[demens]

Hey. I've been trying to add a bit more of a statistical, math based foundation to my picks. My head is spinning a bit after a few days of reading through the threads in this section. I wanted to see if i'm on the right track and what you guys think of capping this way.

I came across a formula that predicts the chances of one team beating another (very simple formula). But before i get into asking question about that i wanna see if i'm even starting at the right place and going in the right direction.

Basically, i started trying to figure out what the Win% number should be to put into the Kelly formula. That number seems to originate out of thin air in most of the posts i've read, i'm not even sure if trying to calculate it has any advantage over just guessing it.

But to figure that out i think you need to figure out what edge you have. Edge is another factor that i think most calculations get about as close to it as pulling a number out of my ass (needless to say i'm not sold on this formula i found being that useful). So anyway, in order to calculate an edge there is got to be a line you can compare to the existing line available. So you have to calculate your own line.

How do you do that? This is where i think this formula can help. I found it here:
http://www.rawbw.com/~deano/methdesc.html
Under Calculating Matchup Probabilities for Teams
Win%A_B = [Win%A*(1-Win%B)]/[Win%A*(1-Win%B)+(1-Win%A)*Win%B]
Basically just taking the Winning % of Team A and Team B gives you the % Team A will win the game. This is the simple version of it, there is another that accounts for home court.

After getting the Win%A vs B there is some way to convert that number into a Moneyline, and then convert the moneyline into a line.

What formula is used to convert % into ML?

Its not really very important because there is the ML to Spread tool on SBR does it, but you have to start with the spread or ML to see the %, so its not direct and you have to guess a few times to hit the right. It works but i would like to know how to calculate it directly.

So what do you guys think of this formula? Is this something useful or is it pure guess work written with numbers.

From reading a bit I know a lot of the "smarter" posters around here are not quick to give answers and reveal their secrets, but the starting point really shouldn't be that big of a secret. I think the real value is in what numbers you end up putting into the formula not the formula or the steps themselves.

Thoughts?

[usernametaken]

I think it is important to measure how much team a hates team b> If the intensity is boiling over= win

[demens]

Hmm, thanks for all the insight.

[gryfyn1]

You can just put you expected win % into the 'implied probability' box and it will spit out the corresponding odds. at least the true odds, which books will adjust for juice.

Or you can do 1+(100- Exp%)/Exp% - that will give you the decimal odds.

[demens]

How do you convert/adjust the "fair line" with juice.

[mmortal03]

I'm not a pro at this, but this is how I calculate my lines at the moment. I'll use the NBA as an example, but this applies just as well to how I would do any basketball or football league in general. I'm very interested in some constructive criticism on this method.

First off, I start with a power rating for each team. There are any number of ways out there on how to calculate a power rating, so I'll leave that up to you to come up with. After I have my particular power ratings of every team, I always transform them into a scale that goes from 0 to 1 (with the worst team being 0, and the best team being 1). This is pretty simple math to do, so I will leave that as an exercise for you.

Next, I calculate my multiplier value, which is used to covert these "normalized" (0 to 1) power ratings into points units. To get the multiplier value, I take the difference between the top scoring margin per game in the league at the time (aka highest point differential) and the worst scoring margin per game in the league at the time (aka lowest point differential).

As a practical example, right now it would be the Miami Heat's 7.260 minus the Cleveland Cavaliers' -9.934, which would equal 17.194.

Using this multiplier value, I can multiply each team's "normalized" power rating by it, and then take the difference of any two teams' resulting values to get my line on a neutral court. After that, I correct for home court advantage and any other situational handicapping circumstances, and I'm good to go.

[demens]

Thats interesting. I have not heard of such a method before. Not sure i fully understand it but maybe if i went thought the math i could see if it makes sense. How do your results come out in terms of the Vegas lines, and if they are off is your edge backed up by actually winning results in your favor? From what i read the spread calcualtions are based on either a combination of ppg, or a teams win%. I'm not quite sure what goes into power ratings.

I tried capping all of todays games (NBA), my results are here:
http://www.sportsbookreview.com/forum/nba-basket...4-05-11-a.html

I used 6 different methods. 3 using the formula i mentioned here already. And 4 using a similar formula Pythagorean expectation that looks like this. P[A] = 1/(1+ (TT:B /TT:A)^X )

The 4 latter methods can really be 2 because the other 2 produced results almost identical to each other. The difference in them was using straight (ppg*opp def ppg/league ave) for team total and using a pace adjustment, using off and defrtgs for each team instead of ppg and then multiplying that by the new game pace that you come up with. It made a small difference when converting to MLs but honestly out of the 13 games it only impacted 2 of the lines and the difference was 0.5....and even that was much closer because i had to round it opposite directions so i feel like the extra steps i took with calculating and adjusting for pace where essentially useless. This is the method i used in the "What i think the line WILL be", after the getting the raw numbers you must adjust for HCA.

The 2nd variation of this, instead of using season ppgs, i used home and away pggs. I really did not like the results it produced at all. All of them were severely scwed to favor the home team. These results obviously do not need to be adjusted for HCA but they were off by 3 or 4 (for some teams even 7) points off the line i got from the 1st method after HCA adjustment. It makes sense to me because the difference between home and away production is very significant. I'm not gonna discard this method until i see the results, it seems way off in terms of the line but maybe thats where it should be and the vegas lines are not adjusting enough for home court? (Doubt it).

The 3 methods based on the equation i listed 1st break down like this. 1 used the equation exactly how it is, 2 used the question that incorporates home and away win% and 3 was a completely subjective method.
In the 1st 2 cases i used the Gaussian # for the teams Win%, for home and away win% i used the league average. The former obviously is adjusted for HCA after the fact. Both version spit out results within a point of each other (more or less), and those results were close to a point off the results from the best method used above.

The last and imo best method i wont get into much. Its using the same equation as the 2nd method (h/A) but i'm being very subjective in what # i use for Win%s.

All these methods produce only a Win Probability. The next step is to covert it into the line which is where i stumbled a bunch.

1 method is i converted that into MLs and used the ML converted on SBR to get the line. The ML i was getting is a No Vig ML, i dont know how to convert it into one adjusted for juice but when i played with the MLs in the tool making it less attractive for both teams it never changed the spread number so i figure its not important. I have read that this tool isn't very accurate because it does not account for game totals. I found another tool that does and honestly i sort of like the SBR tools results better.If anyone has the formula it uses i'd like to know what it is.

The 2nd tool is from this site here, i think its from one of the users on the forum, UVsomething...
http://sportsobjective.com/wordpress/?p=3554
Using that gave me lines that were again, about a point off SBRs tool. The higher the spread the larger the difference was.

So anyway, still waiting on the rest of the lines to see how close i came. And waiting on results to see if there was any advantage. The "what i think the line SHOULD be" section i using that subjective method.

[mmortal03]

How do your results come out in terms of the Vegas lines, and if they are off is your edge backed up by actually winning results in your favor? From what i read the spread calcualtions are based on either a combination of ppg, or a teams win%. I'm not quite sure what goes into power ratings.

My power ratings use a couple different ways of combining strength of schedule, winning percentage, and scoring margin, and that's it. After I convert these to neutral site lines, I then do some systematic adjustments based on scheduling and home court advantage. That's as complex as I've gotten so far in terms of setting a line.

Regarding my edge, it largely depends on which power rating I'm plugging into my line transformation method from above. Some do better than others. There could additionally be some superior way to convert any of my power ratings to points units than what I've come up with, but I haven't sat down and figured that one out yet.

Going back to 2005, my best performing basic power rating across all games after December 15th of each season (5221 games to the present) would have won 51.95% of the time against the closing line. Obviously, that percentage by itself is not enough to make money, but no one seriously bets every game. It's just my starting framework.

In terms of how I do when I am off from the professional lines, at extremes of +-1.5 standard deviations or more, my methods tend to do well with large positive deviations from the line, and not with large negative deviations from it. I couldn't tell you exactly why that is, but it's one thing that I've taken advantage of to enhance my EV: I drop all large negative deviations from my picks.

How big are the deviations in absolute terms? My lines differ with the professional lines by less than a point on average (about 0.75 points), and have a correlation coefficient with it of 0.88. They can vary at the extremes by 10 points or more, but that's a small subset.

I would doubt that my correlation with the pro lines is optimal, but, remember, we're trying to predict game results here, not the line! For example, some of my power ratings have in fact had higher correlations with the line, but produced a worse edge. That's because where you need to be is on the same side of the bookie's line as the game results, whether that is by 1 point or 30 points. Higher correlation with the game result is the superior goal here. With that said, I've tried to do some simple artificial tweaking of my lines to make them better fit the game results, but I haven't seen much success doing that. Instead, I'd probably do better to systematically integrate some completely new information, like injured player info or something of the sort.