1. #1
    Wrecktangle
    Wrecktangle's Avatar Become A Pro!
    Join Date: 03-01-09
    Posts: 1,524
    Betpoints: 3209

    Why the Kelly Criterion is not a good bankroll management system

    I see a lot of postings about the Kelly Criterion posted here at SBR. If you don’t know what the Kelly Criterion is, read http://en.wikipedia.org/wiki/Kelly_criterion . And while the good Dr. Kelly “proved” that his Kelly Criterion was the optimal amount to wager in a Bell Telephone Tech Journal paper published in 1956, you as a sports bettor should NOT use it.

    The chief reason is not that Kelly it is a bad formula, but rather that the Kelly bet fraction you calculate [basically, your losing percentage plus the casino advantage subtracted from your winning percentage, i.e. Kelly = Win% – (Lose% + vig) ] is for all intents incalculable by almost all sports bettors, probably even the Ganchrows of the world.

    Why is this? Because you must calculate that win % as a Bayesian Prior. If you don’t know much about Bayes and his theorem, read http://en.wikipedia.org/wiki/Bayes'_theorem.

    Now, Kelly itself works pretty well in Blackjack, and came to fame in Ed Thorp’s Beat the Dealer, but the Bayesian Prior’s you calculate in BJ are 1) smaller as a whole than most sports win fractions thus less likely to affect your bankroll if you make a mistake 2) much easier because you have a well known (but not easy) statistical method to calculate Kelly Criterion in 21.

    I contend that in sports situations you have four problems to deal with:
    1) understanding Bayesian Priors in the first place – this is actually the easiest part, but if your head is spinning after reading some of Ganchrow’s posts you may have a problem.
    2) calcing your win % given that your particular method may be suspect - I contend that angle players have a real problem here.
    3) the particular sport league you are betting against evolves due to changing rules, changing players, and changing team management i.e. coaching decisions.
    4) the line changes due to market perceptions of these league changes.

    Now, I can write a lot about each of these four points, and I doubt that any of you are convinced of all four, but if you believe any one of the four is in place, you cannot accurately calc your Prior. And without an accurate Bayesian Prior, you cannot do Kelly…period.

    OK, that’s enough to start the fight. Tell me why I’m wrong.

  2. #2
    bookie
    bookie's Avatar Become A Pro!
    Join Date: 08-10-05
    Posts: 2,083
    Betpoints: 1012

    I think you're asking a false question. Or maybe it's false because you're not even asking it.

    The question is: What's the best money management strategy? It's not a question of whether Kelly is good or bad, it's a question of how it stacks up against alternatives.

    Your argument that you can't understand Bayesian Priors could be extended to say that you can't know if you are +EV in any sports betting situation save past-posting, and therefore whether any sports betting anyone does ever is rational.

    Finally no one that I know of argues that a "pure kelly" approach makes any practical sense. Everyone that I know of argues for a partial fractional approach, trying to build in for the fact that we ego-centered fallible handicappers are prone to overestimate our advantage. Whether a sound Kelly Fraction is then a fifth of advantage, a third, a half...that decision is part of the art of handicapping--one which like all decisions made under conditions of uncertainty can be made incorrectly.

    But, just because a kelly estimate can be made incorrectly it doesn't follow that no one should ever make decisions from within the Kelleyian framework.

  3. #3
    MonkeyF0cker
    Update your status
    MonkeyF0cker's Avatar Become A Pro!
    Join Date: 06-12-07
    Posts: 12,144
    Betpoints: 1127

    Well said, bookie.

  4. #4
    Dark Horse
    Deus Ex Machina
    Dark Horse's Avatar Become A Pro!
    Join Date: 12-14-05
    Posts: 13,764

    I'm on the fractional bandwagon. My goal is not just to maximize profit (Kelly), but to find a balance between profit and peace of mind. With a little effort anyone can find the balance that is best suited to their personality. (Someone with a psychology and betting background might be able to devise a test that will give you your fractional Kelly. )

    On a sidenote, if anyone has insight into this, I'd be interested to learn more about doing the same in the stock market. I'm looking for individualized ratios between day trading, long term investing, and options that are suited to different personalities/mindsets.
    Last edited by Dark Horse; 04-26-09 at 11:45 AM.

  5. #5
    Wrecktangle
    Wrecktangle's Avatar Become A Pro!
    Join Date: 03-01-09
    Posts: 1,524
    Betpoints: 3209

    Let's clear up a misconception. I use FULL Kelly, but I'm a hard math guy, up on the latest in model building, time series forecasting, etc, and I have trouble with it. It is provably OPTIMAL, and if you want to maximize your bankroll over any sort of time frame, you have to use it.

    But 99.999% of the bettors don't, can't, and will go broke faster because of it, fractional or otherwise.

    If you guys think you really know how to use it, I think that's great. Its your bankroll, your money, lose it as fast as you'd like. But unless you absolutely keep strict records on every thing you do or thought about doing (i.e. a situation missed your criteria), you are NOT getting decent priors. Unless you do this, and damn few folks do, you cannot make Kelly work. Matter of fact, everyone I check with fails in some regard when pressed.

    And if you don't know your priors, you can't do fractional Kelly either. BTW, Fractional Kelly is NOT Kelly, only Full Kelly is optimal i.e. maximizes BR growth while minimizing risk of BR loss.

  6. #6
    Data
    Data's Avatar Become A Pro!
    Join Date: 11-27-07
    Posts: 2,236

    I find the whole premise of maximizing bankroll via sportsbetting ridiculous due to the shallowness and the shadiness of that market. Kelly is a fine theoretical framework that has no application in sportsbetting as in an investment venue.

    On a side note, it has been discussed many times in this forum that most people, including Kelly followers, do not realize that their Kelly bankroll is two to four orders of magnitude larger than what they think it is.

  7. #7
    tomcowley
    tomcowley's Avatar Become A Pro!
    Join Date: 10-01-07
    Posts: 1,129
    Betpoints: 6786

    Quote Originally Posted by Wrecktangle View Post
    I see a lot of postings about the Kelly Criterion posted here at SBR. If you don’t know what the Kelly Criterion is, read http://en.wikipedia.org/wiki/Kelly_criterion . And while the good Dr. Kelly “proved” that his Kelly Criterion was the optimal amount to wager in a Bell Telephone Tech Journal paper published in 1956, you as a sports bettor should NOT use it.

    The chief reason is not that Kelly it is a bad formula, but rather that the Kelly bet fraction you calculate [basically, your losing percentage plus the casino advantage subtracted from your winning percentage, i.e. Kelly = Win% – (Lose% + vig) ] is for all intents incalculable by almost all sports bettors, probably even the Ganchrows of the world.

    Why is this? Because you must calculate that win % as a Bayesian Prior. If you don’t know much about Bayes and his theorem, read http://en.wikipedia.org/wiki/Bayes'_theorem.

    Now, Kelly itself works pretty well in Blackjack, and came to fame in Ed Thorp’s Beat the Dealer, but the Bayesian Prior’s you calculate in BJ are 1) smaller as a whole than most sports win fractions thus less likely to affect your bankroll if you make a mistake 2) much easier because you have a well known (but not easy) statistical method to calculate Kelly Criterion in 21.

    I contend that in sports situations you have four problems to deal with:
    1) understanding Bayesian Priors in the first place – this is actually the easiest part, but if your head is spinning after reading some of Ganchrow’s posts you may have a problem.
    2) calcing your win % given that your particular method may be suspect - I contend that angle players have a real problem here.
    3) the particular sport league you are betting against evolves due to changing rules, changing players, and changing team management i.e. coaching decisions.
    4) the line changes due to market perceptions of these league changes.

    Now, I can write a lot about each of these four points, and I doubt that any of you are convinced of all four, but if you believe any one of the four is in place, you cannot accurately calc your Prior. And without an accurate Bayesian Prior, you cannot do Kelly…period.

    OK, that’s enough to start the fight. Tell me why I’m wrong.
    You're trying too hard. The ability to bet a line that other people have passed on is a significant indication that your estimate of edge, without the information that they've passed, is too high. None of the other stuff is even relevant (or necessarily correct). If you can't estimate your win%, then you couldn't bet at all, so attacking any staking strategy on those lines is just silly. Your prior win% is X (or distribution of win% if you prefer). Your estimate of win% after you see a bettable line (that people with a clue have passed on) has to be lower, possibly significantly lower.

    When you see some completely dumbass prop, where it's obvious that the bookie is asleep or doesn't even know what he's offering (or the bettors don't know what they're betting on), and it's unlikely that anybody with a clue has been by and passed on it, it's perfectly reasonable to fire full-kelly. I've seen props that are trivial to cap where the offered win% was wrong by more than 20%- not often, but occasionally it happens. Sadly, betting anything near kelly on those wasn't happening after my first few months in the game, but if I could, I'd fire a large kelly fraction without a second thought.

  8. #8
    Dark Horse
    Deus Ex Machina
    Dark Horse's Avatar Become A Pro!
    Join Date: 12-14-05
    Posts: 13,764

    Quote Originally Posted by Wrecktangle View Post
    Let's clear up a misconception. I use FULL Kelly, but I'm a hard math guy, up on the latest in model building, time series forecasting, etc, and I have trouble with it. It is provably OPTIMAL, and if you want to maximize your bankroll over any sort of time frame, you have to use it.
    You speak out against full Kelly, here as well as earlier, yet you use it, and have trouble with it...

    Does that count as two or three contradictions?


    The problem is not so much the objectivity of betting systems, but the objectivity of the individual. At a subtle level the outcome of any bet affects the mind, even if a person thinks it doesn't. A win gives a subtle level of joy, and a loss gives a slight disappointment. This is why I stress peace of mind. Without it, objectivity is lost. And without objectivity, the way we perceive the world is 'tinted'. And with tinted vision, how good are those objective systems? It all begins with the mind.

  9. #9
    bookie
    bookie's Avatar Become A Pro!
    Join Date: 08-10-05
    Posts: 2,083
    Betpoints: 1012

    Quote Originally Posted by Wrecktangle View Post
    Let's clear up a misconception...
    What's the misconception that we're clearing up?

  10. #10
    Wrecktangle
    Wrecktangle's Avatar Become A Pro!
    Join Date: 03-01-09
    Posts: 1,524
    Betpoints: 3209

    I use it because it is optimal, and measure any BR method against it when I backcast a season or forecast a bet. The difficulty lies in the Priors, not in Kelly. And difficulty is no argument to not use it, it simply makes me be more precise.

    If you guys really know what you're doing, then perhaps you're in the .01% "smart" guys I allowed for.

    Data, you are simply wrong. I suppose you think the stock market is less shady? It is largely driven by insider knowledge. I contend that the sports market is much more open than any stock market you might pick.

    The sports market, measured in Billions per year, is much broader than you might think, not nearly as broad as stocks, since we have Trillions.

  11. #11
    Dark Horse
    Deus Ex Machina
    Dark Horse's Avatar Become A Pro!
    Join Date: 12-14-05
    Posts: 13,764

    This thread is named: Why the Kelly Criterion is not a good bankroll management system

    But you seem to be making the case that you know what you're doing, and since you know what you're doing and have trouble with Kelly, therefore, since others couldn't possibly (unless they belong to 0.01% smart guys) know their business better than you, it is a bad system.

    Seems rather subjective, if you ask me. My question would be why you would bet full Kelly if you have trouble with it.

    Full Kelly becomes even more theoretical for larger bankrolls. How are you going to get down 60% of a 100K or 200K bankroll? So, even if a bettor were good enough to use full Kelly correctly, he would soon run out of operating room. Not in the perfect universe held together by flawless theories, but in the real world.

  12. #12
    reno cool
    the meaning of harm
    reno cool's Avatar Become A Pro!
    Join Date: 07-02-08
    Posts: 3,567

    I said before that Kelly is better for games such as Blackjack where you can get many decisions in a short amount of time and your edge is small and can be calculated exactly.
    A typical sport better would be better off figuring out personal goals, needs and tolerance and devising a betting strategy in accordance. Most people don't intend on risking a bankroll= their net worth nor stick to their planned strategy independent from its actual result.

  13. #13
    Data
    Data's Avatar Become A Pro!
    Join Date: 11-27-07
    Posts: 2,236

    Quote Originally Posted by Wrecktangle View Post
    Data, you are simply wrong. I suppose you think the stock market is less shady? It is largely driven by insider knowledge. I contend that the sports market is much more open than any stock market you might pick.
    I would call the property of the markets you were referring to as fairness. I believe that those markets are mostly fair and the impact caused by people with an inside knowledge is small. By shadiness I mean that the bookies operate out of legal boundaries hence the place we are having this conversation at. When a bettor sends a ** to Miguel A. Rivera it is not the same as putting his money on the brokerage account.

    The sports market, measured in Billions per year, is much broader than you might think, not nearly as broad as stocks, since we have Trillions.
    Yes, and most of the money is controlled by the mob but that falls into category above. If you do not deal with the mob and play at semi-legal off-shore books your a very limited in how much you can put in. A US based full Kelly bettor would have to make 6 and more figures bets on a regular basis and the market cannot accommodate that.

    The markets for some stocks a shallow too, by the way. Of course, the markets for other stocks could be pretty active and deep and nothing compares with FOREX in volume.

  14. #14
    Wrecktangle
    Wrecktangle's Avatar Become A Pro!
    Join Date: 03-01-09
    Posts: 1,524
    Betpoints: 3209

    Quote Originally Posted by Data View Post
    I would call the property of the markets you were referring to as fairness. I believe that those markets are mostly fair and the impact caused by people with an inside knowledge is small. By shadiness I mean that the bookies operate out of legal boundaries hence the place we are having this conversation at. When a bettor sends a ** to Miguel A. Rivera it is not the same as putting his money on the brokerage account.

    Yes, and most of the money is controlled by the mob but that falls into category above. If you do not deal with the mob and play at semi-legal off-shore books your a very limited in how much you can put in. A US based full Kelly bettor would have to make 6 and more figures bets on a regular basis and the market cannot accommodate that.

    The markets for some stocks a shallow too, by the way. Of course, the markets for other stocks could be pretty active and deep and nothing compares with FOREX in volume.
    Well, yes since our last President W enacted illegal legislation outside the treaty boundaries of the WTO, it is mostly "shady" in the US. That will most likely be reversed, but probably for tax purposes, not as a freedom.

    6+ figure bets are possible, not just in the same book. In Vegas, higher are done also, just got to live there or closely near-by.

    But I take your point. In the US, our "Nanny Government" will keep us from going wrong.

    Along with a wrong-headed drug, and sex industry policies, the US has a long way to go to achieve real personal freedoms.

  15. #15
    Wrecktangle
    Wrecktangle's Avatar Become A Pro!
    Join Date: 03-01-09
    Posts: 1,524
    Betpoints: 3209

    Points made about the "headroom" of Kelly betting (6 figure bets, etc) I completely agree with. But I'd wager most folks would like to have that problem if their original BR were small. But this helps make my original point.

    I'll try to achieve Kelly if only to grade whatever BR management system I'm currently using. I feel that grading my system against a theoretically optimal cannot be bad. There is little real risk using Kelly for the small BRs I play with, and it sharpens my Prior calculations and that can only be good. And if we ever achieve real freedoms in Nanny State Amerika, maybe I'll achieve the monster BRs you folks play with.

    Back to the original point: for almost everyone reading this forum, Kelly is really not a practical system, and as Data so correctly points out, certainly not here in the Nanny Republic.

    "Thank you, Thank you verry much"

  16. #16
    acw
    acw's Avatar Become A Pro!
    Join Date: 08-29-05
    Posts: 576
    Betpoints: 1564

    Quote Originally Posted by Dark Horse View Post
    peace of mind.
    I really loved this comment! Who's peace of mind?

    With a little effort anyone can find the balance that is best suited to their personality. (Someone with a psychology and betting background might be able to devise a test that will give you your fractional Kelly. )
    Yes, I would be very happy, if someone can give me list of how much bookies are willing to lose!

  17. #17
    u21c3f6
    u21c3f6's Avatar Become A Pro!
    Join Date: 01-17-09
    Posts: 790
    Betpoints: 5198

    A lot of good points above.

    Here’s my take. I find that a lot of people tend to make things more complicated than they need to be. This is not to criticize. If you enjoy the output of energy and/or it gets you the results you want, great. But you can also over analyze something IMO and miss the lesson to be learned. There are not a lot of absolutes for me. As soon as someone says you “have” to or it “has” to be done this way or you “can’t”, I am pretty sure that there are those that don’t do it that way and/or can and are successful. However, understanding how to calculate the amount for full-Kelly has helped me focus on my wagering strategies. If I only focused on whether or not I can use full-Kelly, I would not have learned what I use to develop my wagering strategies.

    What did I learn? I am relatively conservative and forego the “extra” profit at times to safeguard my bankroll against a large hit and large fluctuations. I have found that I perform best when my bankroll is slowly creeping higher with minimal “small” hits than when it goes up a lot and then goes down a lot (even if I still show profit). The funny thing is, if you are successful, there will come a time when you are wagering an amount that used to be your whole starting bankroll! It can be a matter of taste but it is also a matter of mathematics. Based on full-Kelly, you are much better off betting a system with a 10% edge that hits 1 out of 2 times than a system with a 20% edge that hits 1 out of 11 times. Even though theoretically you could make more money with a 20% edge, you can wager more money safely with the 10% edge system than you can with the 20% edge system. A safe amount to wager on a system with a 20% edge that hits 1 out of 11 times is 2% of bankroll. For a $1,000 bankroll the wager would be $20 with an expectation of a $4 gain. A safe amount to wager on a system with a 10% edge that hits 1 out of 2 times is 10% of bankroll. For a $1,000 bankroll the wager would be $100 with an expectation of a $10 gain which is 2 ½ times more profit than the system that has twice the edge. Even if you only bet 5% of bankroll ($50) on the 10% system, you still reap a greater profit ($5 expectation) with greatly increased safety. Of course this is if you are concerned about bankroll growth. Someone that is looking for more “excitement” and/or doesn’t mind large fluctuations of their bankroll might prefer to go for the “bigger” gains. Their choice.

    Full-Kelly is only concerned with the long run, requires accurate input but can also have large fluctuations. I look for strategies that have a higher win rate based on what I wrote above. I also “know” (within reason) my edge for the various wagers I make. My “edge” is different for my U/O wagers compared to my spread wagers and so on. So my % wager is different dependent upon what type of wager I am making. I then use a half-Kelly for a couple of reasons. One, to safeguard against an over-estimation of my edge and/or a possible fall-off in edge % and two, to cut down on the larger fluctuations that full-Kelly can produce, which fits my personality better. I very rarely have more than 3 or 4 wagers “working” at a time so I don’t worry about doing any Bayesian type calculations. In addition, I view my bankroll in plateaus, so I only raise or lower my wagers based on the half-Kelly % when my bankroll enters a new plateau (this means that most often I am actually wagering somewhat less than half-Kelly).

    There is so much more that can be written but I will stop for now.

    Dark Horse: I used the same methodology to develop my sports betting that I used for my stock options trading. I “knew” my edge when I set-up a trade and allocated a half-Kelly of my funds to that trade. Again, I did not have many trades "working" at one time so I didn't need to do any Bayesian type calculations.

    Joe.
    Last edited by u21c3f6; 04-28-09 at 09:44 AM. Reason: Spacing

  18. #18
    Dark Horse
    Deus Ex Machina
    Dark Horse's Avatar Become A Pro!
    Join Date: 12-14-05
    Posts: 13,764

    Quote Originally Posted by acw View Post
    I really loved this comment! Who's peace of mind?
    The guy who says that your maximum bet size is the max amount you're willing to lose without causing you any stress. It's a pretty common approach.

    Wrecktangle, if you have been at this for a number of decades, as you say you have, then certainly, based on age, a bankroll of 200K is not monstrous (although it would depend on your definition of BR, which in true Kelly would include all your assets...). I have no doubt you know your stuff, but you really are all over the place in this thread.

  19. #19
    Dark Horse
    Deus Ex Machina
    Dark Horse's Avatar Become A Pro!
    Join Date: 12-14-05
    Posts: 13,764

    In any case, from what I know of sports and stocks, I believe the place to bet full Kelly is the stock market, where you can work with stop losses, so you either win a lot or lose a little. There is no need to pay full price for being wrong.

    Also, in the stock market you're up against an extremely predictable public (I include brokers here), which sets the price (often based on fear), where the line in sports betting is set by unusually sharp people. Big difference.

    Completing this line of thinking, if you can beat sports, you can absolutely beat the stock market. Sports has the bar set much higher, and as such is an ideal training ground.
    Last edited by Dark Horse; 04-27-09 at 03:09 PM.

  20. #20
    durito
    escarabajo negro
    durito's Avatar Become A Pro!
    Join Date: 07-03-06
    Posts: 13,173
    Betpoints: 438

    Stock markets are far more efficient than sports betting markets.
    Points Awarded:

    WileOut gave durito 6 SBR Point(s) for this post.


  21. #21
    Arilou
    Arilou's Avatar Become A Pro!
    Join Date: 07-16-06
    Posts: 475
    Betpoints: 949

    Kelly doesn't just have a Baysean priors problem, it has a meta problem that those priors are correlated. Any given strategy, and to a lesser extent any given gambler will have an edge on his wagers that is highly correlated between his different bets relative to his estimation. If you're trying to bet Kelly, and you correctly estimate that your edge given what you know is 5%, that includes worlds where it is 2% and ones where it is 8% both with reasonably high probability. Thus if you bet anything close to Kelly based on average edge you are giving yourself a high percentage of worlds where you are betting high enough that your bankroll growth in Kelly terms will be negative. In the meantime, in the 8% worlds where you are underbetting you are getting rich faster, but that additional wealth has less marginal utility to you because you were already going to do well. There's also the problem that Kelly assumes transaction costs are zero, that betting limits don't exist (the marginal value of the ten millionth dollar in your bankroll is essentially 0 unless your edge is huge) and other similar horses-as-three-meter-spheres. I like to calculate the Kelly bet size because it is useful to frame the question of what to bet, and as a point of no return: If you pass this point then you went too far. In more practical terms, Kelly betting or even fractional Kelly betting is a great approach when you know you are right but it makes mistakes and traps far more deadly.

    That brings me to Data's point about your Kelly bankroll being larger. I know where that comes from, but there are several things that I would point out. First, the uncertainty noted above is even more of an issue if you are commiting everything you own and you take it to the next level beyond that if you're willing to beg, borrow and/or steal. Also, people have to pay fixed costs of living and of maintaining their gambling infrastructure. This means that things going very badly, to the point of having to take loans or worse, wipes out Kelly's evaluation of your utility in such scenarios. Such scenarios are very, very bad. On the other hand, if you are indifferent between taking the day job and gambling because your bankroll is too small you can pick up utility by risking that bankroll since the upside is that you can continue to bet whereas the downside costs only the money since gambling "correctly" had no utility for you. I don't know of a sane gambler who applies Kelly to their life bankroll rather than to their gambling bankroll, and with many good reasons.

    As a theoretical construct I love Kelly. It has helped me a lot, in many situations. Fractional Kelly is a far better approach than flat betting, if you apply some common sense. I respect those who love it as a practical guide, but I prefer to think of it as a signpost.

  22. #22
    Wrecktangle
    Wrecktangle's Avatar Become A Pro!
    Join Date: 03-01-09
    Posts: 1,524
    Betpoints: 3209

    DH: I'm not a pro bettor, I'm a math modeler. I've sold my expertise to touts and managed small BRs. In my last instance I took $1000 into $23000, pulled out $21000, paid taxes on it and put it into a rocket hot real estate market where it bloomed into 6 figures with a lot less work and strife than sports betting. The $2000 sank with Aces Gold as I waited too long to get my bux out. Now, I've held on to my best real estate, and have watched it plummet as I bought into Greenspan's "we've got the best managed economy in the world and the upcoming recession will be mild."

    Small potatoes I'm sure compared to many here. But I used Kelly to manage that small BR, in NFL sides only, over 2 seasons, all legally with a ton of work (avg 47 hrs per week not counting watching games). I have also had two seasons where I had less than 1/2 of my original BR by mid-season, when I quit, and several seasons where I was up a little, or down a little, two as I recall where I doubled. But it took several seasons to make Kelly work, and I had to get serious about Priors, and build not a few money management sims along the way. I find most folks don't have the time, math background, programming skills, and/or patience I had to make it work because "real" bettors mostly want action.

    I don't have the fever, but I do like modeling sports, especially NFL sides which is my specialty. When I can do it legally, I do. Otherwise I don't. I don't live in Vegas and in my locale, internet betting is illegal.

    So, this is my Kelly story in a nutshell. I left in 2003 and with the crash of the r.e. market, I'm now back working at something I once was good at.

    So maybe this tale answers a lot of the Qs posed by others too.
    Last edited by Wrecktangle; 04-27-09 at 11:03 PM. Reason: spellos & clarity

  23. #23
    Dark Horse
    Deus Ex Machina
    Dark Horse's Avatar Become A Pro!
    Join Date: 12-14-05
    Posts: 13,764

    Kelly wasn't specifically designed for sports betting. That may not make a difference for strictly mathematical edges, but it does for other angles. I doubt Kelly is the final answer, where it comes to bet sizing in sports betting. I just don't think the best answer is that one-dimensional. Percentages are given 'depth' by Z-factors, and Z-factors in turn get more perspective with action points. In addition, even strong systems don't necessarily produce each season, but can skip a season. It's important to recognize that. And there are always experimental edges, that may or may not develop into systems. Again, Kelly doesn't cover that. To take only a percentage, without any other qualifications, and translate it into a huge bet size is pretty absurd in my mind (unless the angle is purely mathematical). I'm still waiting for a better formula... (am looking at factoring in the action points for losses versus the action points for wins).
    Last edited by Dark Horse; 04-28-09 at 02:52 PM.

  24. #24
    reno cool
    the meaning of harm
    reno cool's Avatar Become A Pro!
    Join Date: 07-02-08
    Posts: 3,567

    Quote Originally Posted by Dark Horse View Post
    Kelly wasn't specifically designed for sports betting. That may not make a difference for strictly mathematical edges, but it does for other angles. I doubt Kelly is the final answer, where it comes to bet sizing in sports betting. I just don't think the best answer is that one-dimensional. Percentages are given 'depth' by Z-factors, and Z-factors in turn get more perspective with action points. In addition, even strong systems don't necessarily produce each season, but can skip a season. It's important to recognize that. And there are always experimental edges, that may or may not develop into systems. Again, Kelly doesn't cover that. To take only a percentage, without any other qualifications, and translate it into a huge bet size is pretty absurd in my mind (unless the angle is purely mathematical). I'm still waiting for a better formula... (am looking at factoring in the action points for losses versus the action points for wins).
    good points.
    also if you don't count your bankroll as all money ever available (which almost no one does) the idea of maximum growth becomes moot.

  25. #25
    BigCap
    BigCap's Avatar Become A Pro!
    Join Date: 02-10-08
    Posts: 189

    Quote Originally Posted by tomcowley View Post
    You're trying too hard. The ability to bet a line that other people have passed on is a significant indication that your estimate of edge, without the information that they've passed, is too high. None of the other stuff is even relevant (or necessarily correct). If you can't estimate your win%, then you couldn't bet at all, so attacking any staking strategy on those lines is just silly. Your prior win% is X (or distribution of win% if you prefer). Your estimate of win% after you see a bettable line (that people with a clue have passed on) has to be lower, possibly significantly lower.
    Example: I pick NFL sides between 51.5 and 57% rate per year over the past 10 years. So either I'm barely breaking even or I'm killing them, so no good estimate of win percentage is available. Does this mean I don't bet at all? Of course not.

    Quote Originally Posted by tomcowley View Post
    When you see some completely dumbass prop, where it's obvious that the bookie is asleep or doesn't even know what he's offering (or the bettors don't know what they're betting on), and it's unlikely that anybody with a clue has been by and passed on it, it's perfectly reasonable to fire full-kelly. I've seen props that are trivial to cap where the offered win% was wrong by more than 20%- not often, but occasionally it happens. Sadly, betting anything near kelly on those wasn't happening after my first few months in the game, but if I could, I'd fire a large kelly fraction without a second thought.
    This is just an academic exercise because anybody with serious br's get capped out very quickly with props. You need to be brain-dead not to max-out mispriced props that you describe with any reasonable br.

  26. #26
    Deledio2Tambling
    Deledio2Tambling's Avatar Become A Pro!
    Join Date: 04-06-09
    Posts: 8

    Seems like a great discussion guys, so sorry if I n00b it up here, but how do you find the 'p' and 'q' value in the Kelly criterion?

    http://en.wikipedia.org/wiki/Kelly_criterion

    Is the p and q what you personally perceive the probabilty of winning to be? Or is it a number that can be found within the odds?

    Sorry if this doesn't make much sense.

  27. #27
    Pancho sanza
    Pancho sanza's Avatar Become A Pro!
    Join Date: 10-18-07
    Posts: 386

    Quote Originally Posted by Deledio2Tambling View Post
    Seems like a great discussion guys, so sorry if I n00b it up here, but how do you find the 'p' and 'q' value in the Kelly criterion?

    http://en.wikipedia.org/wiki/Kelly_criterion

    Is the p and q what you personally perceive the probabilty of winning to be? Or is it a number that can be found within the odds?

    Sorry if this doesn't make much sense.
    Your estimate.

  28. #28
    Deledio2Tambling
    Deledio2Tambling's Avatar Become A Pro!
    Join Date: 04-06-09
    Posts: 8

    Thanks Pancho

    So lets say you yourself Pancho rate a team a 'good', but not 'great' chance to win......what sort of % would you give of that coming off?

  29. #29
    Wrecktangle
    Wrecktangle's Avatar Become A Pro!
    Join Date: 03-01-09
    Posts: 1,524
    Betpoints: 3209

    D2T: If you are having trouble with the Kelly equation in that wiki article, stay with that BR management article I sent you (Century System).

    Newbies have absolutely no reason to be messing about with Kelly. Until you can handicap and have a positive win %, stay with a simple unit system that can progress if you are winning and your BR is growing under it.

  30. #30
    Deledio2Tambling
    Deledio2Tambling's Avatar Become A Pro!
    Join Date: 04-06-09
    Posts: 8

    Quote Originally Posted by Wrecktangle View Post
    D2T: If you are having trouble with the Kelly equation in that wiki article, stay with that BR management article I sent you (Century System).

    Newbies have absolutely no reason to be messing about with Kelly. Until you can handicap and have a positive win %, stay with a simple unit system that can progress if you are winning and your BR is growing under it.
    Thanks for your help mate!

  31. #31
    Dark Horse
    Deus Ex Machina
    Dark Horse's Avatar Become A Pro!
    Join Date: 12-14-05
    Posts: 13,764

    D2T, Justin made a good video about it:

    http://www.youtube.com/watch?v=IyATmCJf4fc

  32. #32
    tomcowley
    tomcowley's Avatar Become A Pro!
    Join Date: 10-01-07
    Posts: 1,129
    Betpoints: 6786

    Quote Originally Posted by BigCap View Post
    Example: I pick NFL sides between 51.5 and 57% rate per year over the past 10 years. So either I'm barely breaking even or I'm killing them, so no good estimate of win percentage is available. Does this mean I don't bet at all? Of course not.
    If you don't estimate your edge over breakeven, you don't bet. If you do, then what number is it? You're dodging the question by claiming that you have an edge but you have absolutely zero ability to estimate it, which is nonsense. You can't have some level of confidence that your win% is over X, but not have any degree of confidence in any number >X. Assuming -110, you're confident enough that you're higher than 52.4, but you have no condidence that you're above 52.5%? 52.6%? GMAFB.



    This is just an academic exercise because anybody with serious br's get capped out very quickly with props. You need to be brain-dead not to max-out mispriced props that you describe with any reasonable br.
    On props, yes. That's exactly what I said. On other stuff that gets stupid, especially if you're taking long odds, you can max out on a decent bankroll.
    Last edited by tomcowley; 04-29-09 at 01:10 AM.

  33. #33
    marcoforte
    marcoforte's Avatar Become A Pro!
    Join Date: 08-10-08
    Posts: 140
    Betpoints: 396

    Having studied Kelly, I find it's not for me. Your issue of bayesian priors is spot-on. While I can estimate a range of my winning % based on over 2 decades of betting, it's only an estimate to be proven in the future. For this reason, I set the bankroll and then flatbet it at 2.5% of the original bankroll on each bet. I'm a high volume bettor playing over 200 bets in a 12 week NFL season starting at week 5. The other thing I don't like is that because I'm a one sport per year bettor, if I start badly, I can never catch up because the season is too short and the bets too small to build back quickly. I prefer to flat bet, at least I have a chance to break even. The best treatise on this in layman's terms is in Buchdahl who at the end of the explanation leaves it to the reader to make up their own mind.

    Durito - I agree but respectfully disagree. At any given moment of time, it is efficient based on available information. It is not efficient in the larger time perspective. Witness the fact of what has happened in the market from the end of February to now. I followed my motto and sunk a ton of cash into the market at the end of Feb. It has almost doubled. The market has a bad perspective on the future depending on how it feels currently.

  34. #34
    BigCap
    BigCap's Avatar Become A Pro!
    Join Date: 02-10-08
    Posts: 189

    Quote Originally Posted by tomcowley View Post
    If you don't estimate your edge over breakeven, you don't bet. If you do, then what number is it? You're dodging the question by claiming that you have an edge but you have absolutely zero ability to estimate it, which is nonsense. You can't have some level of confidence that your win% is over X, but not have any degree of confidence in any number >X. Assuming -110, you're confident enough that you're higher than 52.4, but you have no condidence that you're above 52.5%? 52.6%? GMAFB.
    If I am 99% confident that my range of picks falls between 51.5% and 57%, how does that help me kelly-stake my bets? How can anybody say they pick 53.1%? That is ridiculous.

    Quote Originally Posted by tomcowley View Post
    On props, yes. That's exactly what I said. On other stuff that gets stupid, especially if you're taking long odds, you can max out on a decent bankroll.
    This is why I noted that adding your prop bet example is useless to this discussion.

  35. #35
    tomcowley
    tomcowley's Avatar Become A Pro!
    Join Date: 10-01-07
    Posts: 1,129
    Betpoints: 6786

    In reverse order- you didn't understand it- there are things that get stupid where you can place a max stake on a decent roll. Props is an example everybody understands, but it comes with the limit problem. Not everything has the limit problem.

    Nobody can be 100% confident that they pick 53.1% (well, that they will continue to pick 53.1% going forward). You assign 99% of the probability inside 51.5 and 57. How much do you assign to each? How about between 51.5 and 52% between 52% and 53%? Etc. If you do those numbers, you can figure it out.

123 Last
Top