General Betting
There is currently 1 person viewing this thread.
I have a method that returns 15% net on a sample of 100 actual bets
Backfitting the selection criteria to 1000 historical results returns 14% so I am fairly confident that in current market conditions, I have a winning strategy.
The odds of winning bets range between 1.6 to 2.1 with an average of 1.77
Longest losing run on 100 actual bets is 3 (twice)
The bets are based on a dutching system covering approx 30% of the field.
I am trying to use a Kelly staking calculator
http://www.seriousbet.co.uk/calculators/kelly_staking
.. but I have a problem with the win probability percentage because at the moment I only have an average figure for the actual bets I have placed. I'm a bit concerned that by using my average win percentage going forward I may unearth a glitch because I cannot be certain that there is a good correlation between actual odds and win probability.
So... at the moment I have to enter 65% win probability whether the odds of the bet are 1.5 or 2.1
In future I will be able to get a better idea of win probability for any given odds but at the moment I dont have enough results info.
Any comments or help appreciated. Mainly I would like to know what you would do if you have understood the above.
Backfitting the selection criteria to 1000 historical results returns 14% so I am fairly confident that in current market conditions, I have a winning strategy.
The odds of winning bets range between 1.6 to 2.1 with an average of 1.77
Longest losing run on 100 actual bets is 3 (twice)
The bets are based on a dutching system covering approx 30% of the field.
I am trying to use a Kelly staking calculator
http://www.seriousbet.co.uk/calculators/kelly_staking
.. but I have a problem with the win probability percentage because at the moment I only have an average figure for the actual bets I have placed. I'm a bit concerned that by using my average win percentage going forward I may unearth a glitch because I cannot be certain that there is a good correlation between actual odds and win probability.
So... at the moment I have to enter 65% win probability whether the odds of the bet are 1.5 or 2.1
In future I will be able to get a better idea of win probability for any given odds but at the moment I dont have enough results info.
Any comments or help appreciated. Mainly I would like to know what you would do if you have understood the above.
sort by:
Show
per page
Replies: 20
|
By:
If your edge is 14% after commission then you could use this figure for your kelly staking. A word of warning though: if your edge turns out to be less than 14% then you are likely to crash and burn with full kellly.
|
|
By:
Thanks Bayes
What would you do given that you had the same info? Its a football system so greed is getting the better of me and I am tempted to spin it for the rest of the season. |
|
By:
I would probably assume a conservative edge and stake using some form of kelly, although not full kelly. It depends on the strength of the market and how confident I am that the edge is real and will persist.
|
|
By:
Yeah, I'm going to be very conservative and do half Kelly.
|
|
By:
what about commission? if you are on say 4% then shoudl really take this off your edge and halve it, to get 5%.
Then bet to return 5% of your bank. So if your bank is say £100,000, and the price is 2.0, bet £2,500 If the price is 1.5 then bet £3,333 |
|
By:
FWIW Half-Kelly is not as conservative as it sounds.
|
|
By:
GB thanks - my returns stated are net
|
|
By:
Lori,
How's that? |
|
By:
It just isn't.
If there are even small errors in your estimates, they can snowball pretty rapidly into bigger errors than you expect. I'm forever advocating people use Kelly so it's rare for me to go the other way a touch, but I'd probably start more like a quarter or maybe even less until I was comfortable with it. Another option, given that you've back tested, would be to put half, quarter, tenth into your backtesting and have a look at the swing sizes to see how you feel about them. |
|
By:
or, more importantly use a higher value of kelly (half would be fine) but in conjunction with much lowered edge estimates until you're sure that your edge is uniform across different price bands.
|
|
By:
I also think I may have misunderstood your post.
I took it to mean you thought you were being very conservative using half Kelly. Now I read the context again, if it means being very conservative with edge projections and also using half Kelly, then just ignore me and carry on , as that sounds fine :D |
|
By:
Lori,
Yes, I understand what you say. I have run various worst case scenarios on the parameters and a few percentage points can wipe out all profit. Unfortunately I can't fully test the back data because I dont have actulal odds for all the matches. I do however have a very good correlation between my input data and the actual odds on the bets placed so I am tempted to assume that the same applies to the back data :) Not being ultra conservative by nature I have decided now to go with full Kelly with a stop loss at half bank :) |
|
By:
cheers for comments
|
|
By:
Why do you need a staking plan when you have a good level stake profit ?
|
|
By:
to maximize profit ken.
The good news with Kelly is it isn't sequence dependent and it increases your winnings. Wait, we've done this one before. |
|
By:
Easy to maximise profits at levels. Just increase the level stake.
|
|
By:
http://en.wikipedia.org/wiki/Kelly_criterion
|
|
By:
It's impossible beetle, given him that link many times.
Ken, there's a difference between winning more, and maximizing. This game is about maximizing while you can. Level stakes is a decent approximation of a system like this anyway, but if you don't change as your bankroll changes, you're leaving money on the table. |
|
By:
Just increase the level stake.
That's not level then is it? |
|
By:
As your bank accumulates with a 15% edge increase the 'level' stake. For instance, from a 50 point bank, betting 1 point per bet, when the bank reaches 60 points, increase the 1 point to 1.5 points, then at 72 points up to 2.25 points etc, etc.
|