When kelly worsens your results .......

It likely means the level stakes is carrying too much risk of ruin.

thanks for responses.
i have the odds for 1600 football matches and i have calculated my perceived odds. i think i have an edge in about half the games. i have been betting profitably using level stakes for the last 600 matches . i successfully use kelly in some other stuff i run, but when i applied it to this system it reduced the profit and increased the max drawdown for these 1600 matches. i think the sample is large enough so i am a bit perturbed that kelly hurts the results. i suspect that there are some matches where for some reason i completely  miscalculate the odds and it kicks out a large kelly trade.

You might learn something by looking at the average stakes for each.

Suspect the level bets is higher, it could be other things, but that's the first thing I'd check.

racingguru Joined: 06 Jan 01
Replies: 301 07 Dec 10 22:27   


Kelly is no good for short prices -  period. Stick to a fixed staked system.


While it may be that you're not good at evaluating short prices, this is completely untrue if your pricing is accurate.

You can't argue with mathematics I'm afraid.

Not willing to get into this again with you for long, we've done this dance before. You make some valid points but in the wrong places. Kelly wasn't designed for anything, it's a factual result based on some strict assumptions.

At short prices say I make something 4/7 and i should have made it 4/6 and its price is 4/5 the difference is 18.1% to 10%. ie overstake by 8.1%(full kelly)

Sure, if you're completely awful at pricing, it will fail. Nobody should be pricing things as badly as the example above and gambling, let alone using Kelly.

I do recall a discussion before where it turned out that horse racing is completely different to any normal sport. It may well be that it's harder to price up than other sports. I don't know enough about it to judge, so I'll take your word on that. You should certainly be suicidal if you price things up as badly as your example above in other sports.

I would think pricing to an accuracy of 1% on a regular basis is near impossible. Leading
Bookmakers employ experienced compilers to price up football and it is normal for the best
and worst price to vary about 4%, just look at yesterday's prices, in one European Match
the favourite varied between 11/10 and 13/10.

Regards compilers, I realise some of them just copy other prices, but I've
found 365 to be quite accurate, also VC. I'm talking about football though
and I do my own pricing on match odds, then compare them with the books etc.
Horses I don't touch as there too many people know things I don't, except
perhaps on the big meetings on the Grade1 tracks. Staking; I do my own
staking according to price, the shorter the price, the bigger the stake,
the bigger the price, the smaller the stake. I can't bring myself to have
the same stake on, say a 5/1 as I do on an 1/1 chance. I do ok though. GL.

racingguru

I had a very,very long debate with Lori over Kelly's usefulness in horseracing, with my arguments very much in line with yours. I think the difference of opinion boils down to this. I may be wrong, but Lori operates in markets where there are a small number of outcomes in sports where statistical analysis is reasonably free from huge statistical variance. The problem with horseracing from an odds compiling standpoint is that a) there can be a large number of outcomes  b) crucially there is blowback from the market to the likely chances eg. connections feel someone has taken "their" price and go for an "easy" c) there are wipe-out factors which can kick in which can completely eliminate a horse's chance, many of which are unknown to you and even those that may be known, the effect is immeasurable  d) the number of factors which determine the performance of an individual horse are so numerous that you would need a team of a dozen or more working to input every variable for every race (these teams do exist in Hong Kong where it is an easier task.)

The net result is that one''s tissue is such a "soft" compared to the "hard"figures of price, bank, stake, that using Kelly or even a fraction of Kelly is a complete and utter waste of energy. I would always prefer to bet to the price as a percentage and go up i.e. bet heavier but still proportionate to price, in races where I believe my edge is greatest.

There is a not-so-subtle difference between mathematics and applied mathematics.

To me, Lori AND rg are both correct. You arrive at Kelly from first principles as the optimal Goldilocks stake to place when you know your independent price and the market price. Tis a well behaved function. However a single number, a point estimate, is rarely of any use or representative of the the real world.

You need fuzzyness, as its already there, and won't go away no matter how good you think your estimate is. But it needs to be a specific measure of fuzzyness. Working out Kelly is simple. Determining your system error to a realistic level is were most fall down.

In various systems I have I use Kelly both forwards to estimate appropriate stakes but also backwards as a selection criteria.

I would recommend limiting Kelly to a specific bakc or lay payout (not stake but total return including stake). Something like 15% of your bank, so enough to have a good win when a value bet comes off without risking too much when one has missed something big.

Could 'Kelly' successfully overcome a small level
stake loss to level stakes ?

In theory, yes, but it depends entirely on your ability to assess the 'true' chances of a particular result occurring correctly.

If level stakes are not profitable 'Kelly Criterion' would if implemented correctly, not suggest a bet. Which makes the point mute.
So in theory would be a definate no.

Another thought is that many posts are wasted as in many instances of literature, a person's convictions will not be
changed by words alone. A couple of examples that I can recall ~ 1. from "War and Peace"  I quote "This was his
acknowledgment of the impossibility of changing a man's convictions by words...." 2. Pygmalion "Words, Words, Words..."

Where you can try to convince someone of a theory by words that are contrary to their convictions you will fail, but
if you show them a practical and repeatable example you are quite likely to succeed.

Compound Magic Joined: 27 May 06
Replies: 527 13 Dec 10 07:30 
If level stakes are not profitable 'Kelly Criterion' would if implemented correctly, not suggest a bet. Which makes the point mute.
So in theory would be a definate no.
============================================================================================

CM - how can you say a "definite no"?

Someone backing LEVEL STAKES takes no account of "value" in the bet or the chances / probability of success of EACH BET.......Therefore, IN THEORY, if someone does have an edge ( which I doubt re someone backing LEVEL STAKES!) and applies KELLY to each bet .....at least they should then at least work out the probability of their bet winning more accurately than just betting the crude level stakes way!

Therefore the likelihood is that is LEVEL STAKES produced a small loss.....then there should be an  improvement due to being less CRUDE with staking!

BTW - I don't use KELLY....but logically there should be an improvement .......

Nice post CM.

Where you can try to convince someone of a theory by words that are contrary to their convictions you will fail, but
if you show them a practical and repeatable example you are quite likely to succeed.
And even then it is still difficult! To take an extreme example, if you have lived your life according to a set of principles for a number of years, and someone comes along and suggests you drop/replace some of them, you are not likely to accept the change unless there is a very high degree of certainty that it will improve your life.

DFCIRONMAN, Kelly can't turn a losing system into a winning one. Of course you are correct in saying that if someone successfully followed the procedure you described, they would start to make money. This is not an argument in favour of Kelly though, as it could also be turned into a winning system using level staking with the following one step process:

1) Stop placing bets where you have a negative expectancy.

The Investor - K said that a SMALL loss being made to LEVEL STAKES crude betting ........therefore in such specific circumstances .....applying your brain to bet consider the probability on each bet staked would probably lead to less amounts being applied where the value was lower...and applying more on bets considered stronger bets ( if you have a REAL EDGE ....this would be the position)......then the likelihood would be the SMALL LOSS would be eradicated................

Of course these "hypothetical" circumstances are so vague by K ....that all I am saying is that YOU CANNOT SAY "NO"......

Hi DFC hope you are still making a quid

In regards to my post ~ turning a level stake loss into a profit using 'Kelly'.

The likelihood is that showing a loss at level stakes is that you don't have an advantage % wise on a proportion of your bets so 'Kelly' would suggest no bet on those. So in theory you would not be able to use 'Kelly' for all bets. If you used Kelly on the proportion of bets that did have enough advantage % wise you could make a profit but then using those same bets at level stakes you would also make a profit.

Hope all that makes sense.

You must remember that as 'Kelly' has bigger bets on the losers than on the winners the difference plus commission plus a profit must be made up by 'Kelly' before it suggests a bet.

As the OP queries "Kelly' turning a level stake win into a loss. I suspect the sample is not large enough or that his level stake profit is derived from a higher winning % at the longer prices.

In some cases level stakes can be more profitable than 'Kelly' It all depends on the price ranges that are winning most often or  the edge is to small for 'Kelly' to suggest a bet as big as the level stake bets.

Thanks for views CM ......You say - "You must remember that as 'Kelly' has bigger bets on the losers than on the winners the difference plus commission plus a profit must be made up by 'Kelly' before it suggests a bet."
===========================================================================

I really can't understand what you mean by the bit in bold? Surely , if you have an "edge" you will tend to select bets where there are good value in bets. Kelly presumably would have bigger stakes on such extra value bets.....and if you have an edge....you will get winners with more on.

The bets that lose surely are more likely to be ones where your edge has viewed their chances of winning to be less than "strong chances to win". I mean by that that the extra value would tend to be less generally on such bets.


You also say - "If you used Kelly on the proportion of bets that did have enough advantage % wise you could make a profit but then using those same bets at level stakes you would also make a profit."


I've always thought that using KELLY would mean putting larger stakes on bets where your edge has identified strong chances to win that the market has not! If you just stuck on LEVEL STAKES on such bets , then you would definitely not make as much as you are not varying your stakes to take account of the value in the bet.

Anyway ....I don't use KELLY....though have tried using the KELLY CALCULATOR on HORSE bets.... I find that the TIME taken up by calculating stakes etc can be very wearing when betting on many races.

However, logically it definitely is better to apply your brain to STAKING on every bet ...theoretically....as there is no logic at all to betting LEVEL STAKES IMO.

GL with bets ..no matter how you stake em!

I've just been re-reading this thread and have noticed that racingguru's contributions,to which both Lori and Sandown had responded, have mysteriously disappeared.

Strangely the best explanation of kelly I have ever seen is in the Daily Mirror's book called Better Betting where he takes an example of tossing a coin using kelly staking to show how it is likely to improve your return no matter the outcome (win/lose/draw). Easy to understand explanation and the math works. He then gives racing examples using half Kelly. V interesting couple of chapters in what is, unexpectly, a v interesting book.

@ Compound Magic

As the OP queries "Kelly' turning a level stake win into a loss. I suspect the sample is not large enough or that his level stake profit is derived from a higher winning % at the longer prices.

my sample is 800 bets which I think should be large enough (?), and kelly does not turn a level stakes into a loss, it just generates a lower (and more volatile) return. my conclusion so far,  is that I have an edge, but I am not quantifying it correctly.  somebody suggested introducing a margin of error into the edge being calculated which makes a lot of sense. i guess if you make the margin of error large enough you end up with level stakes.

guess if you make the margin of error large enough you end up with level stakes.

or more likely, no-bet.