[Spartak]

There is something bothering me this days, so I decide to ask the sbr community.
I am watching closely a betting model(unfortunately not mine) that spits value bets on most popular eu soccer games. It is giving own % chance for a bet to win and then if there is value compared to the odds suggest a play. So far it is successful one with flat betting for me, but I am curios can I maximize it.
I mean the value in % is different for every game ( if there is ofc) and I believe it will be more wise to set up my bet according the value. in most cases it is with 5%-15% range but in rare cases it is around 20%. For example it gave 4 plays in last Sunday with only one winner but at +700 so it could be stressful one if i dont optimize the bet size.
There is another example for upcoming game. It says the dog has 20% chance to win and another 21% to avoid defeat. The bookie is giving him just 5% to win and 10% to avoid defeat.
What is the proper way to set up the bet size in my case?
10x in advance.

[danshan11]

you might be better off with Kelly or some form of Kelly. you should read into using kelly

[semibluff]

The Kelly Criterion Wikipedia page would be a good place to start. I would recommend fractional Kelly over full Kelly, (with a fraction of your choice), because I don't like going broke. The closer you get to maximising your winnings the more you risk a catastrophic run of results.

If you are going to follow any form of tipping/advice service then track results that you can personally verify before getting financially involved. Many services will falsify records with odds that were never available or slip in past post winners. It's not impossible to identify a +1900 outcome that should be +400, but such betting opportunities are rare. They're invariably in small markets where you will struggle to bet big anyway. Smaller books that deal with such markets don't always have the best record in paying out.

[HeeeHAWWWW]

Many Most services will falsify records with odds that were never available or slip in past post winners.

Fixed that for you :-)

[Poisec]

Fixed that for you :-)

Yep, the few times I subscribed to services, I always went through these guys https://smartbettingclub.com for transparent and verified results. But even then, it was difficult to get profit, because the most successful services will impact the odds quite a lot once they released their tips (most obvious example is Club GOWI, a shame because he is really good).

[Spartak]

Will give a try with kelly but will track for a while before jumping on. I know how it works and could be real trouble if you don’t have a constant edge.

[danshan11]

question really is knowing what a real edge is, most people perceive an edge as my model shows -220 and the line is -160 and to some that means massive value but in reality if that line closes at -160, there really was no value in most cases

[danshan11]

and to be honest on a staking method, I am firmly a believer in fixed profit
simply divide your roll into 100 pieces and place each bet to win 1 unit
-200 bet 2 to win 1
100 bet 1 to win 1
+200 bet .5 to win 1 and so on
its magic, you want to have more money on bets that are more likely to win. its an ugly cousin of kelly but if you backtest it can show an edge over flat and there is no way really to backtest kelly so who knows if it is better than kelly

[tsty]

Lol back test kelly?

What r u talking about

[danshan11]

yeah can you please tell me how someone can back test kelly?

[HeeeHAWWWW]

It's not enormously practical to backtest staking systems in most sports, you don't have the sample size. Better approach for me is a simulation - say 5000 bets repeated a million times.

[danshan11]

you cant simulate your edge, so how would you backtest, and if you use a flat or avg type number you are truly not using kelly based on edge and that defeats the purpose. I dont doubt kelly usually I just doubt the accuracy of peoples perceived edge

[HeeeHAWWWW]

you cant simulate your edge, so how would you backtest,.....

Use a range of edges then, it's not an issue. The point is to select a staking method.

[danshan11]

Use a range of edges then, it's not an issue. The point is to select a staking method.

you can do that but the whole idea of kelly is to use your edge to determine bet size and this is very difficult to simulate. I truly believe this is a big factor in people not having success with kelly. I think kelly is great IF you know your true edge and reality is it is very difficult to know your edge. I will not debate about testing because honestly I am not a tester. My thing is I do not believe you can actually back test your edge. Now if someone has their old lines and the closing line and results they could use that to see how kelly would do.

[Zocki]

Use a range of edges then, it's not an issue. The point is to select a staking method.

What about this

You "know" your edge. So the stake for backtesting each bet is Kelly: edge / odds-1

Should this not leave you just with wins and losses, you can calculate a yield and so on, that enables you to see if you are on to something with your estimates?

edit:
maybe in 2nd step and if you are happy after a large sample size check the average stake, min, max, frequency and more, translate fractions to betsizes and if this is practical and suits you and go on with step 3...

But I think I got your points danshan and HeeeHawww

[TommieGunshot]

you cant simulate your edge, so how would you backtest, and if you use a flat or avg type number you are truly not using kelly based on edge and that defeats the purpose. I dont doubt kelly usually I just doubt the accuracy of peoples perceived edge

Then don't "simulate" your edge. Just use whatever the actual probabilities are. Estimations are ok, especially if using a fraction of Kelly Criterion.

[danshan11]

you cant use actual probabilities to backtest because you dont have them, we dont bet at the closing number, we bet at a number between there. If you use the closing as the bet line it wont matter what kind of staking method you will lose because you will have to pay the vig. Especially if you are running it against thousands of results.

I think we all can agree kelly is the best method IF you know your edge. I dont think backtesting kelly is necessary. the issue is do you have an edge and if so how big is it? I mean before you do anything you need to know if you have an edge. I have argued before that kelly is not appropriate for sports bets because when we bet we use our perceived edge and or the current line, which we know both of these are probably not consistent with the closing line which is the true probability minus juice. this is why I think fixed profit betting is the most efficient and technically is kelly in a sense, you bet whatever it takes to win 1% of your bankroll.
I am not going to fight whether kelly is good or bad, I have my opinion and to be honest it could be 100% wrong but I am good with this discussion. if you believe in kelly and love it, keep on keepin on!

[TommieGunshot]

you cant use actual probabilities to backtest because you dont have them,

I am perfectly free to use the best estimates I have.

And that's exactly what I do

I mean before you do anything you need to know if you have an edge. I have argued before that kelly is not appropriate for sports bets because when we bet we use our perceived edge and or the current line,

This is absolutely correct. However, the closing line is immaterial. The odds the bet was made at was known. The edge (or even just the perceived edge) is known (otherwise the best bet is zero). Put those into the Kelly Criterion formula and then know for sure what the best bet size is to be maximally aggressive. There is also the option to cut it in half (fractional Kelly) to significantly cut down on risk of ruin as well as mitigate any effects of overestimating the edge.

[HeeeHAWWWW]

What about this
You "know" your edge. So the stake for backtesting each bet is Kelly: edge / odds-1

Should this not leave you just with wins and losses, you can calculate a yield and so on, that enables you to see if you are on to something with your estimates?

The problem is that over usual sample sizes, luck plays an enormous role.

Say you have 1000 bets at around evens, with an edge in the 1% to 7% range. At 1/3 Kelly, your median outcome is +63% bankroll growth, with quartiles at +20% and +117%. If you're in the bottom 5%ile, you'll actually lose 20%. Bottom 0.1%ile, lose 60%. At half-Kelly it's an even wider spread (and full-Kelly is insane).

[Alfa1234]

The problem with "knowing" your edge is, you have no idea what your edge is for every bet, you only know your average edge over a large sample. This makes using Kelly very difficult IMHO. That is, if you base your edge on the closing line. If you know the exact accuracy of your model and disregard the closing line altogether, it's a different thing.

I am an advocate of using a "to win" amount of bankroll regardless of edge %. It averages out long term regardless of edge size.

[danshan11]

The problem with "knowing" your edge is, you have no idea what your edge is for every bet, you only know your average edge over a large sample. This makes using Kelly very difficult IMHO. That is, if you base your edge on the closing line. If you know the exact accuracy of your model and disregard the closing line altogether, it's a different thing.

I am an advocate of using a "to win" amount of bankroll regardless of edge %. It averages out long term regardless of edge size.

the fixed profit or to win method either name is very sound in that you are betting more on probable results and less on less likely to win scenarios, this should stabilize your bankroll and extend your playing time with that bankroll long term. If you flat bet say 1 unit and you end up taking a bunch of really long shots because that is where you perceive value, you could see larger swings of your bankroll and that can have a dramatic effect on your bankroll good or bad. Stability is the key here and fixed profit betting helps with that.

[tsty]

lol there is literally a plethora of papers out there dissecting kelly to it's core yet you guys keep posting this garbage

anyone who has read through those papers would realise that full kelly is the only way to go

kelly maximizes ev to it's fullest potential

[Alfa1234]

lol there is literally a plethora of papers out there dissecting kelly to it's core yet you guys keep posting this garbage

anyone who has read through those papers would realise that full kelly is the only way to go

kelly maximizes ev to it's fullest potential

I wish you the best of luck with your full kelly profit maximising, you'll have to reload several times every year despite making a profit overall, especially at your turnover. I don't understand why you keep posting like you are all-knowing, while a lot of what you post is actually pure garbage...No-one who is serious in this business or has any idea of what they are talking about would advise people to bet full Kelly, the volatility would kill you. Full kelly is great in theory but will not work in practise.

[danshan11]

poor tsty, brother please stop, the more you say the worse it looks

[tsty]

I wish you the best of luck with your full kelly profit maximising, you'll have to reload several times every year despite making a profit overall, especially at your turnover. I don't understand why you keep posting like you are all-knowing, while a lot of what you post is actually pure garbage...No-one who is serious in this business or has any idea of what they are talking about would advise people to bet full Kelly, the volatility would kill you. Full kelly is great in theory but will not work in practise.

it works plenty well in practice

the only difficult part is when you need to place many bets in the day where calculations can get weird

but if you use the calculator enough you can generally get a good feel and not be a nit

it's fairly obvious you have no idea what you are talking about when you think "volatility" is an issue

if you bet enough then it makes no difference

if your bankroll doesn't fluctuate 10-30% daily then you are doing something wrong

[danshan11]

can you explain the 10-30% fluctuation please

[Alfa1234]

it works plenty well in practice

the only difficult part is when you need to place many bets in the day where calculations can get weird

but if you use the calculator enough you can generally get a good feel and not be a nit

it's fairly obvious you have no idea what you are talking about when you think "volatility" is an issue

if you bet enough then it makes no difference

if your bankroll doesn't fluctuate 10-30% daily then you are doing something wrong

You think you are doing great because your bankroll fluctuates 30% on any given day? Do you realise you just said a 4-day bad streak will wipe you out? Are you serious??

[Alfa1234]

can you explain the 10-30% fluctuation please

If you have a high perceived edge of 20-30%, even at a high placed odd (say the odd is 5.0 / +400 but your perceived probability of winning the bet is say 50% - giving you a huge edge, full Kelly will make you bet 37.5% of your total roll. This is obviously a terrible idea to anyone with any common sense as losing a few bets in a row at 50% probability will occur on any given day.

[danshan11]

thanks for the reply and of course anything that makes you bet more than a few% on any single game is a horrible idea and there should be no time every where you have a 20-30% edge. I would say in todays very sharp market that a 20-30% perceived edge would be a model problem in most cases. I am sure there are a few occasions this could occur but not a smart play to bet based on that type of edge. Over a huge sample size there would eventually be situations that would put nearly your entire bankroll in play on a very small number of plays. Again enhances my argument for fixed profit betting and as I have read you concur with me on this. But with that said remember tsty is out playing and betting us all and is playing at another level. it would be like me and you arguing a flying man is not real but we only think that because we are human and not from Krypton like tsty!

[semibluff]

If you have a high perceived edge of 20-30%, even at a high placed odd (say the odd is 5.0 / +400 but your perceived probability of winning the bet is say 50% - giving you a huge edge, full Kelly will make you bet 37.5% of your total roll. This is obviously a terrible idea to anyone with any common sense as losing a few bets in a row at 50% probability will occur on any given day.

Anyone who can quickly replace a bankroll decimated by a string of bad results and betting full Kelly isn't really betting full Kelly in the first place.

[Alfa1234]

Anyone who can quickly replace a bankroll decimated by a string of bad results and betting full Kelly isn't really betting full Kelly in the first place.

Very good point. If you artificially lower your bankroll by only allocating a small portion of it to the roll you base the calculation on, you are automatically only betting a certain smaller Kelly percentage.

[tsty]

You think you are doing great because your bankroll fluctuates 30% on any given day? Do you realise you just said a 4-day bad streak will wipe you out? Are you serious??

Do you understand how stupid you sound?

If I lose 30% of my bankroll every day then I end up with 0%?

RoR on kelly is zero

[tsty]

Anyone who can quickly replace a bankroll decimated by a string of bad results and betting full Kelly isn't really betting full Kelly in the first place.

this does not make any sense

[tsty]

thanks for the reply and of course anything that makes you bet more than a few% on any single game is a horrible idea and there should be no time every where you have a 20-30% edge. I would say in todays very sharp market that a 20-30% perceived edge would be a model problem in most cases. I am sure there are a few occasions this could occur but not a smart play to bet based on that type of edge. Over a huge sample size there would eventually be situations that would put nearly your entire bankroll in play on a very small number of plays. Again enhances my argument for fixed profit betting and as I have read you concur with me on this. But with that said remember tsty is out playing and betting us all and is playing at another level. it would be like me and you arguing a flying man is not real but we only think that because we are human and not from Krypton like tsty!

I was not talking about edge

and finding 30% edge isn't that rare

finding a few every year isn't that uncommon
mayweather vs connor was a lot more than that

[Optional]

Anyone who can quickly replace a bankroll decimated by a string of bad results and betting full Kelly isn't really betting full Kelly in the first place.

this does not make any sense

It does, because technically you are supposed to look at your entire liquid assets as the bankroll when calculating a Kelly bet.