I have spent a fair bit of this year going over the data I have collected over quite a long period of time. In one particluar sport, (a US Sport) Over a sample of 1721 bets that meet the criteria I have a Profit on turnover (POT) of 8.75%. The criteria pushes out an average of 2 bets a day. The average odds for each bet is 2.44 with a max price of 2.99 and a minimum price of 2.0. I have calculated the POT via staking to win £1000. SO a 2.0 shot would be £1000 staked, A 2.5 shot £666.67 etc...
To level stakes it returns 44.57% at ave 2.44 price.
My questions to you guys are...
Using this staking method and assuming a £10,000 bank what % would you risk as a maximum stake on any particluar bet, so in this instance a 2.0 shot?
Would you recommend using a system like Kelly or something similar?
Thanks.
Staking - US bets
-
Nero Tulip
- Posts: 708
- Joined: Wed Apr 15, 2009 5:29 pm
Kelly. Always Kelly. Though, run it in your simulations/data to check that the return does seem optimised, that will give you further confidence that your estimation of fair value is considerably better than the market's.
My advice is don't fanny around with half Kelly, full Kelly is the way to go, but use a 'notional bank' that is a percentage of your total worth, be on the very conservative side and take into account market liquidity, it's not a sprint, you might still question things if/when they go on a dodgy run (I still do even with my best strategies). Your Notional bank does not change in size after winning / losing bets, this is how you avoid the Kelly boom and bust, doing this it makes no sense to go half Kelly etc.
Congrats on an awesome set of results, I'm jealous
!
My advice is don't fanny around with half Kelly, full Kelly is the way to go, but use a 'notional bank' that is a percentage of your total worth, be on the very conservative side and take into account market liquidity, it's not a sprint, you might still question things if/when they go on a dodgy run (I still do even with my best strategies). Your Notional bank does not change in size after winning / losing bets, this is how you avoid the Kelly boom and bust, doing this it makes no sense to go half Kelly etc.
Congrats on an awesome set of results, I'm jealous
!-
Nero Tulip
- Posts: 708
- Joined: Wed Apr 15, 2009 5:29 pm
PS I'd love to know which sport this was!
edit -
I should also say, my recommendation of Kelly assumes you are producing your own fair value figure for each selection... rather than simply using some 'systematic' selection criteria. Kelly will live and die on the quality of your FV, as such it's actually a nice way to measure your success and any improvements you make.
edit -
I should also say, my recommendation of Kelly assumes you are producing your own fair value figure for each selection... rather than simply using some 'systematic' selection criteria. Kelly will live and die on the quality of your FV, as such it's actually a nice way to measure your success and any improvements you make.
Thanks for your reply, The problem I have with Kelly is some times I have 5 bets on at the same time, or at least need placing at the same time and Kelly spits out some scary stakes! In an evening I could have 60-70% of my bank on the 5 bets using full kelly. Not something I can do as its not uncommon to get 5 or more losers in a row, This would really set back the stakes and optimization. The model is obviously not 100% accurate so in instances where it is wrongly out of kilter with the market, kelly will compound the over value the of the model and leave me over exposed.
I am going to run some numbers using % of kelly and see how that goes. Cheers
I am going to run some numbers using % of kelly and see how that goes. Cheers
-
Nero Tulip
- Posts: 708
- Joined: Wed Apr 15, 2009 5:29 pm
Notional bank solves that problem by being small enough to reload. Remember, this is not a bank that you are fearful of losing, its simply your individual bet operating float.. which is why it needs to be small, as you are using it for every bet. Think along lines of 1-5% of your big bank roll, test it and look for risk of ruin, then be a bit more conservative - sleep and a steady p/l are underrated
If you have that much on then your FV is seriously a long way from the market, which is amazing if its right, but leaves me just the other side of sceptical - that's not to be disparaging, sometimes the best results data spits out is the aggressive sort, that rewards you in clarity of direction more than it does in slight inaccuracy, which can always be refined.
edit-
fwiw, I use full kelly on my notional, and it's no bother. Remember by using a % of kelly, all you are actually saying is FV is this % closer to the market than my calculations. In which case, work on improving your estimation of FV.
If you have that much on then your FV is seriously a long way from the market, which is amazing if its right, but leaves me just the other side of sceptical - that's not to be disparaging, sometimes the best results data spits out is the aggressive sort, that rewards you in clarity of direction more than it does in slight inaccuracy, which can always be refined.
edit-
fwiw, I use full kelly on my notional, and it's no bother. Remember by using a % of kelly, all you are actually saying is FV is this % closer to the market than my calculations. In which case, work on improving your estimation of FV.