the last 5 mins before off for even the dullest races...it's big syndicate money...they only play once there's serious liquidity in the markets, and they aint playing on-course at Wolves!

Agreed. It's off course influence( real or ' imagined ' )

Jul 9, 2014 -- 12:02PM, OliasOfSunhillow wrote:

I think this is an interesting topic. You could perhaps feed a set of best morning prices and SP's (although R'4 would need accounting for unless you simplifying by training only on races with no NR's) into a net along with a host of factors you feel may be relevant to price movement and then see what the NN could come up with. This would be a situation where an understanding of the masses as well as form factors would be useful.

Hi OliasOfSunhillow, Yes I also think morning prices or tissue prices + other form factors should show some patterns.

Hi Vikas its a data set I have been examining for the past 3 years. There are some pointers amongst the data although I have not as yet fed it into a NN.

IS it data that you have collected yourself using Betfair APIs OR from external sources liek racingpist/timeform or combination of both?

^^^^

RACING PIST

The former Vikas

iam on large print forum how do i get back to the original small print  forum

weight of money as a reason for sps is a bit of an obvious statement, finding the reasons for the money is the objective, eg when a horse is heavily backed and then is seen in the race with an easy lead for the first time that is insider knowledge, it would be nice to think you could model these things but in reality you just dont know as an outsider, things that can be modelled will usually be factored in much earlier than the last 15 minutes

I'm more into looking at ways to predict those I think are going drift or shorten during the day rather than try to predict actual prices

@pawras I think you and I meant the same thing, so if I can predict SPs to certain degree of accuracy and price 20 minutes before the race is either very high or low then my prediction then it should steam or drift ideally ! If outcome of ML is identifying steamers and drifters then we will have to define how much movement from current price will qualify horse to be drifter or steamer and for that we will need some number, that's what I meant by SP, I agree in some cases horse may drift and them steam back to same price.

the key thing is identifying a good criteria(s) as to why they're going shorten or drift.

Jul 9, 2014 -- 12:22PM, Rider wrote:

predicting the market in the last 15 minutes is difficult, its about how the horses look, travel to post and the insider angle on what the jockeys orders are

True pawras, We all know various parameters and can probabaly identify steamers and drifters by spending hours every day and can also make money but is it possible to do this automaticaly using code or other mathematics like data/machine learning etc is what I am trying to figure out.

if the criteria for believing a runner is going to shorten or drift is part of normal available form information, e.g. has had negative results at the distance when others have had positive results, then it can be done automatically.
but you do need a db which holds historical results and todays decs so that you can test your ideas and spit out which of todays runners you should be looking at.

Interesting thread, imo.
I still think that, when dealing with a shi t of thing known laughingly as "Starting Price", your data is more than likely to be corrupt.

It always reminds me of when we used to volunteer to bring up the core samples from the long-hole drillers in the gold mine in Australia.
We used to have to rush to do this during "smoko" which was often less than half an hour.
"Accidents" and mistakes were often made ( trays dropped and the samples just thown back in, any old way).

Geologists back in civilisation would do the analysis.
Because their raw data was corrupt, 'tis no wonder that many  a ace else.false gamble on the Sydney and London stock markets  occurred.

"Rubbish in; rubbish out", I guess.

Btw, just like in the racing game, mining companies used to cheat like fk wherever they could: the name of the game was "kid the investors".

Some mines produced nothing but dirt. None of us cared till the money ran out. Then we went and did it all again some place else.

Anyway, enough of the reality. Good luck to all ye guys who try to formulate and automate your betting predictions!
However, remember this, although 'tis often stated that no bstrd on the btf forum ever loses - "sharper minds" and all that -
the harsh liklihood is that the gambling companies - huge syndicates and bookmakers - are light-years ahead on this kind of thing.

All the best!

ps
btw,
I'm only a traveller off the side of the road, but I think it's "one criterion, two criteria".

ps2,

apologies for the rough spelling.

The point remains valid: check your data.

Today a bookie told me that an expensive horse with Dubai in its name had been punted from 25s to 7/2. It came nowhere.

ps3

All this stuff should probably not be here but in General betting, imo.
It is very interesting.

to do any useful analysis re steamers and drifters I'd you need at least a couple of years of betfair data on hand and have that linked to a form database. that way you can do some data mining on criteria for price changes.

There used to be some guy called something like betfairy fantasist who was always raving about the benefits of betfair sp.
Apparently, it is much better than betting-shop sp ( from the punter's point of view, that is).

Obviously, somebody like myself wouldn't have a clue as to how to predict this. After the event, it almost always becomes " backfitting", imo.

Good luck to those who can predict sp, and, more importantly, profit from their skills/methods/bots, or whatever!

Sean,
Unfortunately,the poster known as BFF was utterly certifiable who lied faster than a horse could gallop and did more Banks than Ronnie Biggs

Jul 10, 2014 -- 11:21AM, pawras wrote:

the key thing is identifying a good criteria(s) as to why they're going shorten or drift.

So this can be done in 2 ways, One is build logic and test it on daily basis and fine tune it till it becomes good enough to be profitable OR get as much data as possible using betfair APIs, timeform APIs or racingpost, build own DB and feed it to ML algos and see if it can find some patterns.

I kind of think using ML will give some patterns that can help to build logic, as combination of both should have accurate prediction.

Jul 10, 2014 -- 5:47AM, Rider wrote:

weight of money as a reason for sps is a bit of an obvious statement, finding the reasons for the money is the objective, eg when a horse is heavily backed and then is seen in the race with an easy lead for the first time that is insider knowledge, it would be nice to think you could model these things but in reality you just dont know as an outsider, things that can be modelled will usually be factored in much earlier than the last 15 minutes

Rider, Do you mean money that comes in last 15 minutes may not have anything to do with form/current conditions and if I am sitting at home I will be at disadvantage as I will be an outsider? 90% of money comes in last 15-20 minutes so I was under the impression and serious money comes at that point and if we are betting on form then we should be profitable in long term, if there are people with insider's knowledge and betting on every race then long term nobody other then them will be profitable, I am thinking 3-5% of races may always have people with insider knowledge.