Edge

**thom321** · 01-30-12, 01:14 PM

My understanding of the Kelly formula is that it is calculating the optimal bet (theoretically), which is a combination of long term return and risk of ruin, factoring in the edge AND the probability of a win. So, on your -300 bet, the higher probability of a win more than offsets the greater edge on your -140, which is why the formula tells you to bet bigger on the -300 bet.

Personally, the standard Kelly formula almost always suggest higher stakes on high probability bets than what I am comfortable with but I think the concept in one form or another has value.

**Cicima6709** · 01-30-12, 01:22 PM

Thanks.

Any recommendations for incorporating the edge/winning probability in the unit size? I was thinking that I would just bet based on the edge...are there any downsides to that? I guess i dont understand how you can have more long term success betting on the probability as opposed to the edge.

Also, is the bet size that the Kelly calculator gives you based on to win or to bet amounts?

**LT Profits** · 01-30-12, 01:24 PM

First, your model is crap because you should never be off by that much. That aside though, Kelly says to bet TO WIN your edge, so bets on underdogs are inherently smaller.

**LT Profits** · 01-30-12, 01:26 PM

Originally posted by Cicima6709

Thanks.

Any recommendations for incorporating the edge/winning probability in the unit size? I was thinking that I would just bet based on the edge...are there any downsides to that? I guess i dont understand how you can have more long term success betting on the probability as opposed to the edge.

Also, is the bet size that the Kelly calculator gives you based on to win or to bet amounts?

The Calculator gives you the BET amounts. If you work out the numbers though, you'll see that it's really to win your edge.

**Cicima6709** · 01-30-12, 01:28 PM

Originally posted by LT Profits

First, your model is crap because you should never be off by that much. That aside though, Kelly says to bet TO WIN your edge, so bets on underdogs are inherently smaller.

I have been hitting at a very similar rate for this season with my model, but thanks for the hint.

So for example if it says to bet $100 on the -165 and $50 on the +140, its saying i should do $165 to win $100 and $35 to win $50? Just kinda surprised by that, if so...doesnt seem like it makes sense to risk that much more on something that has such a smaller edge. Am i missing something here?

EDIT: Oh, nevermind, i guess i am wrong. So the calc gives you the amounts that you would bet? So to simplify, if the #s it spat out were $100 and $50, it would be $100 to win $60 and $50 to win $70 as the optimal bets according to the Kelly? Thanks for your help.

**LT Profits** · 01-30-12, 01:33 PM

Originally posted by Cicima6709

I have been hitting at a very similar rate for this season with my model, but thanks for the hint.

So for example if it says to bet $100 on the -165 and $50 on the +140, its saying i should do $165 to win $100 and $35 to win $50? Just kinda surprised by that, if so...doesnt seem like it makes sense to risk that much more on something that has such a smaller edge. Am i missing something here?

No, if Kelly spits out 100 on the -165, that means 100 to win 61. Likewise, the +140 would be 50 to win 70. Read my previous two posts, the numbers that Kelly spits out are the risk (or bet) amounts. If you work out the numbers though, you'll see that they are the risk required to win your edge.

**Cicima6709** · 01-30-12, 01:37 PM

Originally posted by LT Profits

No, if Kelly spits out 100 on the -165, that means 100 to win 61. Likewise, the +140 would be 50 to win 70. Read my previous two posts, the numbers that Kelly spits out are the risk (or bet) amounts. If you work out the numbers though, you'll see that they are the risk required to win your edge.

I am just a bit surprised that despite my edge being double on the + money, my bet is close to identical. I was under the impression that just the edge would be a decent way to predict unit size. Is there a downside to basing it just on edge?

Thanks, I appreciate your help LT.

**FreeFall** · 01-30-12, 01:40 PM

Yes to your edited question.

**LT Profits** · 01-30-12, 01:51 PM

Originally posted by Cicima6709

I am just a bit surprised that despite my edge being double on the + money, my bet is close to identical. I was under the impression that just the edge would be a decent way to predict unit size. Is there a downside to basing it just on edge?

Thanks, I appreciate your help LT.

You get what, 33.74% at -165 and 28.57% at +140? But look at what that yields to win: 33.74%/1.65 = 20.45%; 28.57% * 1.40 = 40%. So you do WIN twice as much, you win your exact edge.

**Cicima6709** · 01-30-12, 02:01 PM

Originally posted by LT Profits

You get what, 33.74% at -165 and 28.57% at +140? But look at what that yields to win: 33.74%/1.65 = 20.45%; 28.57% * 1.40 = 40%. So you do WIN twice as much, you win your exact edge.

Thanks, that makes sense.

Where did you get the 33.74 and 28.57 though? No matter what kelly multiplier i use, i get different numbers. Although its the same ratio...just curious how you got those percentages.

**LT Profits** · 01-30-12, 02:09 PM

You entered as two simultaneous events, which makes the calculation more complicated (but the theory the same). If you enter one at a time as stand alones, or if you enter ANY single event alone, the result will always be risk required to win edge at full Kelly.

**JohnGalt2341** · 02-02-12, 06:09 PM

Originally posted by Cicima6709

So in my model, i just calculated the edge i have on two bets. One of the bets was a -165 in which i thought the real value was -300, and the other was a +140, in which i thought the real value was -140. I calculated the edge to be 20.45% on the -165 and 40% on the +140 bet.

I'm confused as to how you determined your Edge. I'm sure I'm wrong about this but this is the way I see it... in the first bet if you believe the real line should be -300 that should indicate that you think you have a 75% chance of winning right? However, if you are getting the line at -165 you would only have to win 62.26% of the time to break even with that line. So wouldn't that make your edge 12.74%?

In the 2nd bet if you feel the true line is -140 that should indicate that you feel that you have a 58.33% chance of winning. If you are getting the line at +140 you would only have to win 41.67% to break even. So wouldn't that make your edge 16.66%?

Like I said... I'm sure I'm not looking at it correctly but if you or LT Profits could explain how you determine your edge I would appreciate it. My question has nothing to do with Kelly by the way.

**FourLengthsClear** · 02-02-12, 07:19 PM

Originally posted by JohnGalt2341

I'm confused as to how you determined your Edge. I'm sure I'm wrong about this but this is the way I see it... in the first bet if you believe the real line should be -300 that should indicate that you think you have a 75% chance of winning right? However, if you are getting the line at -165 you would only have to win 62.26% of the time to break even with that line. So wouldn't that make your edge 12.74%?

In the 2nd bet if you feel the true line is -140 that should indicate that you feel that you have a 58.33% chance of winning. If you are getting the line at +140 you would only have to win 41.67% to break even. So wouldn't that make your edge 16.66%?

Like I said... I'm sure I'm not looking at it correctly but if you or LT Profits could explain how you determine your edge I would appreciate it. My question has nothing to do with Kelly by the way.

75.00% divided by 62.26%
= 1.2046

Therefore +EV of 20.46%

**hutennis** · 02-05-12, 01:19 AM

Originally posted by Cicima6709

Im just trying to figure out the best way to analyze my model...

Maybe, for starters, you should make sure that you have a model in a first place?

Here is a hint for you.

A successful model must be able to make predictions which will be farther confirmed, under controlled conditions, by statistically significant experimental results.

Any backward looking only “models” are nothing more but a research, collection of data that is used to form a hypothesis.

This hypothesis must be tested by confirmation of a specific prediction by science fair experiment to make sure that your hypothesis is not wrong and your research is not negatively affected and thus rendered useless by confirmation bias, forward bias, hindsight bias, Texas sharpshooter fallacy, data fishing, misunderstanding of the law of large numbers, misusing regression to the mean, gambler’s fallacy, illusion of correlation and who knows how many more different cognitive, logical, statistical and decision-making biases and fallacies.

Any positive results achieved by using research data only (or what great many people erroneously call "my model"), can and should be safely attributed to luck.