As some of you may have noticed, Przeczpsan (can't find the right spelling!) has pointed out a problem with our method here which means that it can be exploited by people who shortprice homewins down to a price of 1.00 (absolute certainty).
I was toying around with alternatives I think the basis of the argument is sound, but it falls down because it can be exploited by pricing that would be useless in the real world, i.e. pricing at the extremes. On the other hand, reverting to a higher order method (e.g. using the square of the error as last year) biases it towards the middle of the spread which isnt relevant for fixed odds betting.
I think the answer may be in using a notional staking strategy as this is really the driving force behind what we are doing I suppose seeking to make a profit by betting informed by our pricing.
Simple fixed staking, or even fixed profit staking will suffer from the same problem as the MAE measure there is no incentive to be close to the true price, once you indicate value. There needs therefore to be some sort of value-based (i.e. Kelly) staking involved IMO.
My suggestion is to take the players odds and the bookies odds and to calculate Kelly stakes based on a fixed £100 bank (with stakes confined to be £100 max). The total profit realized in each week would be the alternative to the current error measurement.
I put a spreadsheet together to do the sums and sent it to OddsAgainst yesterday for comment, but I haven't heard back.
Anyone else got any thoughts?
I was toying around with alternatives I think the basis of the argument is sound, but it falls down because it can be exploited by pricing that would be useless in the real world, i.e. pricing at the extremes. On the other hand, reverting to a higher order method (e.g. using the square of the error as last year) biases it towards the middle of the spread which isnt relevant for fixed odds betting.
I think the answer may be in using a notional staking strategy as this is really the driving force behind what we are doing I suppose seeking to make a profit by betting informed by our pricing.
Simple fixed staking, or even fixed profit staking will suffer from the same problem as the MAE measure there is no incentive to be close to the true price, once you indicate value. There needs therefore to be some sort of value-based (i.e. Kelly) staking involved IMO.
My suggestion is to take the players odds and the bookies odds and to calculate Kelly stakes based on a fixed £100 bank (with stakes confined to be £100 max). The total profit realized in each week would be the alternative to the current error measurement.
I put a spreadsheet together to do the sums and sent it to OddsAgainst yesterday for comment, but I haven't heard back.
Anyone else got any thoughts?
