Margin of error in prices

10% is a good cut off figure to use imo, though a bit frustrating when there are a lot 0f 7 or 9% possibles that win when you leave them.

Using this calculator you can still bbet on the lower margins but it lowers the stake

http://www.seriousbet.co.uk/calculators/kelly_staking

Thanks mate, I thought 10% would be pretty safe tbh... guess I'll start with that and work my way down if I can or need to! Cheers

i suggest that for events that you are most experienced in pricing and feel you have (within reason) as much information as the people making the market you could confidently bet to 4% margin (ie you think it has a 54% chance of occuring and can bet with no commission at even money). the less you know about the event relative to the market the more chance that you are wrong. perhaps in events you have less experience with 10% is more sensible but there comes a point that if you don't know much about something and are wildly different to a liquid market, that you have to accept that your estimates are very unlikely to be correct.
what sports and markets are you applying this to?

Yes there is a danger that if you set your margin too high, you'll only get on when your prices are out.

I produce ratings for football, and due to improving excel skills I've given them a tweak and an overhaul - i'm about to test them against Football Data's stats over a few season to determine the optimum margin for them, though i'll need to automate the process first.

I don't bet unless the price is at least 20% better, that is for example 2.20 against my price of 2.0. In theory that givers me a profit margin of 7% after commission (hopefully !)

Thanks for your input guys I thought I might miss the boat at 10% tbh. I think you are right that 4-5% percent sounds more realistic. I'm looking at doing this on tennis but also footy when I can get my ratings sorted. I actually feel the prices are less accurate in the tennis markets so will be testing this first.

Kennilworth if you bet when prices are 20% better I think this is approx the same as being 4% better probability e.g. 54% proability bet at 50% is 1.85 bet at 2.00 about 20% increase in odds - so looks like the experts agree!

Thanks for your input guys I thought I might miss the boat at 10% tbh. I think you are right that 4-5% percent sounds more realistic. I'm looking at doing this on tennis but also footy when I can get my ratings sorted. I actually feel the prices are less accurate in the tennis markets so will be testing this first.

Kennilworth if you bet when prices are 20% better I think this is approx the same as being 4% better probability e.g. 54% proability bet at 50% is 1.85 bet at 2.00 about 20% increase in odds - so looks like the experts agree!

scrooge, you flatter me, but I ain't no expert !!

Haha, you seem to have thought about and experienced a fair bit on this from what you post to me!

Although not sure what I said above is correct because if you look for 20% better in odds at 1.04 you would bet at 1.05 but this is only 1% difference in probability. So I am now thinking that the it's the percentage increase in probability you need to measure against for your margin and not the percentage increase in odds because this isn't consistent against probability throughout the scale off odds. Would you guys agree?

scrooge, I never go down anywhere near the likes of 1.05 chances etc. Anyone who can tell the difference between 1.04 and 1.05 shots is a lot smarter than I am.

scrooge - this % value thing is a bit of a nonsense in terms of being theoretically optimal for the exact reason you have highlighted; at extreme prices it does not work.
is a 100/1 shot that should be 8/1 better or worse than a 6/4 shot that should be evs? both apparently offer10% value.

imagine that on 100 unrelated occasions you have £1 at 100/1 on something with an 11% chance of occurring. you will only come out behind over this sample if you back no winners. the chance of this happening is 0.89^100 = 0.0009%. if you back one winner you roughly break even, this will happen 0.01% of times. so you are 99.99% to be in profit. the expected profit on this set of bets is a shade over £1000.

imagine that on 100 unrelated occasions you have £1 at 6/4 on something with a 50% chance of occurring. you will come out behind over this sample if you back anywhere between 0 and 39 winners. the chance of this happening is 1.7% and the chance of you breaking even (exactly 40 winners) is 1.1% so you are 97.2% to be in profit. the expected profit on this set of bets is £25.

on both fronts, the 100/1 that should be 8/1 appears better, and, if you were offered an unlimited number of these as opposed to an unlimited number of 6/4s that should be evens, you would undoubtedly opt for the former.

in practice there are a couple of things to consider.
firstly you are unlikely to find 100 of either of these, particularly the 100s should be 8s. let's say you find 10.
in the 100s should be 8s case there is a 31% chance of you losing all 10 bets.
in the 6/4 should be evens case there is a 0.1% chance of you losing all 10 bets.
it is obvious from this that you can afford to risk a bigger proportion of your bankroll on the 6/4 chances than the 100/1 chances.

in my experience 6/4 chances that i think should be evens come along fairly regularly, perhaps once in every five limited overs cricket matches for example. Horse racing aside, 100/1 shots that should be 8/1 might come along once a month, and are invariably specials with bookmakers who will not allow you to stake as you wish.

to summarise, using this definition of 'value' the bet is worth more to you the further the price is from evens (works for odds on as well as odds against chances). however, at these extremes mistakes and therefore opportunities are less likely.

personally, and this is in no way adhering to kelly or any other strategy, if i was a) certain of being paid out and b) forced to run the bet to expiry with no option to trade out of the position, i would stake about the same on a 100/1 shot that should be 8/1 as on a 6/4 shot that should be evens. if i was certain that my percentages were correct i think between 2 and 3% of my bank.

is a 100/1 shot that should be 8/1 better or worse than a 6/4 shot that should be evs?

You can bet 10.22% of your bankroll on the 100/1 shot for an expected growth of 18.86% and an expected profit of 104.47%

You can have 16.67% of your bank on the 6/4 shot but your expected growth is only 2.062% and your expected profit is 4.17%

Makes the 100/1 shot clearly better.

out of interest, are those the suggested stakes by kelly or some derivative of kelly?

Full Kelly.

and yes, I would be sweating buckets watching 10% of my bank go on something that wins 11% of the time, but I'd do it :D

i have never bothered with a formal staking system for a number of reasons and i'm sure kellys merits have been discussed on here a million times - but - do you know if kelly is based on the premise that you will always find value of some (or a paticular) sort in the future?

I think that question delves more into the realms of philosophy.

Technically it doesn't, it just maximizes your growth, however if someone wanted to argue that the growth is only relevant if you continue to gamble, then I wouldn't really argue strongly back at them (though I would disagree )

It does allow for factors such as you buying a lottery ticket when the payout is double the odds (I did the calcs once and cant remember, but I seem to remember you shouldn't buy one unless your net worth is large, even if the jackpot is double the chances of winning)

A decent article on it with enough maths to be non-pathetic and little enough maths to be understandable is written by Ganchrow here and explains why Kelly is so strong.

http://forum.sbrforum.com/handicapper-think-tank/29009-expected-value-vs-expected-growth-kelly-criterion-part-i.html#post250260

The best bit of that article is that he explains how it can lead to bankroll shrinkage if you bet too much on a +EV shot.

cheers for that.

Does the margin of error really matter at all? If the current odds (taking commission into account, of course) are bigger than your own priced-up odds, shouldn't you always back it?

When pricing up an event, you will always have a margin of error in your prices, e.g. what you calculate is a 2.50 chance might really be a 2.30 or 2.70 chance. However, as long as your errors tend to overestimate as well as underestimate by similar amounts, the irregularities should even themselves out.

In other words, sometimes you'll be making bets with a negative value, and sometimes they'll have a very big positive value. But as long as the average is a positive value, you'll profit in the end. This all relies on your ratings being any good, of course!

good thread.

Thanks for your input guys, I understand that in theory less smoney should be staked on long shots than short prices due to the favourite/underdog bias and also because we are not betting an infinite amount of times but I won't be backing anything over 10-1 most probably and rarely over 5-1 for the most part.

Beagel 3 the point about prices being more accurate around the extremes is something I hadn't considered but am actually starting to realise with my testing. Not sure why this is exactly but most of the value I find exists around the even money shots. Any ideas??
Here 10% better than my prices exists but around the extremes it doesn't. Maybe subconsciously this is why a lot of people steer clear of these prices - they are very accurate and it is hard to profit from these.

I trialled the kelly for a while but then I realised I could never be 100% confident in my prices and even if I could I might find a 1.14 shot that I calculate as a 1.01 shot which according to kelly should be a 90% bank job but be on the receiving end of a 10 in 100 times it lost. If you start again and again you could possibly reach a traget easier but in sports betting one factor like weather or injuries could be more important in your 100% banker bet than usual in your otherwise consistent system. It's just too unmeasurable. This is why I am trialling 2% of bank and then recalculating 2% upon every 25% growth now.

Mr Magoo I thought that they should even themselves out but again if your prices allow for 10% then 4% inaccuracy won't be as damaging on one end and even more profitable on the other. I guess it again comes down to the fact that we don't have an infinite amount of bets or necessarily a model that fits for all and that can be measured easily, so is better to err on the side of caution.

Mr Magoo -
i suppose there are two ways of looking at this -
the first is that you are betting on a toss of a coin, and someone offers you 5/4 on one toss and someone offers you 6/4 on a different toss. you would undoubtedly stake more on the 6/4 toss, kelly suggests 16.7% of your bank on the 6/4 and 10% on the 5/4 (excusing that this calculation does not account for the simultanious placement of bets but this is not important). so from the point of view of bankroll optimisation it is important to know what margin you are dealing with.
the second is that in practice you are not betting on anything that has a known probability. assume all bets are made on betfair and your commission rate is 3%. you must have over a 1.5% edge to have a positive expectation at prices around even money (eg the market is evens both teams so true probability of the team you back must be 51.5%+). if you do your ratings completely independently, have one team 52% and so bet on them, you are giving zero respect to the other opinions in the market. i'm sure even most extremely successful punters would agree that on non-derivative markets if they compared their ratings to the markets the longterm probabilities would turn out to be somewhere between the two. if your sample is big enough you will know what weighting to use. say it is 50-50 then you would need to make it a 53% chance before expecting to break even. it is likely that the further you are from the market the more chance there is information you are unaware of. maybe you could weight it market price 70% and your price 30% whenever you are 10% away form the market price and 80-20 when you are 15% away. of course it depends on the event. if it's on a tennis player prematch you have obvious injury worries when you are so far different whereas if it is the same bet at the end of the first set that is televised live, you are far less likely to be unaware of something that could make your rating so wrong.
The point is that by recognising a level of competence by the 'market' and taking a weighted average between your price and the market price i think you will avoid the extremely marginal bets that may not cover the cost (commission and time) of placing them.

scrooge - i think it comes down to that the way we define 'value' is not very satisfactory. as illustrated above what we call 10% at the extremes is far more valuable to us than at even money. perhaps in reality 3-4% at 1/10 or 10/1 is as valuable as 10% at evens in terms of bankroll growth and security. you might expect to find those errors on an equal basis.

I thought it was more value the higher the odds to be honest? Am I wrong??

I think 20% better than 1.1 would give us less % profit than 20 better at evens... I know this is probably because statistically you will win more at 20% better than 1.1 but you have to win a lot more bets to profit at this price - I think this is why a lot of pro's don't back low odds regularly.

I have come to the conclusion that there is more edge around evens because these are the markets where it is obviously very difficult to come to a conclusion and any slightly better analysis or info is very valuable. This added to the fact you still have a good chance of winning your bet and you need to make less winning bets since each winning bet will cover the previous loss is why many gambler choose this strategy.

beagel 3, I take it you've read the paper 'Computer Based Horse Race Handicapping and Wagering Systems: A Report' by William Benter?

In that, he makes the same point that the public odds can reflect hidden knowledge, and so combining the current odds with your own tissue prices can therefore give you better profit.

I admit I find this terribly confusing - after all, by most standards, 'value' is where your own opinion differs from the available odds. So you've got this bizarre situation where a massive difference in prices can mean you've got great value, OR you've missed something critical and the market knows better. Or somewhere between the two.

One thing that Benter doesn't explain well is *when* to consider the publicly available odds to use them as a guide. With horse racing, the prices can fluctuate so much that their guidelines will give contradictory results. When does the 'live money' hit the market? Probably not first thing (only a few odds compilers having priced the markets up), probably not at the off (the gambles have already taken place, and been overbet), but somewhere between the two.

mr magoo, can i read that without subscribing? please post a link if you have one.