Kelly bettors, how do you figure out win%

[durito]

but again, (for 2nite example, just takign random numbers but ill use them for the sake) i got celts -7.5 over clips. i think celts should b -10(just a number that seems good..not really wat i've come up with) so im getting celtics +2.5 better than i think ther line should b at, how do i plug this in 2 find how much i should b betting according to the kelly formula?

also, it seems like u and ganch are using two different approaches, he comes off with ML and your using spreads. i no i can use the converter to get from ML to SPREAD, would that help me when i develop my edge of +2.5? (i think thats confusing but i'll see if u get it or not)

If we assume that -10 (-110/-110) is the fair line. (and I will contend that no matter how accurate your line's prove to be, that it wont be this accurate -- ie you are not 100% correct and the market 100% wrong...assuming you are sharper than the market the real line is somewhere in between)

We can then use Ganchs 1/2 pt Calculator. Put in NBA and -10 -110/-110 as the line. Now if we put in -110 for -7.5 we get 16.53% as your edge (in fairness the 1/2 pt value calculator will get less accurate the further away from the spread you get) -- Nevertheless, that's a substantial edge.

If this were my play, I would use whatever historical data I have to see how accurate my predictions were vs. the spread which would enable me to get a more accurate prediction.

also, it seems like u and ganch are using two different approaches, he comes off with ML and your using spreads. i no i can use the converter to get from ML to SPREAD, would that help me when i develop my edge of +2.5? (i think thats confusing but i'll see if u get it or not)

I'm a bit curious myself as to why Ganch appears to mostly bet money lines. Though i think I can guess.

[Ganchrow]

but again, (for 2nite example, just takign random numbers but ill use them for the sake) i got celts -7.5 over clips. i think celts should b -10(just a number that seems good..not really wat i've come up with) so im getting celtics +2.5 better than i think ther line should b at, how do i plug this in 2 find how much i should b betting according to the kelly formula?

also, it seems like u and ganch are using two different approaches, he comes off with ML and your using spreads. i no i can use the converter to get from ML to SPREAD, would that help me when i develop my edge of +2.5? (i think thats confusing but i'll see if u get it or not)

It's just two sides of the same coin.

If you think the Celtics should be -10 then using the Half-Point calculator set to NBA Spreads, you'd enter a spread of 10 and set both fave odds and dog odds to +100. You'd then scroll the calculator up to 7.5 and enter -110 (or whatever you paid) into the "Fave Odds" box in the grid next to 7.5 and hit "Enter". This then gives you an edge of 16.53%.

One word of caution, however. The HPC is quite accurate when dealing with small deviations in spread. The larger the deviation (such as 2½ points as in your example) the less accurate it becomes, tending to overstate edge. That said unless you're betting stale lines or possibly on the open or maybe with lineup information pending it's highly unlikely that you've identified an NBA spread off by a full 2½ points.

[B1GER1C828]

im not to sure how 2 use the calculator, i guess ill hav to mess around with it for a bit, ill try some searches or something, see if i can come up with the process

[B1GER1C828]

If we assume that -10 (-110/-110) is the fair line. (and I will contend that no matter how accurate your line's prove to be, that it wont be this accurate -- ie you are not 100% correct and the market 100% wrong...assuming you are sharper than the market the real line is somewhere in between)

We can then use Ganchs 1/2 pt Calculator. Put in NBA and -10 -110/-110 as the line. Now if we put in -110 for -7.5 we get 16.53% as your edge (in fairness the 1/2 pt value calculator will get less accurate the further away from the spread you get) -- Nevertheless, that's a substantial edge.

If this were my play, I would use whatever historical data I have to see how accurate my predictions were vs. the spread which would enable me to get a more accurate prediction.




I'm a bit curious myself as to why Ganch appears to mostly bet money lines. Though i think I can guess.

well thank u vey much for explaining how to use the calculator, i was just about to do all the work of trying to figure it our myself thank you durito.

[B1GER1C828]

That said unless you're betting stale lines or possibly on the open or maybe with lineup information pending it's highly unlikely that you've identified an NBA spread off by a full 2½ points.

ya im aware of that, it was just to throw out an example and those numbers came 2 mind. thx a lot guys. i think i actually get it now. if i think line should b 8.5(still using celts -7.5) i get an edge of 8.52...so i plug this into the equations and get my betting amount, thank you very much guys! my question is(prob an irrelevant question but curiosity)

if thats my edge, is the only difference between the half-point calc and the equation of edge=probability(odds)-1 that one using ML and other uses spread?

p.s i owe yas big! probably was annoying with all my questions but you guys were amazing!

[eidolon]

I just "BET" on the line I think will win.
I only have a 2 bet system. If you want to talk about units, it would be a 1 unit play for almost all my bets, and than a 5 unit bet on like maybe 1 or 2 a month (probably only around 3-5% of my picks).

Because how I look at it: If I think I see an edge, I'm going to put my bet down, and if I see a huge edge, like wtf are thinking edge, I put down the 5 unit one.

Think about it: People always talk about a 1-5 unit betting spread, or more, but if you are a full time bettor you probably only win 55-57% of the time (not counting the 60% and 50% years); so if you are only winning 2-4% over the vig, changing your betting, will show that, if you are going to make a bet just put down the same bet everytime, because if you think you are going to win, bet the amount your bankroll will allow you.

My view on bankroll: bet 1 or 2 %. and recalculate bankroll every 3,6, or 12 months.

[donjuan]

I just "BET" on the line I think will win.
I only have a 2 bet system. If you want to talk about units, it would be a 1 unit play for almost all my bets, and than a 5 unit bet on like maybe 1 or 2 a month (probably only around 3-5% of my picks).

Because how I look at it: If I think I see an edge, I'm going to put my bet down, and if I see a huge edge, like wtf are thinking edge, I put down the 5 unit one.

Think about it: People always talk about a 1-5 unit betting spread, or more, but if you are a full time bettor you probably only win 55-57% of the time (not counting the 60% and 50% years); so if you are only winning 2-4% over the vig, changing your betting, will show that, if you are going to make a bet just put down the same bet everytime, because if you think you are going to win, bet the amount your bankroll will allow you.

My view on bankroll: bet 1 or 2 %. and recalculate bankroll every 3,6, or 12 months.

Yeah, your view is great if you hate money. If you want to learn about being optimal, do a search for Ganchrow's posts and read pretty much all of them multiple times until you understand the concepts presented.

[LT Profits]

What I absolutlely HATE about kelly is that if you have the same edge on an underdog as you have on a favorite, Kelly will always tell you to bet less on the dog. That would be very disconcerting for someone like me, as at least 95% of my Money Line bets are on underdogs. Every time I run my numbers based on all edges for all my bets being equal, I always see that I made out much better units-wise using my % of BR RISKED approach than I would have had I used Kelly, which basically would have forced me to bet those plays TO WIN the amount that I actually risked.

Conclusion: I am sure that the math behind Kelly is 100% correct, and the reason my results are what they are is probably because my edge in my underdog bets has probably been much better than the edge in my chalk bets, and that edge was probably better as the odds got higher. Thus, running Kelly at those true edges would have yielded a recommended risk amount closer to what I actually risked, putting the Kelly results and my actual results more in line.

All that said, sticking with % of BR is easier for me since it is a one-second calculaion and yield similar results to Kelly without the headache of precisely quantifying my edge. However, I'm sure my opinion would be the total opposite if I was a chalk player, so if you bet predominantly ML faves,
Kelly is the way to go.

[Ganchrow]

What I absolutlely HATE about kelly is that if you have the same edge on an underdog as you have on a favorite, Kelly will always tell you to bet less on the dog. That would be very disconcerting for someone like me, as at least 95% of my Money Line bets are on underdogs. Every time I run my numbers based on all edges for all my bets being equal, I always see that I made out much better units-wise using my % of BR RISKED approach than I would have had I used Kelly, which basically would have forced me to bet those plays TO WIN the amount that I actually risked.

Conclusion: I am sure that the math behind Kelly is 100% correct, and the reason my results are what they are is probably because my edge in my underdog bets has probably been much better than the edge in my chalk bets, and that edge was probably better as the odds got higher. Thus, running Kelly at those true edges would have yielded a recommended risk amount closer to what I actually risked, putting the Kelly results and my actual results more in line.

All that said, sticking with % of BR is easier for me since it is a one-second calculaion and yield similar results to Kelly without the headache of precisely quantifying my edge. However, I'm sure my opinion would be the total opposite if I was a chalk player, so if you bet predominantly ML faves, Kelly is the way to go.

On a per-dollar wagered basis underdogs carry more risk than favorites and big underdogs carry more risk than small underdogs. For example, assuming no vig, a $100 bet at +1000 odds is more than three times risker (10x the variance) than a $100 bet at +100 odds. Just as you expect to experience more volatile bankroll fluctuation the larger the percentage of bankroll you wager, you also expect to experience more volatile bankroll fluctuation the greater the proportion of dollars you wager at longer odds (i.e., "more positive") versus shorter odds.

All else being equal, however, there is a specific advantage to betting at longer odds versus shorter odds and that is that the longer your odds the more positively skewed your payouts become. "Positive skewness" refers to the property of long odds bets such that you risk a little to occasionally win a lot. A lottery ticket is the canonical example of a positively skewed bet. You risk a very, very small amount in order to very, very occasionally win a very, very large amount. A negatively skewed payout would be just the opposite -- risking a lot in order to frequently win a little. Going all-in pre-flop with a junk hand in deep stack no-limit hold 'em is a good example of a negatively skewed bet -- most of the time you'll collect the blinds although occasionally you'll be called and lose your stack (and less frequently you'll be called and win with the weaker hand -- this increases the skewness).

The reason that negative skew is in itself bad for growth is because it places the bettor in a situation where his bankroll could potentially suffer a large hit, thereby necessitating a reduction in bet size going forward. Any state of affairs where an advantage bettor has substantially less funds with which to wager will make it more difficult to continue to grow bankroll, which is what a Kelly bettor tries to avoid. Kelly recognizes this and will attempt (again, all else being equal) to steer a bettor as much as possible away from negative skew and towards positive skew. That said, the acquisition of expected return and the avoidance of variance nevertheless factor in more heavily for a Kelly bettor than skew.^{<a href="#foot1">1</a>}

State Lotteries are very -EV but many people still play them because the idea of risking a little to occasionally win a lot is extremely enticing. Conversely, you don't frequently see people wagering $1 million dollars every week for a 99.9999% probability of winning $1, even though at 0 EV it's a much better bet expectation-wise than a state lottery.

These same preferences also frequently drive the recreational sports bettor. He notes that by betting the large money line underdog he'll occasionally have a nice surprise. He'll tend to remember these large wins and will tend to forget about the many seemingly inconsequential losses he encounters. He'll also note that when he bets large favorites he'll face many small seemingly inconsequential wins and will occasionally suffer very large, very heart wrenching losses. He'll remember these big losses and will vow to avoid large favorites going forward. Perversely, he'll consider this careful risk management when in reality all he's doing is taking on more skew at the cost of greater variance.

In fact, in a somewhat roundabout way the greater variance of underdogs might actually be an advantage for the recreational sports bettor -- especially those attempting to fool themselves into believing theirs bet are +EV. Because of the lower risk inherent in betting favorites it's much less likely for a -EV bettor betting favorites to show a profit over any given stretch of time than a -EV bettor betting underdogs.

Consider, for example, two bettors each playing at 4.545% vig over 1,000 trials, one betting only +1000 dogs and the other betting only -1000 faves. If each player were to risk 1 unit on every wager then both players have the same expectation, namely -45.45 units. However, the player betting at -1000 has only a 0.0022% probability of showing a profit over 1,000 wagers while the player betting at +1000 has a 33.3257% of showing profit over the same stretch. ^{<a href="#foot2">2</a>} (The +1000 player also has a 6.3989% probability of winning 100 or more units. The -1000 player? A 5.3627×10^-56% probability.)

So we have two players, each with the same expectation, but the first with a de minimis probability of showing a profit and the second, no more or less skilled as a bettor than the first, with nearly a 1/3 probability of facing an outcome that permits him the self-delusion of believing he has an advantage as a sports bettor even when none exists. This is yet another reason why recreational bettors so frequently prefer to bet the large underdog.

Now I'm certainly not trying to claim that it's impossible to be an advantage player and bet underdogs. Far from it. The shorter the odds at which one bets the lower one's maximum possible theoretical edge. As such while it may very well be possible to find say a 10%+ edge betting at +1000 odds, it's mathematically impossible to find higher than a 10% edge when betting at -1000 odds. As such, while each unit of edge is more valuable at shorter odds, it's also harder to come by. If you have a choice between pursuing a strategy that yields 5% edge on -1000 bets versus one that yields 5% edge on +1000 bets the choice should be very clear (the former).

But it's not usually so cut and dry.

More realistically you'll find yourself choosing between strategies that might yield ~1% edges at -1000 and ~10%-15% edges at +1000. With choices like that it's not so clear.

Risk management is crucial to long-term sports betting success. A key component of this is recognizing the greater risk to one's bankroll on a per-dollar wagered basis that one need accept when betting large underdogs. Furthermore, due to this risk it's much easier to convince one's self that an edge has been secured when in fact none does exists. This makes it more difficult to be confident of the long term prospects of an underdog strategy than of a favorite strategy. On the other hand, because one faces potentially greater losses when betting at shorter odds, incorrectly deducing an edge can be considerably more dangerous with favorites than with underdogs.^{<a href="#foot3">3</a>}

Whatever the segment along the odds contiuum on which the advantage player concentrates, it's crucial that he recognize and account for these fundamental structural odds issues when determining his risk management strategy.

<hr>
<a name="foot1">¹</a>Taking it a step further is the concept of "kurtosis", which may be loosely thought of as the "riskiness of risk". The greater the extent to which risk may attributed to very extreme occurrences, the higher the kurtosis. A bankroll growth maximizing bettor will (once again, all else being equal) seek to minimize kurtosis just as he seeks to minimize variance and maximize skew and expectation. This effect will tend to make a bettor favor odds close to even (slightly higher as edge increases).

<a name="foot2">²</a>It's for this very same reason that when evaluating the records of a handicapper it's important not only to take into account the range of bet sizes, but also the range of payout odds at which bets took place. For example a handicapper up 24.5 units over 250 1-unit bets (+9.8%) at -200 has a more statistically significant record (p-value=1.56%) than a handicapper up 60 units over 250 1-unit bets (+24%) at +400 (p-value=3.70%).

<a name="foot3">³</a>In other words, while it's easier to overestimate one's edge betting at longer odds, the potential hazard from doing so is greater at shorter odds.

[B1GER1C828]

what i realized just now which for w/e reason i didnt b4, is you can use win prob to calculate kelly...

is win% just the implied prob divided by the two implied probabilities added together?

[Ganchrow]

is win% just the implied prob divided by the two implied probabilities added together?

This corresponds to the win probability implied by the market assuming that the same % vig is paid by all sides of the market (which may or may not be a valid assumption).

Note that that the edge implied by the combination of the line and the market probability will always be negative.

[B1GER1C828]

ok with that being said...how do u find win% without doing this. because when i was doing this, your right i kept getting a negative edge and it was really questioning me

[donjuan]

ok with that being said...how do u find win% without doing this. because when i was doing this, your right i kept getting a negative edge and it was really questioning me

As I posted earlier, in the thread, you would find lines off market value if you wanted to get an edge by doing this.

[3put]

In other words, while it's easier to overestimate one's edge betting at longer odds, the potential hazard from doing so is greater at shorter odds.

Ganchrow, thank you for this excellent post.

I have a handicapping model focusing on favorites in the range -300 to +100. The edge is normally 1% - 8% for my bets.

In the beginning I used full Kelly, but I soon realized that I was not comfortable with the risk-amount on very favorable bets. Also the quoted hazard in overestimating the edge, led me to try Kelly with a factor 0.33.

That has worked out very fine, but still I'm a chicken on the very best of my bets.
Occasionally I find a line with an edge of 15% but ironically I cut down on the $$ risked, arguing to myself that I still have a very favorable situation.

I would very much appreciate a post from you, discussing issues like:

- Full Kelly versus fractional Kelly
- More on consequences of overestimating the edge.
- Size of bankroll versus Kelly Factor.

[Dark Horse]

I would differentiate between winning percentages per angle (based on precise records), and long term fluctuations/streaks of up to one year.

Based on my observation it is entirely possible that a near 60% bettor over a decade can still experience a year of 40% winners. Try throwing that at Kelly.

Mathematical explanations as to the likelihood (or great unlikelihood) of that happening are flawed, because real life can't be isolated into a test tube of math. Most importantly, (present day) math does not account for the magnetic fluctuations of Time.

[durito]

I would differentiate between winning percentages per angle (based on precise records), and long term fluctuations/streaks of up to one year.

Based on my observation it is entirely possible that a near 60% bettor over a decade can still experience a year of 40% winners. Try throwing that at Kelly.

Mathematical explanations as to the likelihood (or great unlikelihood) of that happening are flawed, because real life can't be isolated into a test tube of math. Most importantly, (present day) math does not account for the magnetic fluctuations of Time.

I'll repeat the question others have asked: What money management system then will hold up better during the year at 40%?

[B1GER1C828]

Ganchrow, thank you for this excellent post.

I have a handicapping model focusing on favorites in the range -300 to +100. The edge is normally 1% - 8% for my bets.

In the beginning I used full Kelly, but I soon realized that I was not comfortable with the risk-amount on very favorable bets. Also the quoted hazard in overestimating the edge, led me to try Kelly with a factor 0.33.

That has worked out very fine, but still I'm a chicken on the very best of my bets.
Occasionally I find a line with an edge of 15% but ironically I cut down on the $$ risked, arguing to myself that I still have a very favorable situation.

I would very much appreciate a post from you, discussing issues like:

- Full Kelly versus fractional Kelly
- More on consequences of overestimating the edge.
- Size of bankroll versus Kelly Factor.

what type of model are you using? and i think ganch explains the difference in one of his posts already, but i dont remember which(which isnt helpful at all..lol)

[donjuan]

I'll repeat the question others have asked: What money management system then will hold up better during the year at 40%?

You stop betting because it's the 40% year LDO.

[Dark Horse]

I'll repeat the question others have asked: What money management system then will hold up better during the year at 40%?

Something equivalent to management in BJ card counting. A drastic reduction in bet size, in inverse form of Kelly. (or perhaps better, no bets at all).

[donjuan]

Something equivalent to management in BJ card counting. A drastic reduction in bet size, in inverse form of Kelly. (or perhaps better, no bets at all).

That is ridiculous. "Hotness" and "coldness" quite simply do not exist.

[Dark Horse]

If you believe that ignorance of a phenomenon means it must not exist, then you are absolutely right.

[donjuan]

If you believe that ignorance of a phenomenon means it must not exist, then you are absolutely right.

It's not a phenomenon but rather a trick played on the mind.

Let's say we have a coin that has been been flipped 50,000 times and heads has come up exactly 25,000 times. However, the last 12 flips have been tails. Would you give me +105 on heads for the next flip?

[Dark Horse]

Never mind. I'm not inclined to enlighten you on the subject. Believe what you want.

[TLD]

A drastic reduction in bet size, in inverse form of Kelly. (or perhaps better, no bets at all).

If you think reducing your bets as your bankroll shrinks during a losing streak constitutes the inverse of Kelly, then again you simply are mistaken.

It sounds like at most you differ on the degree of reduction. But you have no reason for doing so. In fact if anything, your “or perhaps better, no bets at all” indicates you’ve fallen victim to the illogic that donjuan was lampooning with his remark that one is best off just not betting during the “40% year.”

Reference to the “magnetic fluctuations of Time” in this context is nothing more than obfuscation through use of a nonsense phrase.

Not that it matters to persuade you to change your beliefs; the reason for my remarks is more any readers new to the Kelly system who might be misled by your claim that Kelly somehow is flawed in practice because “it doesn’t account for streaks.” Since you’ve sidestepped all opportunities to explicate or defend this claim, it carries exactly the same weight as if I were to arbitrarily assert “Kelly is flawed in practice because with Kelly you end up betting too many home teams.”

(Also, for what it’s worth, a blackjack card counter does not adjust his bet size in response to “streaks” but to 1) the distribution of remaining cards of the various denominations, and 2) bankroll size (a la Kelly).)

[MrX]

TLD,

Thanks for saving me a post and saying it better than I would have.

[Dark Horse]

Reference to the “magnetic fluctuations of Time” in this context is nothing more than obfuscation through use of a nonsense phrase.

Thank you for proving my point that it is far better to not even touch upon a subject that is hidden from the masses.

You may now have a few laughs on my behalf.

And please do continue with your test tube version of reality, and with full Kelly betting.

[donjuan]

TLD hit the nail on the head when he said this:

Reference to the “magnetic fluctuations of Time” in this context is nothing more than obfuscation through use of a nonsense phrase.

[failheart]

Ways to Forecast a Line

This is the reality of the line. The line maker has Chicago giving six points to Cleveland. He has, in his opinion; brought that game to a perceived standoff the points make it so.

Suppose Kansas City was playing at Chicago. And each team was sporting a win-loss record 4-6, each was locked into a three-game losing streak and both teams had scored less than 17 points in their past four games. Their near identical records make it a hard choice. But you still figure you have an idea of what the line will be, and make Chicago -3.

The three-point edge is strictly based on home-field with a nod towards Chicago kicking game. Then you compare your Chicago -3 with the official line. You could see Kansas City -1 with a dog getting the nod on the road. That's a four point spread.

You can either pass on the game. Or you can take Chicago getting +1 since you already had them at -3. The decision is yours.

This is all I do. Figuring out my forecast line is me just looking back to the previous times the teams have met and see how they performed. Also checking out the teams current ATS, O/U, Power vs. Dog and what their win/loss record is being the Favorite and Dog at home or on the road.

[B1GER1C828]

wow i had alot of posts in this thread...3 years ago wowsers. rereading them, i was one dum fuk.

[Melloweitsj]

You figured out your edge yet?