[pico]

from Ganch's article:

**broken link**

Given a bankroll of $10,000 and an edge of 5%, then on a bet at odds of +100 one should wager 5% / (2-1) = 5% of bankroll, or $500.
Given a bankroll of $10,000 and a win probability of 55%, then on a bet at odds of -110, one should wager $10,000 * (55% * 1.909091 - 1) / (1.909091-1) = 5.5% of bankroll, or $550.
Given a bankroll of $10,000 and a win probability of 25% then on a bet at odds of +350, one should wager $10,000 * (25% * 4.5 - 1) / (4.5-1) ≈ 3.57% of bankroll, or about $357.
Given a bankroll of $10,000 and a win probability of 70% then on a bet at odds of -250, one should not wager anything because edge = win prob*odds = 70%*1.4 = 98% < 1.

how do you figure out that you have a 5% edge on certain games? why not 10% or 30%. i think it is pretty impossible to figure out what is your edge is.

[donjuan]

how do you figure out that you have a 5% edge on certain games? why not 10% or 30%. i think it is pretty impossible to figure out what is your edge is.

If you can't somewhat reasonably quantify your edge, what makes you think you have any edge in the first place?

[LT Profits]

I guess the best way to explain it (and I'm no donjuan or Ganch) would be to calculate what you think the spread and/or money line and/or total on a game should be, and compare that to the actual line. The percent off should theoretically be your edge.

I know that's the simplified version, but just wanted to paint a picture of what it entails.

[Ganchrow]

calculate what you think the spread and/or money line and/or total on a game should be, and compare that to the actual line. The percent off should theoretically be your edge.

I'll just point out that while the percentage by which an actual line is off from your forecast line might be a good indicator of edge it isn't your actual quantitative "edge" as far as Kelly is concerned. What you really want to do is take your line forecast and then use that to deduce a win probability for the offered line. (You might find a tool such as the Half-Point Calculator useful in that regard.) Using win probability and payout odds you then calculate your edge using the equation:

edge = odds * probability - 1, where odds are expressed in decimal format

For more on how to calculate edge see An introduction to expectations and theoretical hold.

From a Bayesian perspective, edge as calculated above merely represents a prior expectation, which should then be tempered by whatever new information (such as the market line) becomes available.

If you can't somewhat reasonably quantify your edge, what makes you think you have any edge in the first place?

A fine question.

[B1GER1C828]

does betting like this help u win alot more money in a longrun? my betting scheme is finding teams that look like they can cover...but doing this increase ur profits/win%?

[curious]

from Ganch's article:


how do you figure out that you have a 5% edge on certain games? why not 10% or 30%. i think it is pretty impossible to figure out what is your edge is.

Don't let these guys fool you. It is IMPOSSIBLE to determine the actual edge in a sports contest. The best you can do is guess. Some people are better at guessing than others. There is no way that anyone can state with anything approaching mathematical certitude that "With the current line Team A has a n X% chance of covering the spread against Team B". Having said that, the only you can "guess good" is to have some sort of prediction model that is reliable to some degree better than picking teams at random. There are too many variables in a sports contest to know what your true edge is.

[20Four7]

There's nothing wrong with taking winnable dogs vs the spread as you have 2 ways to win, if they win outright you win and if they lose by less than the number you win. Again it comes down to finding those dogs who can win. By taking the fav, you HAVE to win by more than the spread.

[pico]

If you can't somewhat reasonably quantify your edge, what makes you think you have any edge in the first place?

i can't. like most bettors, by looking at stats and from using my intuition, i can only say, i think the dog can cover this game or the fav can cover the spread. if i can really say, hey i think the line is wrong, the real line should be at -11 not -6, then i would probably quit my day job and gamble full time.

[pico]

I guess the best way to explain it (and I'm no donjuan or Ganch) would be to calculate what you think the spread and/or money line and/or total on a game should be, and compare that to the actual line. The percent off should theoretically be your edge.

I know that's the simplified version, but just wanted to paint a picture of what it entails.

thanks. i got the picture. i guess if you want to go pro, you have to think like those wall street analysts...i think the the stock price next year is oging to be 60, so i am going to long 5 lots.

[curious]

thanks. i got the picture. i guess if you want to go pro, you have to think like those wall street analysts...i think the the stock price next year is oging to be 60, so i am going to long 5 lots.

I have spent hundreds of hours writing software to search for patterns so that I could get a handle on figuring out the edge. I have several friends who are smarter than me at this, we spend a lot of time on the phone and via email examining possible edges. This is harder than anything I've ever been involved in.

[pico]

I have spent hundreds of hours writing software to search for patterns so that I could get a handle on figuring out the edge. I have several friends who are smarter than me at this, we spend a lot of time on the phone and via email examining possible edges. This is harder than anything I've ever been involved in.

well, the lines are set by LVSC, and those people have shitload of data and time to make up the lines. by saying you have an edge, you're pretty much saying those guys' line are wrong. and if you can figure out the edge, you're saying, hey not only you guys are wrong, i can figure out how much % you're wrong by. that is freaking hard to do, imo.

i think i'll leave the hardcore capping to people like ganch.

[B1GER1C828]

ganch must be friggen loaded by now then
i just dnt see how u can just do this consistently and make profits, if ur not good at finding edge, u basically can b throwing alot of money away...

[20Four7]

The only sport I seriously cap is NFL football. When I cap football I analyze the game and come up with a "projected" score. For the superbowl, my "projected" score was NE 24 Giants 21. With this I say the line should have been NE -3 and the total 45. I keep stats to know that when my line is 9.5 away from the book makers line, I have a large edge. I know what % I win these game (same with the totals). These are the games I want to be betting on. I usually don't consider anything 6 or less away from the book, as even when you believe you have an edge, an injury, fumble, interception can turn everything around very quickly.

I believe if you can't set a line on the game you can't effectively cap the game.

I agree with curious, effectively determining your edge is probably the hardest part of being a winning gambler. And if using Kelly I always ALWAYS underestimate my edge.

[Arilou]

There are two important things to keep in mind when you want to figure out your edge. First, you use your model or your capping or your results, or a combination of all of them, to figure out what the line "should" have been. 20Four7 used the example of -3 in a -12.5 game. Once you have that, you can use your other data to figure out that means you're 77% to win.

The second thing to keep in mind is that the previous calculation was wrong. That was best case, this is reality. You're not going to have the edge you think you have, except for when you have tons of actual results to base it on, so cut your edge by at least half. Remember, it's not a tragedy to underestimate your edge, it's far worse to overestimate.

[Ganchrow]

The only sport I seriously cap is NFL football. When I cap football I analyze the game and come up with a "projected" score. For the superbowl, my "projected" score was NE 24 Giants 21. With this I say the line should have been NE -3 and the total 45.

I'll just point out that (especially in football) if you determine your "projected line" from the final score you consider most likely then there's a good chance you're incorrectly calculating your edge.

NFL MOVs are dominated by "attractors" such as the 3, 7, and 10. In reality this means that for games predicted to be close the most likely outcome will be favorite winning by exactly 3 irrespective of the actual spread. Even with spreads as high as 10, for example, many score distribution models still judge a favorite MOV of 3 as the most likely game outcome.

So when looking at a projected final score, in order to determine edge one needs to be sure to consider that projection in light of the peculiar scoring distribution inherent in NFL football.

A game line corresponds to the median MOV outcome, a "projection" (in common parlance) references the modal score outcome, while an edge represents the mean dollar outcome of a wager. While all are related concepts they are nevertheless distinct and should be considered independently of one another.

[Ganchrow]

The second thing to keep in mind is that the previous calculation was wrong. That was best case, this is reality.

If you're referring to the edge calculation posted above the point it is that it's indeed correct conditioned on the estimate of win probability. If your win probability estimate is flawed then so too will the resultant edge estimate. "Garbage in, garbage out."

You're not going to have the edge you think you have, except for when you have tons of actual results to base it on

This corresponds to a frequentist methodology for edge/win probability estimation. A Bayesian, however, would point out that one can speak of the subjective likelihood of an event occurring even without ever actually having witnessed the event in question This might allow, for example, a bettor using Bayesian inference to determine the likelihood of a hypothetical 28½-point NFL home favorite winning by fewer than 3½ points even though not a single NFL team has ever been favored by 28½ points (let alone a "ton" of such actual results existing).

This is the key to successful estimation of one's edge and is often difficult for those too accustomed to frequentist methodology to fully accept. The half-point calculator, for example, is a tool I built (more or less) on frequentist estimation. This means it's fairly good at estimating likelihoods surrounding events it's already seen many times in the past (NFL spreads around 3, for example) but nearly useless at estimating likelihood surrounding events it's seldom seen (NFL spreads of 15+, for example).

But just because it's difficult (or impossible) for a frequentist estimation procedure to succeed when lacking a complete and detailed data set does not mean that a process of Bayesian inference would be similarly doomed to failure. What's necessary is the efficient usage of as much available information as possible in order to influence one's prior beliefs (which certainly could be grounded in frequentist sampling).

This is how many e-mail SPAM filters work (in particular those utilizing what's known as "naïve Bayesian classification"). Even though a filter may never have seen a particular flavor of SPAM it may still make a judgment as to the subjective likelihood of that piece of mail being SPAM ("subjective" because if, for example, I work in a Viagra factory what's likely to be SPAM for you might be a whole lot less likely to be SPAM for me.)

[curious]

If you're referring to the edge calculation posted above the point it is that it's indeed correct conditioned on the estimate of win probability. If your win probability estimate is flawed then so too will the resultant edge estimate. Garbage in, garbage out.

This corresponds to a frequentist methodology for edge/win probability estimation. A Bayesian, however, would point out that one can speak of the subjective likelihood of an event occurring even without ever actually having witnessed the event in question This might allow, for example, a bettor using Bayesian inference to determine the likelihood of a hypothetical 28½-point NFL home favorite winning by fewer than 3½ points even though not a single NFL team has ever been favored by 28½ points (let alone a "ton" of such actual results existing).

This is the key to successful estimation of one's edge and is often difficult for those too accustomed to frequentist methodology to fully accept. The half-point calculator, for example, is a tool I built (more or less) on frequentist estimation. This means it's fairly good at estimating likelihoods surrounding events it's already seen many times in the past (NFL spreads around 3, for example) but nearly useless at estimating likelihood surrounding events it's seldom seen (NFL spreads of 15+, for example).

But just because it's difficult (or impossible) for a frequentist estimation procedure to succeed when lacking a complete and detailed data set does not mean that a process of Bayesian inference would be similarly doomed to failure. What's necessary is the efficient usage of as much available information as possible in order to influence one's prior beliefs (which certainly could be grounded in frequentist sampling).

This is how many e-mail SPAM filters work (in particular those utilizing what's known as "naïve Bayesian classification"). Even though a filter may never have seen a particular flavor of SPAM it may still make a judgment as to the subjective likelihood of that piece of mail being SPAM ("subjective" because if, for example, I work in a Viagra factory what's likely to be SPAM for you might be a whole lot less likely to be SPAM for me.)

I did something similar with NCAAF with underdogs over +250. I did not have any previous data to go by. I would guess the "average" ML of these big dogs was +400. I guessed (correctly) that these big dogs do not lose as often as this line would predict. And by using a strategy of betting 2X on the spread and X on the ML, lines over +250 would be profitable to bet on. I was right. I did also handicap these teams to determine if they had any chance of winning. I didn't follow them blindly. Of course, then SBG stole my winnings and BJO doesn't take ML over about +180 so I had to abandon the experiment.

[20Four7]

I'll just point out that (especially in football) if you determine your "projected line" from the final score you consider most likely then there's a good chance you're incorrectly calculating your edge.

NFL MOVs are dominated by "attractors" such as the 3, 7, and 10. In reality this means that for games predicted to be close the most likely outcome will be favorite winning by exactly 3 irrespective of the actual spread. Even with spreads as high as 10, for example, many score distribution models still judge a favorite MOV of 3 as the most likely game outcome.

So when looking at a projected final score, in order to determine edge one needs to be sure to consider that projection in light of the peculiar scoring distribution inherent in NFL football.

I realize the frequency of such scoring plays in NFL football and realize a projected score of 8 to 2 for example to be an extremely un likely event while a projected score of 9 to 3 is certainly far more likely although not necessarly a realistic score for a game either. I may not be exactly quantifying my edge, I do know how my data has performed over the years and know what my win rate is for various "projected" outcomes. I have always believed if you are going to make an error to err on the conservative side.

While I don't have my figures here at work if I recollect 3 is the biggest MOV followed by 7, 10, 6 and 4. I would also say just under 50% of all NFL games are decided by 7 or less points which makes the 3 and 7 huge since they are the top 2 in MOV in NFL. While I certainly don't have your inherent math skills, I am not afraid of numbers and certainly am willing to work with them to try and quantify any edge I have.

As always your posts are extremely helpful and certainly cause me to look into how I calculate edges etc more closely.

[donjuan]

The only sport I seriously cap is NFL football. When I cap football I analyze the game and come up with a "projected" score. For the superbowl, my "projected" score was NE 24 Giants 21. With this I say the line should have been NE -3 and the total 45. I keep stats to know that when my line is 9.5 away from the book makers line, I have a large edge. I know what % I win these game (same with the totals). These are the games I want to be betting on. I usually don't consider anything 6 or less away from the book, as even when you believe you have an edge, an injury, fumble, interception can turn everything around very quickly.

Edges in sports betting are almost never this big. If you think closing lines are frequently off by 6+ points in the NFL you need to get your head checked.

[20Four7]

I never routinely said lines were off by 6 or more. I said I knew I had a large edge when I found they were. I had weeks this year when there were no bets or "my lines" were close to the books lines. Sports betting is about making the most of your opportunities. As a general rule the books do a very good job with the numbers, that being said the way to make money is to take advantage when you feel you have the advantage. Superbowl Sunday was one of those days when I felt I could take advantage.

The books wouldn't be in business long if their lines were routinely off by large amounts. Remember the purpose of the line is to drive action both ways. It's not the actual number the book believes the game will be decided by. Certain teams can have inflated numbers for any number of reasons. Certainly there was a time when the yankees were favoured by more than they should have been, and the same goes for the cowboys or the 49er's since the bigger number is needed to "force" action on their opponent.

Most gambling edges that are in the 1 to 2% range is big enough , and any edge larger you have to begin to question how you arrived at that. A poker player with a 2% edge is way above the competition at the table.

[B1GER1C828]

i just want 2 take ganch's mind and put it on mine....

but i must say, these last like 5-6 posts have cleared it up alot...do i totally understand it? no, but atleast im way more ahead then when i was.

ganch can u give us a good example of how u would find ur edge? like use an example of a basketball game tonight or something we can get a base off of? ive looked at ur kelly thread and ive tried to look at it at peieces which has helped, only thing im short on is finding the edge, so hopefully u can just fill that last bit in.

thx alot ganch

[MrX]

Here's one way you might have a good handle on your edge:

Say you have an idea for a condition that may lead to a positive betting opportunity. Better yet, say you have multiple ideas. You gather lots of historical statistics and create a database. Each record contains one game, the results of each indicator (taking great care that the indicators are independent of each other), the betting line, and the results. Perform a regression analysis. Now you will have a formula describing your edge, if any, in terms of your indicators. You test this formula against a large sample of past results that were not included in the original regression. If they are in line, you now might have a pretty good estimate of your edge going forward.

[Ganchrow]

i just want 2 take ganch's mind and put it on mine....

but i must say, these last like 5-6 posts have cleared it up alot...do i totally understand it? no, but atleast im way more ahead then when i was.

ganch can u give us a good example of how u would find ur edge? like use an example of a basketball game tonight or something we can get a base off of? ive looked at ur kelly thread and ive tried to look at it at peieces which has helped, only thing im short on is finding the edge, so hopefully u can just fill that last bit in.

thx alot ganch

MrX's example above is on point. Here's a simple example using stale lines:

Let's say your local bookie rarely updates his odds after hanging them each morning. This morning he hung the Delaware NCAA Basketball money line at (rotation# 802) and is still offering that now.

Looking at SBR Odds you find that the current prevailing market price for the game is now Wilmington /Delaware .

Q: Based on the current market what would be your edge were you to take Delaware at with your local?

A: The current market implies a zero-vig win probability of about 59.330% for Delaware (if you don't understand how to calculate that refer to An introduction to expectations and theoretical hold) while the US price of +125 offered by your bookie implies decimal odds of . Hence, your edge on the Delaware -125 bet would be about 59.33% × 1.8 - 1 ≈ 6.794%, and the full Kelly stake (assuming no hedge placed on WIlmington) would be about 6.794% / 0.8 ≈ 8.493% of bankroll. (If you were to hedge then the full-Kelly stake on each leg of the arbitrage would be equal to the probability of that leg winning. Hence, you'd bet 59.330% of bankroll on Delaware at -125 and 40.670% of bankroll on Wilmington at +140.)

[RageWizard]

What is Ganch got like a PHD in math or something. I could follow this math in this post, but some of the other stuff in other threads is beyond me, and I've got a Mech. Engineering degree from Lehigh Univ.

[B1GER1C828]

Your numbers are awful. I hope you aren't betting large.

no need for that for that in this thread, GANCH i tink that example finally knocked it into my head... using an nba example( just cuz im more comfortable with it) 2nite lakers are -165 and hawks are +145.

implied -165= 62.26
implied +145= 40.82

62.26/(62.26+40.82)= 60.40 %
40.82/(62.26+40.82)= 39.60%

so using those numbers...

60.40(odds in decimal of -165)= 60.40(1.6061)= -2.99%..how can u have negative?

if i was 2 bet on +145, i get 39.60(2.45)-1= -2.98%

where did i mess up/whats the deal?

[donjuan]

where did i mess up/whats the deal?

You assumed an efficient market and bet an efficient line. You want to find an off-market number like +165 or -145 (doesn't need to be arb-able to be profitable though).

[B1GER1C828]

You assumed an efficient market and bet an efficient line. You want to find an off-market number like +165 or -145 (doesn't need to be arb-able to be profitable though).

dont think i exactly know what u mean.

[donjuan]

Think about Ganchrow's post in more depth and get back to me. I'm here to help, but not to hold your hand.

[B1GER1C828]

Think about Ganchrow's post in more depth and get back to me. I'm here to help, but not to hold your hand.

...id prefer u didnt help

[B1GER1C828]

im not a math genius and this has been the most helpful thread to me so far really don't need u to come and be doing that

[pico]

MrX's example above is on point. Here's a simple example using stale lines:

Let's say your local bookie rarely updates his odds after hanging them each morning. This morning he hung the Delaware NCAA Basketball money line at (rotation# 802) and is still offering that now.

Looking at SBR Odds you find that the current prevailing market price for the game is now Wilmington /Delaware .

Q: Based on the current market what would be your edge were you to take Delaware at with your local?

A: The current market implies a zero-vig win probability of about 59.330% for Delaware (if you don't understand how to calculate that refer to An introduction to expectations and theoretical hold) while the US price of +125 offered by your bookie implies decimal odds of . Hence, your edge on the Delaware -125 bet would be about 59.33% × 1.8 - 1 ≈ 6.794%, and the full Kelly stake (assuming no hedge placed on WIlmington) would be about 6.794% / 0.8 ≈ 8.493% of bankroll. (If you were to hedge then the full-Kelly stake on each leg of the arbitrage would be equal to the probability of that leg winning. Hence, you'd bet 59.330% of bankroll on Delaware at -125 and 40.670% of bankroll on Wilmington at +140.)

Ganch, is this the reason some books will ban or limit steam players? if you believe the market is moving in the right direction, then hit the slow book hard...double pop that ****er. or full kelly that ****er, if you have a small bankroll.

[TLD]

I think you’re a little confused. The tools don’t provide an edge; they’re more to measure an already existing edge you discovered by other means.

So you can’t take a game that’s +145/-165 and say “Now I’ll use Ganchrow’s tools to tell me which side I should bet because it’s the side with an edge.”

No, you would use them more for a situation where, say, you handicapped a game and determined that the Knicks should be a 5 point favorite over the Celtics. Then you shop lines and see that you can get as much as +7 on the Celtics. So you want to know, “Is the difference between what I think the line should be (+5) and what it is (+7) big enough to make it profitably bettable?” “What amount of vig should I be willing to pay at +7 and still have an edge?” “-105? -107? -110?” “And if I do have an edge, what is the precise size of the edge?” (which is useful in deciding bet size). And you would use Ganchrow’s tools to address those questions.

Or let’s say you don’t handicap, but you’re of the opinion that the market consensus is typically accurate on the NBA, and you want to use that information to your benefit. You might find that almost all books have that game at 5.5 or 6, so the market consensus is right in there, but you do have this one oddball book where you can get +7. And again you want to know if you should bet the +7 when the line “should be” +5.5 or +6, so you use Ganchrow’s tools.

But you’re skipping steps if you’re trying to somehow use the tools alone to discover an edge. You have to have something you’ve already come up with through independent means to then compare with the line using the tools.

[B1GER1C828]

I think you’re a little confused. The tools don’t provide an edge; they’re more to measure an already existing edge you discovered by other means.

So you can’t take a game that’s +145/-165 and say “Now I’ll use Ganchrow’s tools to tell me which side I should bet because it’s the side with an edge.”

No, you would use them more for a situation where, say, you handicapped a game and determined that the Knicks should be a 5 point favorite over the Celtics. Then you shop lines and see that you can get as much as +7 on the Celtics. So you want to know, “Is the difference between what I think the line should be (+5) and what it is (+7) big enough to make it profitably bettable?” “What amount of vig should I be willing to pay at +7 and still have an edge?” “-105? -107? -110?” “And if I do have an edge, what is the precise size of the edge?” (which is useful in deciding bet size). And you would use Ganchrow’s tools to address those questions.

Or let’s say you don’t handicap, but you’re of the opinion that the market consensus is typically accurate on the NBA, and you want to use that information to your benefit. You might find that almost all books have that game at 5.5 or 6, so the market consensus is right in there, but you do have this one oddball book where you can get +7. And again you want to know if you should bet the +7 when the line “should be” +5.5 or +6, so you use Ganchrow’s tools.

But you’re skipping steps if you’re trying to somehow use the tools alone to discover an edge. You have to have something you’ve already come up with through independent means to then compare with the line using the tools.

hmmmm i think i no what ur saying. so u dnt calculate to find edge with the equations, its coming off ur knowledge(sorta) to figure out what you think the line should b and calculate from there. i got u thx alot

[B1GER1C828]

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)

[TLD]

I believe one of the tools is a half-point calculator. So what you could do is use that to see what –10, +100 (which you’ve deemed the correct line to be) equates to at –7.5. I’m too lazy to look it up, but let’s say it equates to –7.5, -130. So what that means is if your initial projection of 10 was accurate, then a wager of –7.5, -130 should break even.

So you shop and find that the best line available to you is –7.5, -105. So now your question is: “When I can bet something that should be –130 and I only have to pay –105, how much of an edge is that?” And you use Ganchrow’s tools to calculate that, and that gives you what you need to plug into the Kelly formula to determine bet size.

But I’ll leave it to Ganchrow or whomever to walk you through it in more detail if necessary, as I don’t have the tools in front of me and I’m just trying to summarize it off the top of my head.