Using Standard Deviation For Totals

[HonestD]

Hi everyone.

Just wanted to share some interesting stuff I've been working on real hard the past few days.

It's always been a point of mine to use SD in sports betting. I firmly believe it can be a viable method if your foundation is accurate.

Basically what I've been doing is generating a mean (average) total and then figuring out the SD. For those who aren't math junkies, in a nut shell its how far up or down values will fall away from the average. This creates a window. For example if you have a total of 60 and your SD is 5, that's 60-5 and 60+5. So you've got a window of 55 all the way to 65.

I then look at the line and see where the total falls in relation to my window. I'm looking for games heavily weighted to one side of the window.

For example, if my window is 55-65 and the line is 57, I'm taking the over. It's basically a -2 on the bottom and +8 on the top. You can call that a 2:8 and divide 2 by 8 for 25%, which leaves a 75% margin on your window at the top. Your bet will always be going towards the bigger number. You want low %s.

If a window is 55-65 and the line is 60, that's basically a 5:5 or 100% which is dead in the middle and is a toss up.

Things can change the more data I acquire, but as of right now I'm only looking for games with a 5 or less SD. The lower the SD the closer your values are to each other. I'm also looking for no less than a 66% margin on the other side.

I've attached part of a screen shot of last weeks games just to get an idea of what my spreadsheet looks like. It's sorted by SD value from lowest to highest. Bottom and Top are how far the end of the window is away from the line. We are looking for a low ratio between top and bottom as well as the line being close to the low or high.

Any questions feel free to comment, I may not have explained it thoroughly enough but I did my best.

 

[B.A.R.]

Interesting...

 

[M.O.T]

I think it’s a solid approach, but at the core it seems you’ll need to be confident in whatever C,B,S, and A are in your example....if I’m understanding it correctly.

 

[booksbestfriend]

Would teases come into play with this?

 

[HonestD]

I think it’s a solid approach, but at the core it seems you’ll need to be confident in whatever C,B,S, and A are in your example....if I’m understanding it correctly.

Precisely. I run those in a "consensus" system where I'm playing only games that they all agree on. Only 2-3 of those games came up on the SD system as far as top plays that I would risk money on.

This is all only off one week and I will keep everything updated for anyone interested,

 

[HonestD]

Would teases come into play with this?

It could. Out of the 14 losses the system had last week, a teaser would have saved 8 of those losses.

 

[M.O.T]

Precisely. I run those in a "consensus" system where I'm playing only games that they all agree on. Only 2-3 of those games came up on the SD system as far as top plays that I would risk money on.

This is all only off one week and I will keep everything updated for anyone interested,

Definitely interested, I’m a math nerd. I’ve been working to improve my PRs the last few years but my total plays are still largely based on feel. I’m definitely looking for ways to take a more quantitative approach.

I understand if you don’t want to share this as it really appears to be the foundation the process is built on, but is the consensus system a projection from 4 other sources or some sort of output from one model for the same game...like 4 different simulations?

 

[HonestD]

Covers, Bleacher Report, Accuscore, and Scoreboard.

Been using all but Scoreboard for years now, and they are good sources in terms of being in the right ball park the majority of the time which is what's most important for this.

 

[CHSGTMD]

thanks for posting HonestD
Will be following your posts and how this works out week to week

 

[HonestD]

thanks for posting HonestD
Will be following your posts and how this works out week to week

My pleasure.

 

[TahoeLegend]

I like the idea of using math as much as possible in handicapping so thanks for posting.

My question is the same as MOT. It appears to me everything depends on the accuracy of the "consensus." If that is off then everything will be off.

Are you using projections from those sites of what the line might be to form your consensus or is it the total they have after allowing the public to bet into it? The reason I ask is that I never see--and have never seen--a 9-point deviation between low and high total as you show in the TT/Houston game, no matter how many sites I check.

Or maybe I just don't understand your process.

 

[HonestD]

I like the idea of using math as much as possible in handicapping so thanks for posting.

My question is the same as MOT. It appears to me everything depends on the accuracy of the "consensus." If that is off then everything will be off.

Are you using projections from those sites of what the line might be to form your consensus or is it the total they have after allowing the public to bet into it? The reason I ask is that I never see--and have never seen--a 9-point deviation between low and high total as you show in the TT/Houston game, no matter how many sites I check.

Or maybe I just don't understand your process.

For sure. Everything depends on the accuracy of the source. They're just projections from the sources. The S source, which you're alluding to is the most volatile of them all, but its totals record individually is pretty respectable. With that said, that's why I'm using the standard deviation calculation. I'm not JUST concern with the consensus. I'm more concerned with accuracy in relation to the predictions from the four sources being similar in value which in turn will hopefully lead to an accurate estimate for a bet, In my opinion the SD will help factor out some of the luck and variance.


I'm strictly taking the total they predict and using that as my foundation. I've been using these sites for several years now so I have a feel for them already as far as being legit. Never in this finite detail, but time will tell right?

 

[Johnnyonthespot]

I like using math and derivations, its factual.

l hate using consensus input, it's not factual. Particularly those consensus sources. Your putting value on voodoo

My recommendation is to never tie a thoughtful model with a consensus filter from covers.

Nothing personal, I wish you luck but this is borderline ridiculous.

 

[HonestD]

This should give you an idea of what the consensus sheet loosk like. All the results are taken off the highlighted 5 pm line as that's when I usually place most my bets.

A open box means its a consensus pick SU and the closes boxes means it's a unanimous pick SU. Highlighted blue boxes are what I'm calling power plays. Those are plays where all four systems agree on an ATS or Total bet.

 

[Pyppurs]

Interesting approach, appreciate you willing to discuss something like this on the board/public

3 quick question if you feel like answering.

1) Have you decided whether you'd be attacking opening or closing numbers? Why?
2) Would you be weighting the input params as the season goes on, if one source becomes better at predicting the total than the others?
3) Couldn't you assign the model to Team Totals as well?

 

[HonestD]

I like using math and derivations, its factual.

l hate using consensus input, it's not factual. Particularly those consensus sources. Your putting value on voodoo

My recommendation is to never tie a thoughtful model with a consensus filter from covers.

Nothing personal, I wish you luck but this is borderline ridiculous.

None taken at all. All input is welcome. For me most of this stuff is purely experimental. Ridiculous? Maybe but that wouldn't be the first nor last time I partake in something like this I'm sure.

 

[HonestD]

Interesting approach, appreciate you willing to discuss something like this on the board/public

3 quick question if you feel like answering.

1) Have you decided whether you'd be attacking opening or closing numbers? Why?
2) Would you be weighting the input params as the season goes on, if one source becomes better at predicting the total than the others?
3) Couldn't you assign the model to Team Totals as well?

1. I can't attack opening numbers because I don't have the data to input that early since the sources aren't my own.
2. I've talked to a friend about that but but it's going to take more games to have a better idea.
3. Yeah you can. I spend a few hours late at night messing around with stuff, but I'm a one man band and can't work out everything, but yeah, it would be the same approach.


I can take criticism. At the end of the day I'm just here to share my thoughts and hopefully eventually we'll all stumble upon something we can mold or take little bits and pieces from to all make money.

 

[Johnnyonthespot]

It's kinda like this, there is a company named sportsinsights that developed this market and idea about consensus, they never made a profit as a company and sold out more than a year ago, they made over 600 baseball plays this year based solely on their primary algos and experience with line movement, they are down money, and they are thought / propaganda leaders, its noise

 

[HonestD]

It's kinda like this, there is a company named sportsinsights that developed this market and idea, they never made a profit and sold out more than a year ago, they made over 600 baseball plays this year based solely on their primary algos and experience with line movement, they are down money, and they are thought / propaganda leaders, its noise

Yeah I know who they are. I subbed to them for a few months for their data last year. The difference between them and me though is that I'm just a normal guy willing to try anything for an edge. I don't make claims and don't take peoples' money. I just try whatever ideas cross my mind and sometime I decide to share.

 

[DBURGH]

Thanks for the info.

 

[HonestD]

Week 4 NCAAF

I have included plays and the window to use based on what line you get. All my plays listed have an SD of 7 or less and a ratio of 33% or less. For example, the Northern Illinois/Florida State window is 32-46. The total is 45. That sits 13 away from the 45 on the bottom side and 1 away from the top on the top side. We're always going towards the bigger number which is 13, aka the under. 1/13 = 7%

Notre Dame/Wake Forest under 60 (45-59)
Northern Illinois/Florida St. under 45 (32-46)
TCU/Texas over 46.5 (50-60)
Kansas St./West Virginia under 60.5 (49-63)
South Carolina/Vanderbilt over 54.5 (53-63)
Akron/Iowa St. over 47.5 (46-54)
Tulsa/Temple over 53.5 (52-60)
Pittsburgh/UNC over 48.5 (48-54)
South Alabama/Memphis over 67 (70-82)

TCU/Texas, Notre Dame/Wake Forest, and South Alabama/Memphis will be 1.5x plays for me since the line sits outside the entire range. Last week plays like this went 2-0.

 

[TahoeLegend]

Thanks for posting your theory.

I'll be interested in how it does as the season progresses.

 

[loggy]

I did something similar years ago. Tracked I think 8-10 sources that provided exact score predictions and took the average vs. the spread and total. It actually did pretty well, around 55%. Agree w/ Johnny about the "ridiculousness" of having faith in their predictions, but the averaging seemed to control it somewhat. If I remember right, it seemed like the games with the greatest deviation from the actual number did worse, aka everyone thinks a certain dog will win outright.

I gave it up because it became tedious to wait and constantly check for the sources to post their predictions, even though the results were profitable.

 

[HonestD]

I did something similar years ago. Tracked I think 8-10 sources that provided exact score predictions and took the average vs. the spread and total. It actually did pretty well, around 55%. Agree w/ Johnny about the "ridiculousness" of having faith in their predictions, but the averaging seemed to control it somewhat. If I remember right, it seemed like the games with the greatest deviation from the actual number did worse, aka everyone thinks a certain dog will win outright.

I gave it up because it became tedious to wait and constantly check for the sources to post their predictions, even though the results were profitable.

I hear what you're saying and I totally agree. I chose to go with four that I trust "somewhat". Over time it may dwindle down to three or two or the entire thing may go kaput, who the heck really knows, but if we can get into the "ball park" more times than not with the totals the low SD threshold should help control the variance and help pinpoint solid games. I mean they all can't be wrong all the time.

 

[HonestD]

Thanks for posting your theory.

I'll be interested in how it does as the season progresses.

It'll be fun to see. At least it gives me action tonight :abouttime:

 

[Johnnyonthespot]

That's not to say that there aren't strong indicators in line movement or important information to be gleaned from baneling with Pinnacle. I just think consensus is too much of an estimate with little proof to build a decision. I appreciate you sharing. It's not easy.

 

[HonestD]

That's not to say that there aren't strong indicators in line movement or important information to be gleaned from baneling with Pinnacle. I just think consensus is too much of an estimate with little proof to build a decision. I appreciate you sharing. It's not easy.

I'm not using line moves at all for this