I'm backing testing a system and it has thrown up a couple of points of profit which would you regard as the "strongest" or the one which would you be happier following or are they equal.
So i have found 125 instances where if i were to lay at 1.73 for £100 it would produce a total produce after commission of £625 or £5 per instance
Now compare this if i lay at 1.12 i also get £625 or £5 per instance.
I maybe be wrong here but i see that even though both return the same profit that both numbers have their own weaknesses. The problem i see with the laying at 1.73 is that if your strike falls its return per instance can fall very quickly whereas the 1.12 lays return per race will hold up for longer but the numbers could be skewed by a few "false" wins.
What number is strongest?
Yes the 1.12 is much more scalable which is a positive but i would just like people opinions regarding the actual merits of each number.
As i said im worried that the 1.12 results could be inflated as the sample size is not massive.
What's your view Peter?
As i said im worried that the 1.12 results could be inflated as the sample size is not massive.
What's your view Peter?
Yes it's a difficult one to call to be honest.
I'm running a test system at the moment in the background that has run approx 4000 races. Its about 3K up but I'm still not convinced whether it is a long term profitable system as the losing periods can be lengthy. (Approx £800 of that was in one day, so luck??)
I've also run into problems with scale as I'm now finding that sometimes only part of my stake is being matched.
I would have thought that there would be more liquidity at the higher odds?
Regards
Peter
PS If you are paying PC dont forget to take that into account
I'm running a test system at the moment in the background that has run approx 4000 races. Its about 3K up but I'm still not convinced whether it is a long term profitable system as the losing periods can be lengthy. (Approx £800 of that was in one day, so luck??)
I've also run into problems with scale as I'm now finding that sometimes only part of my stake is being matched.
I would have thought that there would be more liquidity at the higher odds?
Regards
Peter
PS If you are paying PC dont forget to take that into account
I'm not paying the PC at the moment due to some substantial losses i accrued on my account with some bookmaker arbs. Im no expert at statistical analysis i know the greater the sample the increased confidence in the results but im wrestling with the sample size and at what amount do i have enough confidence to starting placing real money and then when to step it up.
I think that statistically speaking, the former version where you are laying at 1.73 is stronger, however I think you don't really have enough data to come to a proper conclusion about whether or not you have an advantage. You need to do a monte carlo hypothesis test (in excel) and then see whether or not your results are "statistically significant" and then take it from there. Maybe I should chuck up a spreadsheet for people to use hmmm...
Hi Buyshirts,
A few thoughts.
How are you back testing the system or getting results? Have you wrote a bot or excel sheet to interact with the market? Or are you simply using Timeform or Betfair free data?
Firstly, the problem with Timeform is the accuracy of the numbers reported is terrible, on many occasions the actual odds matched on betfair is different and the BSP is not actual price at off but a calculation based on orders and overround of selections.
I believe the official Timeform line regarding inplay odds is that £100 must be matched to register on their site. Your £8 - 60 lay for £5 profit could easily fall into this £100. Watch a few races and then compare the data on their website to your observations.
Secondly, the betfair inplay market is extremely difficult to trade, even with a bot/excel, price will often gap past your order or be taken when the true probability of the event is far less if laying. So you may find your 1.12 is only taken when a horse has 95% chance of winning (1.05) not 89% (1.12) and if a horse falls at the last when leading by 3 lengths (a 1.12 scenario, perhaps) the price will go pretty much straight to 1000. For this reason alone, 1.73 should be an easier number to work with in real life as it is much more likely to get matched regularly.
Another thing, I am not sure how it is possible that in this sample you can make profit? Simply picking a number to lay at, will, over 1000s of bets, tend towards the expectancy of your money back less 5% betfair commission. Ask your self if there is a genuine edge going forward or simply found a data set with an edge? Why is the market wrong at these values? What do you know that the market doesn't? Have you manipulated the data in a way that would not be possible if you didnt already know the result? Sound stupid thing to ask but it is easily done by over fitting past data to guarantee profit.
well done if you have a genuine edge, keep it guarded, if you are simply playing with numbers though, be careful before putting money in.
Hopefully, this is not too negative, experience has taught me past data and inplay markets are a receipe for disaster. My ultimate advice would be find an edge immediately before the off and ignore inplay, huge amounts of money are matched at t=0 and you are pretty much guaranteed to be matched. This means any research you carry out is not in vain.
Good luck
A few thoughts.
How are you back testing the system or getting results? Have you wrote a bot or excel sheet to interact with the market? Or are you simply using Timeform or Betfair free data?
Firstly, the problem with Timeform is the accuracy of the numbers reported is terrible, on many occasions the actual odds matched on betfair is different and the BSP is not actual price at off but a calculation based on orders and overround of selections.
I believe the official Timeform line regarding inplay odds is that £100 must be matched to register on their site. Your £8 - 60 lay for £5 profit could easily fall into this £100. Watch a few races and then compare the data on their website to your observations.
Secondly, the betfair inplay market is extremely difficult to trade, even with a bot/excel, price will often gap past your order or be taken when the true probability of the event is far less if laying. So you may find your 1.12 is only taken when a horse has 95% chance of winning (1.05) not 89% (1.12) and if a horse falls at the last when leading by 3 lengths (a 1.12 scenario, perhaps) the price will go pretty much straight to 1000. For this reason alone, 1.73 should be an easier number to work with in real life as it is much more likely to get matched regularly.
Another thing, I am not sure how it is possible that in this sample you can make profit? Simply picking a number to lay at, will, over 1000s of bets, tend towards the expectancy of your money back less 5% betfair commission. Ask your self if there is a genuine edge going forward or simply found a data set with an edge? Why is the market wrong at these values? What do you know that the market doesn't? Have you manipulated the data in a way that would not be possible if you didnt already know the result? Sound stupid thing to ask but it is easily done by over fitting past data to guarantee profit.
well done if you have a genuine edge, keep it guarded, if you are simply playing with numbers though, be careful before putting money in.
Hopefully, this is not too negative, experience has taught me past data and inplay markets are a receipe for disaster. My ultimate advice would be find an edge immediately before the off and ignore inplay, huge amounts of money are matched at t=0 and you are pretty much guaranteed to be matched. This means any research you carry out is not in vain.
Good luck
buyshirts wrote:As i stated before im light on statistically knowledge, So if you could post a spreadsheet that would be a step in the right direction.
Cheers
Alright sure, I'll do that I just need to knock it up for you. Basically, what the spreadsheet will do is assume that you don't have an advantage on your bets, then calculate the chance that you made the same or more money on your bets. In other words, the sheet will assume the odds you're giving represent the true odds, and it'll then work out the chance you'd make more than you have betting on the true odds. If this figure is low (like lets say under 1%) then it makes sense that there is a very good chance you have an advantage. I have some free time this evening, I'll sort it out and post it tomorrow.
-
JollyGreen
- Posts: 2047
- Joined: Sat Mar 21, 2009 10:06 am
I assume this is in running? Can I ask a couple of questions?
1. Have you actually tested this live or is it from data back testing - this may have already been asked?
2. Are you aware of the result bias that occurs on some tracks?
JG
1. Have you actually tested this live or is it from data back testing - this may have already been asked?
2. Are you aware of the result bias that occurs on some tracks?
JG
Yes it is in running
It is back testing, i am using Timeform's in running price (I understand it's shortcomings as highlighted by BengalV earler)
I have never really looked into system's/angles in such a way where i have had to collect and collate a large amount of data and then analysis it in fine detail. I assume as like most people on here and Betfair we have to start off from a position of a lot of idea's how things work and all fit together but not a lot of the tools needed analysis and dissect the available information.
As you can see by my previous post i have cobbled together a set of results thrown up by certain criteria and have come across (what i assume is a problem for most people) a problem. Firstly what amount of "instances " are enough to validate the results and how best interpret them.
As regards results bias i have started from the position of no track bias but i assume from your question you are aware of some
It is back testing, i am using Timeform's in running price (I understand it's shortcomings as highlighted by BengalV earler)
I have never really looked into system's/angles in such a way where i have had to collect and collate a large amount of data and then analysis it in fine detail. I assume as like most people on here and Betfair we have to start off from a position of a lot of idea's how things work and all fit together but not a lot of the tools needed analysis and dissect the available information.
As you can see by my previous post i have cobbled together a set of results thrown up by certain criteria and have come across (what i assume is a problem for most people) a problem. Firstly what amount of "instances " are enough to validate the results and how best interpret them.
As regards results bias i have started from the position of no track bias but i assume from your question you are aware of some
Not everything that may work in theory is executable in practice and I'm just talking about pre-race. The in-play is even trickier. If you're looking at timeform or betfair data on a spreadsheet for historic event and giving you a positive return doesn't mean that you will be able to replicate this in practice.
Option 1) Laying in person each race - this is where courtsiders and in-play betting exchange have an upper hand
2)spreadsheet or BA automation triggers - you will find that not all bets will be matched because of delaying in execuation and in-play delay. So those that do get matched will reduce your profitability and even cause you an overall loss.
3)Lay the field & keep in-play - again not all will be taken and therefore overall loss
Efficient execution will the key.
Option 1) Laying in person each race - this is where courtsiders and in-play betting exchange have an upper hand
2)spreadsheet or BA automation triggers - you will find that not all bets will be matched because of delaying in execuation and in-play delay. So those that do get matched will reduce your profitability and even cause you an overall loss.
3)Lay the field & keep in-play - again not all will be taken and therefore overall loss
Efficient execution will the key.
You may want to play around with the spreadsheet I attached in this post
viewtopic.php?f=2&t=6017
Regards,
Switesh
viewtopic.php?f=2&t=6017
Regards,
Switesh