General Betting
Welcome to Live View – Take the tour to learn more
Start Tour
There is currently 1 person viewing this thread.
30 Nov 10
23:41
Joined:
Date Joined:
24 Apr 08
I have been pricing up my own odds for a while now and have been using kelly staking on everything I've priced up, even when the edge is minimal. Recently though results have been less good particularly on those where the edge is small.
I've looked at my results and it seems I would get a lot better results if I had a minimum edge threshold. This would see me still making a lot of bets but only about a quarter of what I am now. I thought I'd be subject to greater variance making less bets but the results don't seem to indicate this.
My question is, mathematically what is the best way to decide on the threshold? Should it be a minimum kelly bet e.g. anything greater than 10% of bank, a minimum edge in terms of odds e.g. where my odds are 10% better or a 10% minimum edge in terms of percentage probability of winning e.g. team 60% likely to win and 50 probability offered?
Any ideas appreciated!
Thanks
I've looked at my results and it seems I would get a lot better results if I had a minimum edge threshold. This would see me still making a lot of bets but only about a quarter of what I am now. I thought I'd be subject to greater variance making less bets but the results don't seem to indicate this.
My question is, mathematically what is the best way to decide on the threshold? Should it be a minimum kelly bet e.g. anything greater than 10% of bank, a minimum edge in terms of odds e.g. where my odds are 10% better or a 10% minimum edge in terms of percentage probability of winning e.g. team 60% likely to win and 50 probability offered?
Any ideas appreciated!
Thanks
Show More
At a glance it would suggest you're accuracy isn't as good as you think it is.
Maybe, we'll see I guess.
I think the kelly formula has already taken into account both odds and percentage probability so a minimum kelly stake should cover everything thinking about it. Stupid question on reflection.
I think the kelly formula has already taken into account both odds and percentage probability so a minimum kelly stake should cover everything thinking about it. Stupid question on reflection.
No not really because I split into many banks and do the full kelly on these.
I have been pricing up my own odds
No need to do that, the bookies will do it for you.
No need to do that, the bookies will do it for you.
I have a set maximum percentage per bet. Which I then apply kelly to. The issue with kelly is that small changes in estimation result in large swings in the corresponding stakes.
That is why you need to apply error bounds on your estimation and calculate the resultant stakes for estimate, estimate+error and estimate-error. This should tell you whether within your system error the bet is still a sensible one.
That is why you need to apply error bounds on your estimation and calculate the resultant stakes for estimate, estimate+error and estimate-error. This should tell you whether within your system error the bet is still a sensible one.
Thanks Zola, makes sense. I already had a max which accounted for aby rare mistakes but I think a minimum stake was require dto make sure I was getting value.
Looks good against past results so hopefully should improve profits in the future.
Looks good against past results so hopefully should improve profits in the future.
Seriously, I know Kelly is theoretically sound, but who honestly has a dedicated betting bankroll?
To phrase the question another way. Who would honestly stop betting if what they deem to be their betting bankroll is wiped out?
To phrase the question another way. Who would honestly stop betting if what they deem to be their betting bankroll is wiped out?
True RG, didn't quite think about it properly. However, to rephrase again, and I see you have made pretty much the same point. There is surely a stage for punters at which the bankroll has diminished to the extent that the punter is not satisfied by the suggested stakes and the majority in this position will increase their betting bankroll from other funds.
Am I right in saying Kelly assumes you know your edge? Haven't read up on it for a while as decided it wasn't for me. Always dangerous as, generally, the bigger your edge seems the more likely it is to be incorrect and I'm not sure that using 1/2 or 1/4 Kelly solves that particular problem.
Am I right in saying Kelly assumes you know your edge? Haven't read up on it for a while as decided it wasn't for me. Always dangerous as, generally, the bigger your edge seems the more likely it is to be incorrect and I'm not sure that using 1/2 or 1/4 Kelly solves that particular problem.
racingguru
Date Joined: 06 Jan 01
Add contact | Send message
When: 02 Dec 10 17:59
Joined:
Date Joined: 06 Jan 01
| Topic/replies: 270 | Blogger: racingguru's blog
Further to that the problem that most people make with Kelly is that they don't have the bank size to bet to a level they feel is worthwhile ie: Kelly tells the bet to be much smaller than they would usually do so my advice to anyone thinking of starting to use Kelly is that wait till you have a bankroll 100x your normal bet size before starting.
I see people are agreeing, I can't help but disagree. What you say is quite wrong actually, I found that Kelly advises too big a bankroll to risk in some circumstances, 10-20% of your current bankroll.I don't think professional gamblers can afford to risk such a large percentage of their bank.
In my case I am not comfortable risking more than 1% of my bank at a given time, so I have to pass on Kelly.
Not to mention, what happens if you use Kelly and you hit a losing streak of 4 bets, which is quite probable?People say "oh I'm only betting a percentage of my CURRENT bank so I can't go bankrupt" Let's assume the bets are "good value" and Kelly advises 20% of bank on each. Bank of 1000.
First bet lost800
Second bet lost 640
Third bet lost 512
Fourth bet 409
Good luck reviving your gutted bank then.
Date Joined: 06 Jan 01
Add contact | Send message
When: 02 Dec 10 17:59
Joined:
Date Joined: 06 Jan 01
| Topic/replies: 270 | Blogger: racingguru's blog
Further to that the problem that most people make with Kelly is that they don't have the bank size to bet to a level they feel is worthwhile ie: Kelly tells the bet to be much smaller than they would usually do so my advice to anyone thinking of starting to use Kelly is that wait till you have a bankroll 100x your normal bet size before starting.
I see people are agreeing, I can't help but disagree. What you say is quite wrong actually, I found that Kelly advises too big a bankroll to risk in some circumstances, 10-20% of your current bankroll.I don't think professional gamblers can afford to risk such a large percentage of their bank.
In my case I am not comfortable risking more than 1% of my bank at a given time, so I have to pass on Kelly.
Not to mention, what happens if you use Kelly and you hit a losing streak of 4 bets, which is quite probable?People say "oh I'm only betting a percentage of my CURRENT bank so I can't go bankrupt" Let's assume the bets are "good value" and Kelly advises 20% of bank on each. Bank of 1000.
First bet lost800
Second bet lost 640
Third bet lost 512
Fourth bet 409
Good luck reviving your gutted bank then.
bib - Have you actually attempted to use KELLY CALCULATOR for bets? I suspect not.
Can Befair just pretend this thread doesn't exist before the stupidity contained within it causes the universe to be sucked in to a tiny ball?
my advice is to use triple kelly, basically you just work out the stake then triple it, thereby giving you 3 times as much profit as you would get under normal kelly
I think the important thing about kelly is that you are staking higher on the picks where you have found most value and have the biggest edge. Some people do this naturally but a consistent approach backed up by stats is the ideal.
I have found that having your bank several times over and performing kelly to just one of these solves any problem of overstaking but still maximises profits.
I have found that having your bank several times over and performing kelly to just one of these solves any problem of overstaking but still maximises profits.
racingguru I am sorry I missed the previous post where you said you were using fractional Kelly. So what is the formula for Kelly/4 for example? Just use a quarter of what full Kelly suggests?
Well I said I'm not using Kelly because it advises too high a percentage to risk, I haven't even tried it, sorry for my ignorance.
So I went and did my homework on Kelly, I'm a math illiterate but I can see where it's going, it makes a fair point in that - where you think you have more value, the stake should increase accordingly. Bottom line, I agree that decimal Kelly can be useful for bettors that can accurately estimate the chances of a given outcome occurring.
I generally trade in play and there are so many variables during a (tennis) match that it's almost impossible to use Kelly, for instance I make some bets with view of trading out a couple of minutes later, or at pre defined odds. Then something can happen during the match and you have to scratch the trade or go all red, as I said too many variables.
That's the reason why I generally use low stakes (1% of bank max) even in cases where I think I have a solid edge.
Sorry for my bad English.
I generally trade in play and there are so many variables during a (tennis) match that it's almost impossible to use Kelly, for instance I make some bets with view of trading out a couple of minutes later, or at pre defined odds. Then something can happen during the match and you have to scratch the trade or go all red, as I said too many variables.
That's the reason why I generally use low stakes (1% of bank max) even in cases where I think I have a solid edge.
Sorry for my bad English.
My ignorance may shine through now, but maybe you can convince me otherwise, I do not use Kelly for the main reason: With sports betting how can anyone know what their true edge is? It isn't like a roll of a dice, the exact probability cannot be ascertained! You can look at your previous results and work out your edge over the bookie BUT this is the past and cannot be compared to the upcoming event. I suspect most Kelly advocates will look at their previous results to try and determine what their edge was and apply to future (which is what I disagree with)? Because of the inexact science of betting (btw - I am arguing all this from a position taker's perspective), people will then decide to use a fractional Kelly because they can't work out their edges correctly! Oh and while we are here we notice, we seem to be overstaking on horses with odds under 4 or the handicap chases so we will adjust again for these! All these changes end up transforming the Kelly into a non-Kelly, beast of a complicated staking mechanism.
My ignorance may shine through now, but maybe somebody can convince me otherwise,
It's true that in many cases you cannot ascertain the exact probability of an outcome but you do not need to be dead accurate. If your estimations are correct within a margin of 2-5 percent give or take, I think Kelly is still useful.
For example, I'm sure that if you know a little about soccer you can give a ballpark figure as to what Real Madrid's chances or winning tomorrow are. You could say" between 50% and 80%" but surely if you look at the relevant stats and analyze the match, you can come up with a more accurate estimation.
Then make the same estimation for 100 matches let's say, and when you draw the line you can determine whether your estimations are correct, and when they are off, then by how much.
That's the way I see it.
For example, I'm sure that if you know a little about soccer you can give a ballpark figure as to what Real Madrid's chances or winning tomorrow are. You could say" between 50% and 80%" but surely if you look at the relevant stats and analyze the match, you can come up with a more accurate estimation.
Then make the same estimation for 100 matches let's say, and when you draw the line you can determine whether your estimations are correct, and when they are off, then by how much.
That's the way I see it.
I suspect most Kelly advocates will look at their previous results to try and determine what their edge was and apply to future (which is what I disagree with)?
So... are you really saying you use no past knowledge of yopur previous performance in making future bets?
So... are you really saying you use no past knowledge of yopur previous performance in making future bets?
Because i very much doubt that. Unless of course you have alzheimers.
Getafix Joined: 13 Jul 10
Replies: 26 03 Dec 10 21:21
My ignorance may shine through now, but maybe you can convince me otherwise, I do not use Kelly for the main reason: With sports betting how can anyone know what their true edge is? It isn't like a roll of a dice, the exact probability cannot be ascertained! You can look at your previous results and work out your edge over the bookie BUT this is the past and cannot be compared to the upcoming event. I suspect most Kelly advocates will look at their previous results to try and determine what their edge was and apply to future (which is what I disagree with)? Because of the inexact science of betting (btw - I am arguing all this from a position taker's perspective), people will then decide to use a fractional Kelly because they can't work out their edges correctly! Oh and while we are here we notice, we seem to be overstaking on horses with odds under 4 or the handicap chases so we will adjust again for these! All these changes end up transforming the Kelly into a non-Kelly, beast of a complicated staking mechanism
===========================================================================================
Replies: 26 03 Dec 10 21:21
My ignorance may shine through now, but maybe you can convince me otherwise, I do not use Kelly for the main reason: With sports betting how can anyone know what their true edge is? It isn't like a roll of a dice, the exact probability cannot be ascertained! You can look at your previous results and work out your edge over the bookie BUT this is the past and cannot be compared to the upcoming event. I suspect most Kelly advocates will look at their previous results to try and determine what their edge was and apply to future (which is what I disagree with)? Because of the inexact science of betting (btw - I am arguing all this from a position taker's perspective), people will then decide to use a fractional Kelly because they can't work out their edges correctly! Oh and while we are here we notice, we seem to be overstaking on horses with odds under 4 or the handicap chases so we will adjust again for these! All these changes end up transforming the Kelly into a non-Kelly, beast of a complicated staking mechanism
===========================================================================================
It isn't like a roll of a dice, the exact probability cannot be ascertained!
======================================================================================
A roll of a dice is gambling WITHOUT any "edge" !
======================================================================================
A roll of a dice is gambling WITHOUT any "edge" !
I agree that it is extremely "hard" to "know" what your "true edge is"................
However .......betting based on -
1.you believe you have an edge .......
2. based on "value" on the odds .....
is surely far better than NOT! eg LEVEL- STAKING type bets...............
However .......betting based on -
1.you believe you have an edge .......
2. based on "value" on the odds .....
is surely far better than NOT! eg LEVEL- STAKING type bets...............
Thanks for responses, I was playing devil's advocate a little.
I don't have alzheimers lol, and yes I will use previous results to try and determine a stake size I am comfortable going forward in order to protect my bank. I am just not convinced by the use of Kelly in staking over say fixed/perc stakes/liab. In this thread and many others I have read, nobody ever uses full Kelly (for sports betting - position taking). The reason being their real edge cannot be determined. So what happens is people split up their strategies in order to apply Kelly i.e., use a x fractional kelly for say handicap chases but if odds only over 4, when one starts breaking down into sub categories this is actually proof of the flaws of kelly i.e., overstaking/understaking when one works out their edge wrong.
Just to clarify a few points, I should have used a better example than rolling dice, I was thinking more from the perspective of a casino working out what odds they should offer - given that they would always offer a book over 100%.
I.e., this is a situation where I think Kelly would be beneficial i.e., exact odds (and therefore exact edge) known before the event.
I am not dismissing Kelly, just airing my views, in fact the reason I posted on this thread is because I would like to be proved wrong with my views on this.
I would ask if any of you do analyse yearly what your pnl would have been had you staked using say percentage liability staking etc instead of Kelly? Though I suspect I know what the answer would be because I don't do post analysis. In fact I should! A job for next week.
I don't have alzheimers lol, and yes I will use previous results to try and determine a stake size I am comfortable going forward in order to protect my bank. I am just not convinced by the use of Kelly in staking over say fixed/perc stakes/liab. In this thread and many others I have read, nobody ever uses full Kelly (for sports betting - position taking). The reason being their real edge cannot be determined. So what happens is people split up their strategies in order to apply Kelly i.e., use a x fractional kelly for say handicap chases but if odds only over 4, when one starts breaking down into sub categories this is actually proof of the flaws of kelly i.e., overstaking/understaking when one works out their edge wrong.
Just to clarify a few points, I should have used a better example than rolling dice, I was thinking more from the perspective of a casino working out what odds they should offer - given that they would always offer a book over 100%.
I.e., this is a situation where I think Kelly would be beneficial i.e., exact odds (and therefore exact edge) known before the event.
I am not dismissing Kelly, just airing my views, in fact the reason I posted on this thread is because I would like to be proved wrong with my views on this.
I would ask if any of you do analyse yearly what your pnl would have been had you staked using say percentage liability staking etc instead of Kelly? Though I suspect I know what the answer would be because I don't do post analysis. In fact I should! A job for next week.
even my 2nd example of casino is crap! I'm sure you can understand what I am trying to get at?
What do Kelly advocates do when the price drifts? For example you place £100 @ 10 the price drifts to 20, do you top up the bet or stick to the original stake?
Level stakes beats most staking plans but if you can accurately price up events kelly will optimise your returns. Most people who benefit from this will be those who produce statistical models where little or no human input has been involved on deciding the kelly stake.
This is because inevitably we all think that Barcelona are 1.01 against some cr4p and stake nearly full bank and they draw/lose. 2 important lessons here:
1) Having several banks or using fractional kelly is a MUST is sports betting as you say futur results odds can never be 100% accurate, even when helped by past data.
2) If you have the same opinion as almost everyone else then the odds are more than likely going to be too low on your selection and long term, a lay of that selection would be more benficial.
Good luck!
This is because inevitably we all think that Barcelona are 1.01 against some cr4p and stake nearly full bank and they draw/lose. 2 important lessons here:
1) Having several banks or using fractional kelly is a MUST is sports betting as you say futur results odds can never be 100% accurate, even when helped by past data.
2) If you have the same opinion as almost everyone else then the odds are more than likely going to be too low on your selection and long term, a lay of that selection would be more benficial.
Good luck!
What do Kelly advocates do when the price drifts?
Kelly determines the optimal proportion to stake given an independent estimate and an instantaneous price.
If you take the current price, you apply the current kelly stake. If the price subsequently changes (and in most cases it will) this should have no bearing on your independent estimation. The question now is, is the new price still value given your unchanging independent estimation and do you want to place another bet
It is simply a device to determine whether a bet at a particular instant in time is value - and if so how much to stake.
This is one of the reasons why it beats level stakes. It is a sensitive measure and should only be used by those who understand it and their own system and their flaws.
Kelly is an incredibly powerful weapon. Just don't point it at your own face.
Kelly determines the optimal proportion to stake given an independent estimate and an instantaneous price.
If you take the current price, you apply the current kelly stake. If the price subsequently changes (and in most cases it will) this should have no bearing on your independent estimation. The question now is, is the new price still value given your unchanging independent estimation and do you want to place another bet
It is simply a device to determine whether a bet at a particular instant in time is value - and if so how much to stake.
This is one of the reasons why it beats level stakes. It is a sensitive measure and should only be used by those who understand it and their own system and their flaws.
Kelly is an incredibly powerful weapon. Just don't point it at your own face.
Thank you all for your thoughts/help on this and apologies for hijacking the thread.
I have finally got round to doing some testing on my last 2 years bets. The results are looking extremely promising but I am wary of comparing apples and pairs......
I thought due to the nature of kelly, it will be very difficult to compare to fixed stakes because of the very nature of each plan. I think the only way each can be compared is by getting the ROI for each staking plan. To be clear, as this often causes debate, my interpretation of ROI is as follows:
ROI: total profit and loss / total liability
For my current dataset, using kelly improves the ROI by a whopping 60% in comparison to fixed stakes! An astonishing gain which is why I have to question this!? Would anyone agree that these staking plans can be compared in this way? Any feedback much appreciated.
I have finally got round to doing some testing on my last 2 years bets. The results are looking extremely promising but I am wary of comparing apples and pairs......
I thought due to the nature of kelly, it will be very difficult to compare to fixed stakes because of the very nature of each plan. I think the only way each can be compared is by getting the ROI for each staking plan. To be clear, as this often causes debate, my interpretation of ROI is as follows:
ROI: total profit and loss / total liability
For my current dataset, using kelly improves the ROI by a whopping 60% in comparison to fixed stakes! An astonishing gain which is why I have to question this!? Would anyone agree that these staking plans can be compared in this way? Any feedback much appreciated.
oh I almost forgot (in true Columbo style)...
for each bet used in the ROI calculations for Kelly, I have not increased my bank with wins and losses. I set the staking to always think the bank is 1 (number irrelevant, the important point being this does not increase/decrease with wins and losses).
for each bet used in the ROI calculations for Kelly, I have not increased my bank with wins and losses. I set the staking to always think the bank is 1 (number irrelevant, the important point being this does not increase/decrease with wins and losses).
Anyone got any book recommendations for Kelly? Thinking of getting myself this for christmas
http://www.amazon.com/gp/product/0809046377/ref=cm_rdp_product/188-9474167-1155851
Poundstones other works look interesting to, anyone read him?
http://www.amazon.com/gp/product/0809046377/ref=cm_rdp_product/188-9474167-1155851
Poundstones other works look interesting to, anyone read him?
Hi Brendanuk1
Yes, have read Fortune's Formula - interesting book (in places) but could have done with a decent sub-editing job because the author does wander off the point many times which makes it a bit confusing. As I have said before, there is a really good explanation of Kelly in (of all things) the Daily Mirror book Better Betting. It also has very good chapters on in-running betting.
Yes, have read Fortune's Formula - interesting book (in places) but could have done with a decent sub-editing job because the author does wander off the point many times which makes it a bit confusing. As I have said before, there is a really good explanation of Kelly in (of all things) the Daily Mirror book Better Betting. It also has very good chapters on in-running betting.
this one?
http://www.amazon.co.uk/Better-Betting-Mirror/dp/1907324097
Is it all horsey horsey stuff? i only do football. Happy to find good info anywhere though
http://www.amazon.co.uk/Better-Betting-Mirror/dp/1907324097
Is it all horsey horsey stuff? i only do football. Happy to find good info anywhere though
Brendan, I would recommend "Fixed Odds Sports Betting: Statistical Forecasting and Risk Management - Joseph Buchdahl",
http://www.amazon.co.uk/Fixed-Odds-Sports-Betting-Statistical/dp/1843440199
this contains many staking plans (inc Kelly - though I skimmed that section last time) which he tests using monte carlo for risk etc. From memory he gives his examples using football, the other chapters well worth the read too regarding overround etc. One of my best books, up there with Peter May's work imo.
http://www.amazon.co.uk/Fixed-Odds-Sports-Betting-Statistical/dp/1843440199
this contains many staking plans (inc Kelly - though I skimmed that section last time) which he tests using monte carlo for risk etc. From memory he gives his examples using football, the other chapters well worth the read too regarding overround etc. One of my best books, up there with Peter May's work imo.
Joe is the football data guy isnt he? will have a look, cheers
Post Your Reply
Please login to post a reply.
Wonder
Instance ID: 13539