It's one year to the day when I joined pickmonitor.com, now line.com. I wanted to write something in depth, more then just general guidelines. Something beginners could read and maybe help those people get a jump start most of us never had.
This is my system, it's not how I arrive at every play, but it's how I arrive at most. If you grasp the concepts I discuss here you will be profitable long term. This is applicable to NFL, NCAA BB, and NBA. It's not purely statistics so certain areas will still be left to your mind to filter, but the fundamentals are all here.
If you don't know how to build a predictive model using power rankings you need to find a site that does. Plenty exist in every sport. Some sites do multiple sports, others specialize in one sport. The foundation of the system is to be able to look at two teams playing and with stats alone get as close to the real winning% as possible. I always utilize 3 separate sets of power ratings. One always comes from me, the other two come from sources I feel are most relevant.
Once you have broken down the board to reflect your anticipated winning % convert the winning % into a money line. Add the actual Vegas line to each game. Sort the spread sheet by your highest favorite, copy that entire column. Sort by the highest Vegas favorite, and paste your highest favorite next to what Vegas shows. Essentially what this does is establish what games have the highest separation from your anticipated winning % compared to Vegas anticipated winning %. The final result will look something like this.
Prediction
1
2
3
ML
Line
Vegas
Diff
Team A
98%
87%
93%
-1265.5
-19
-2010.3
744.7
Team B
89%
90%
90%
-861.9
-15
-1247.3
385.3
Team C
97%
94%
95%
-2010.3
-19
-2374.1
363.9
Team D
89%
79%
84%
-533.9
-13
-895.1
361.2
Team E
86%
79%
82%
-474.8
-13
-826.7
351.9
Team F
92%
75%
84%
-512.9
-13
-861.9
349.0
Team G
88%
90%
89%
-826.7
-14
-1139.6
312.9
Team H
94%
91%
92%
-1203.7
-16
-1455.4
251.7
Team I
95%
90%
93%
-1247.3
-18
-1467.5
220.2
What I show here in order, Team, 3 power rankings, the anticipated moneyline, the vegas line, the anticipated vegas moneyline (based on line), and the difference between the anticipated moneyline and the vegas moneyline.
What I look for with this chart is games in which my Numbers and Vegas numbers differ by at least 50 dollars on the ML. These are the games I wan't to analyze more. Further, within these games I eliminate large spreads. It depends on the sport, but I wan't somewhat comparable match ups, avoiding best vs worst type games. If you notice in the spreadsheet example I posted, all those games would be eliminated. It's very easy to see a 50 dollar disparity with big lines. That's why I eliminate them. It's much more rare (so long as your model is accurate) to see that disparity as lines draw closer to even.
So what is left is supposedly competitive games with a disparity of $50 or more. From that point I break them down further into 3 categories.
-Games not backed by the public, positive value (ideal)
-Games not backed by the public, negative value (given consideration)
-Fade games that show positive value and are backed by the public
The fundamental theory is pretty simple. It is my belief that most games oddsmakers over juice the public for small amounts. Public % information is available from numerous sources but it helps knowing what direction will be sided with prior to all that information being released. Anticipating that is certainly an attainable skill, once you start breaking down games as I show you will start to be able to guess 90% or better what side most the action will come from. This will allow you to hit the overnight lines more. Unfortunately with ML's you need to wait until the morning.
The final process is justification of data. I take my remaining games and review them in depth to prove the data is relevant.
Let's give some examples of the three groups I mention for clarification.
Showing Value not backed by public
I show Team A is playing Team B. I have Team A at -175, Vegas has them at -120. The public is backing Team B. So in this situation I show almost 55 cents in value toward Team A, yet the public still likes Team B. This is a great situation to start with. I will take this play and analyze it in depth. The goal being, disprove why I should be getting 55 cents. If I can't disprove it, I feel confident I'm on the side of value. Vegas didn't have any reason to Juice me, the public is already backing the team that is getting worse odds.
Showing Negative value not backed by public
I show Team A is playing Team B. I have Team A at -120, Vegas has Team A at -175. The public is backing team B. In this case I'm showing value in the same game the public is backing. Showing value with statistical models on the same team the public is backing should serve as a major red flag. Vegas does not need to make Team A -175, my odds show -120, and people are still backing Team B regardless. This is an awesome situation to break down further. The goal is to find out why the heck Vegas feels so strong about Team A. If you find supporting evidence, you have a play.
Showing positive value and backed by the public (FADE)
Team A is playing Team B. I have Team A at -175, Vegas has Team A at -120. The public is backing Team A. This is another great set up. I show 55 dollars in value on Team A, the public is backing Team A. Everything says Team A. Yet Vegas gives us a better line on Team A? I don't think so. This turns into a potential fade situation. My goal at this point is to see why Vegas is so cold on Team A that they would be willing to give myself and the public a better price for a team they already know will get heavy action. Tread lightly here, it takes practice to take this past break even but it certainly can be done.
Reading some of this might sound confusing. I really hope I was able to convey the approach in an understandable way. Thanks for a great 1st year to Patrick, and the community. This site has become a major part of my life and I really enjoy the people who are a part of it. Let's make 2012 an even better year!
Probably because I suck lately and it's a really long post. Kidding aside, it has a few concepts that are very important to understand, I didn't expect a lot of responses.
Just for discussions sake lets kick around an idea that's been eating at me.
How is value truly established? Not just the perception of value, but guaranteed value? I think you start by looking at how lines are made. The line is power rankings, sprinkled with a dash of perception, history,injuries, and circumstances. So how can we attack that process in a way nobody else does?
The key to me is injuries, but not in the way people usually analyze injuries.
Lets take a random football team and place them in a very common scenario. They enter camp and two starters go out. After week 2 they have 5 starters out. They stay pretty banged up for several weeks muddling along to a below average record. Maybe a couple decent profile players, not the QB, but these are starters. Then all of a sudden they start getting people healthy. For the rest of the season the statistics people will use, which also make up a huge % of the Vegas Power Rankings, are not actually the real statistics of this team when healthy.
People do not consider the previous injuries because the players previously injured don't show up on the current injury report. But you can rest assured they show up in the teams statistics. How can't they? Looking at NCAA BB this year, thinking to Xavier when they had suspensions. The top 3 guys were out. Now they're back. Yet Xaviers statistics are not based on them playing in every game. So even if we have the most accurate power ranking system known to man, how can we say a teams statistics are accurate?
I love the idea of building a database to attack this angle because the concept of value is just about indisputable.I don't know anybody who has a database that attempts to take this into consideration. The problem is to build such a database would take probably a year at the least. Maybe longer. Finding the info would be pain staking. But gosh do I think it's worth pursuing. If you could apply this angle with other methodologies you could be very dangerous.
Hi Mike! Great pieces. I do have some questions for you:
What equation or formula do you use to derive the winning % on your own for your power rankings?
What sources do you go to (or are available, if you don't want to reveal the 2 you use) that offer their winning % for their power rankings?
How do you convert the %s into a money line?
I understand clearly about your three groups; that part was easy enough to understand. This is definitely different than how I normally approach things and has piqued my interest enough where I would like to learn more; sorry if my questions sound noobish.
Also to touch upon your interest in building a database that takes injuries into consideration. I am doing something a bit similar right now. I'm currently building a database of the NBA players of each team and their estimated contribution to the team and efficiency to determine the effect of the loss of any player due to injury or other reasons. It's not really a historical database that could be used to look back at past results, but more of a current database that is updated regularly and can be used as a tool to help with game analysis. Another project I'm working on is a database of all NHL and NBA games this season to analyze the effects of exhaustion and fatigue on a team due to back to back games that involve traveling and more than 3 games in a week with at least 2 games requiring traveling across states. This would be more of a historical database that can be useful in finding strong trends that could be used to make strong wagers in games with similar variables.
Mike, you just described the New York Giants perfectly.
That banged up defensive line was gone until a few games left in the season. When healthy, they played the best defense pound for pound then anyone has all year. Many still had the assumption of New York having a bad defense, even though they were a top 3 by playoff time.
@Next, very true, many people looked at the Giants as if they barely deserved to be in the playoffs. This spot occurs a lot in many sports. Just for example, a few days ago, the leafs got back Liles and Armstrong, not superstars, but above average players. They also played Reimer who made a solid outing a few weeks back, but didn't get playing time because of Gustavson's solid performance. All three players added to the team resulted in them nearly beating Pittsburgh twice, the first loss coming without Reimer, and killing Ottawa 5 to nothing last game. There is value in underrated players.
If you watched the replays, you'd probably admit that the collapse was in large due to poor goal tending. That was really bad luck, I mean, Kessel even missed an open net that would have put them up by one if he had just lifted the puck.
It's Toronto/New York a few years ago when Toronto was up 4-0 with 5 minutes to go and lost in regulation. Certain team have great comebacks and others every once in a while, great collapses.
Then of course on the 2nd of the home and home, they beat Pittsburgh 1-0 the following night. Gotta live with those, they will come back in your favor eventually.
Every team has their "identity" long term. In MLB, Philadelphia is known for playing average until the last few innings. They lead the league in comeback wins over the past 2-3 seasons.
I have been apart of quite a few. Every time I bet on them, they take me for a ride. It usually ends well so not a bad ride.
1- It varies from sport to sport. The college sports have major differences to account for the large disparity in talent. I usually run correlation data with points without attempting to factor actual points. Essentially combining highly correlated statistics with points together to form one number.
2. I can't reveal most my sources, that would be taken this to far. But converting any set of rankings, even if it's teams ranked 1-30 is possible. For example one I use for NFL is from profootballoutsiders. They don't convert the winning % for you, but based on the rankings and historical rankings they show, you could create a formula to assign a winning% with good accuracy.
3. Just for simplicity sake I only convert favorites, two separate formulas to convert favorites and dogs. For favorites the formula is (-100*win %)/(1-win%). I throw in -10 at the end to account for juice, but as moneylines grow that number grows as well.
As for your comments about the injury post I think your on the right track. Any work involving travel distances, time zone changes, fatigue resulting from scheduling are all very much valid. Oddsmaker's certainly adjust to fatigue based on scheduling, probably still not enough, but the data is indisputable that it does adversely effect performance so once it deemed definitive they will factor it in.
Your work with injuries is smart. It's not exactly the angle I'm looking at, but smart none the less. I think the historical context is what will provide the exceptional value I'm seeking (also provide the ungodly amount of time to create). Certainly for the database to be effective you would have to find a way to establish some lose value for a player or player category. For example, starter, all star etc. One thing I have noticed with injuries over the years is if it's high profile, fading that team is generally 50/50. This is very common when a QB goes down. People always fade teams without the starting QB but that is to high profile to get value out of in most cases (certainly not Peyton this year).
Just as an example here is Wednesdays card. 2 plays qualify.
I project Tennessee at -295 the line indicates -350. This is an example in which I show positive value on South Carolina and the majority of wagers will be on Tennessee.
I project Buffalo at -386 the line indicates -263. This is an example in which I show negative value on Central Michigan, but the majority of wagers are on Buffalo. So it becomes a fade Buffalo play.
It won't be an on going thing Hopper, I was just trying to give people the tools to replicate it without necessarily being able to do it EXACTLY like I do it. But yes those are the plays. I would still suggest SC +7, but now the Central line has dropped to +5 vs +6. So that's now a no play at that number. Most of the plays that qualify for this will shift against me overnight. So it's best when utilizing this to hit the games as early as possible.
It's one year to the day when I joined pickmonitor.com, now line.com. I wanted to write something in depth, more then just general guidelines. Something beginners could read and maybe help those people get a jump start most of us never had.
This is my system, it's not how I arrive at every play, but it's how I arrive at most. If you grasp the concepts I discuss here you will be profitable long term. This is applicable to NFL, NCAA BB, and NBA. It's not purely statistics so certain areas will still be left to your mind to filter, but the fundamentals are all here.
If you don't know how to build a predictive model using power rankings you need to find a site that does. Plenty exist in every sport. Some sites do multiple sports, others specialize in one sport. The foundation of the system is to be able to look at two teams playing and with stats alone get as close to the real winning% as possible. I always utilize 3 separate sets of power ratings. One always comes from me, the other two come from sources I feel are most relevant.
Once you have broken down the board to reflect your anticipated winning % convert the winning % into a money line. Add the actual Vegas line to each game. Sort the spread sheet by your highest favorite, copy that entire column. Sort by the highest Vegas favorite, and paste your highest favorite next to what Vegas shows. Essentially what this does is establish what games have the highest separation from your anticipated winning % compared to Vegas anticipated winning %. The final result will look something like this.
Team B
What I show here in order, Team, 3 power rankings, the anticipated moneyline, the vegas line, the anticipated vegas moneyline (based on line), and the difference between the anticipated moneyline and the vegas moneyline.
What I look for with this chart is games in which my Numbers and Vegas numbers differ by at least 50 dollars on the ML. These are the games I wan't to analyze more. Further, within these games I eliminate large spreads. It depends on the sport, but I wan't somewhat comparable match ups, avoiding best vs worst type games. If you notice in the spreadsheet example I posted, all those games would be eliminated. It's very easy to see a 50 dollar disparity with big lines. That's why I eliminate them. It's much more rare (so long as your model is accurate) to see that disparity as lines draw closer to even.
So what is left is supposedly competitive games with a disparity of $50 or more. From that point I break them down further into 3 categories.
-Games not backed by the public, positive value (ideal)
-Games not backed by the public, negative value (given consideration)
-Fade games that show positive value and are backed by the public
The fundamental theory is pretty simple. It is my belief that most games oddsmakers over juice the public for small amounts. Public % information is available from numerous sources but it helps knowing what direction will be sided with prior to all that information being released. Anticipating that is certainly an attainable skill, once you start breaking down games as I show you will start to be able to guess 90% or better what side most the action will come from. This will allow you to hit the overnight lines more. Unfortunately with ML's you need to wait until the morning.
The final process is justification of data. I take my remaining games and review them in depth to prove the data is relevant.
Let's give some examples of the three groups I mention for clarification.
Showing Value not backed by public
I show Team A is playing Team B. I have Team A at -175, Vegas has them at -120. The public is backing Team B. So in this situation I show almost 55 cents in value toward Team A, yet the public still likes Team B. This is a great situation to start with. I will take this play and analyze it in depth. The goal being, disprove why I should be getting 55 cents. If I can't disprove it, I feel confident I'm on the side of value. Vegas didn't have any reason to Juice me, the public is already backing the team that is getting worse odds.
Showing Negative value not backed by public
I show Team A is playing Team B. I have Team A at -120, Vegas has Team A at -175. The public is backing team B. In this case I'm showing value in the same game the public is backing. Showing value with statistical models on the same team the public is backing should serve as a major red flag. Vegas does not need to make Team A -175, my odds show -120, and people are still backing Team B regardless. This is an awesome situation to break down further. The goal is to find out why the heck Vegas feels so strong about Team A. If you find supporting evidence, you have a play.
Showing positive value and backed by the public (FADE)
Team A is playing Team B. I have Team A at -175, Vegas has Team A at -120. The public is backing Team A. This is another great set up. I show 55 dollars in value on Team A, the public is backing Team A. Everything says Team A. Yet Vegas gives us a better line on Team A? I don't think so. This turns into a potential fade situation. My goal at this point is to see why Vegas is so cold on Team A that they would be willing to give myself and the public a better price for a team they already know will get heavy action. Tread lightly here, it takes practice to take this past break even but it certainly can be done.
Reading some of this might sound confusing. I really hope I was able to convey the approach in an understandable way. Thanks for a great 1st year to Patrick, and the community. This site has become a major part of my life and I really enjoy the people who are a part of it. Let's make 2012 an even better year!
Thanked by
Probably because I suck lately and it's a really long post.
Kidding aside, it has a few concepts that are very important to understand, I didn't expect a lot of responses.
Just for discussions sake lets kick around an idea that's been eating at me.
How is value truly established? Not just the perception of value, but guaranteed value? I think you start by looking at how lines are made. The line is power rankings, sprinkled with a dash of perception, history,injuries, and circumstances. So how can we attack that process in a way nobody else does?
The key to me is injuries, but not in the way people usually analyze injuries.
Lets take a random football team and place them in a very common scenario. They enter camp and two starters go out. After week 2 they have 5 starters out. They stay pretty banged up for several weeks muddling along to a below average record. Maybe a couple decent profile players, not the QB, but these are starters. Then all of a sudden they start getting people healthy. For the rest of the season the statistics people will use, which also make up a huge % of the Vegas Power Rankings, are not actually the real statistics of this team when healthy.
People do not consider the previous injuries because the players previously injured don't show up on the current injury report. But you can rest assured they show up in the teams statistics. How can't they? Looking at NCAA BB this year, thinking to Xavier when they had suspensions. The top 3 guys were out. Now they're back. Yet Xaviers statistics are not based on them playing in every game. So even if we have the most accurate power ranking system known to man, how can we say a teams statistics are accurate?
I love the idea of building a database to attack this angle because the concept of value is just about indisputable.I don't know anybody who has a database that attempts to take this into consideration. The problem is to build such a database would take probably a year at the least. Maybe longer. Finding the info would be pain staking. But gosh do I think it's worth pursuing. If you could apply this angle with other methodologies you could be very dangerous.
Thanked by
Hi Mike! Great pieces. I do have some questions for you:
I understand clearly about your three groups; that part was easy enough to understand. This is definitely different than how I normally approach things and has piqued my interest enough where I would like to learn more; sorry if my questions sound noobish.
Also to touch upon your interest in building a database that takes injuries into consideration. I am doing something a bit similar right now. I'm currently building a database of the NBA players of each team and their estimated contribution to the team and efficiency to determine the effect of the loss of any player due to injury or other reasons. It's not really a historical database that could be used to look back at past results, but more of a current database that is updated regularly and can be used as a tool to help with game analysis. Another project I'm working on is a database of all NHL and NBA games this season to analyze the effects of exhaustion and fatigue on a team due to back to back games that involve traveling and more than 3 games in a week with at least 2 games requiring traveling across states. This would be more of a historical database that can be useful in finding strong trends that could be used to make strong wagers in games with similar variables.
Thanked by
Mike, you just described the New York Giants perfectly.
That banged up defensive line was gone until a few games left in the season. When healthy, they played the best defense pound for pound then anyone has all year. Many still had the assumption of New York having a bad defense, even though they were a top 3 by playoff time.
It's Toronto/New York a few years ago when Toronto was up 4-0 with 5 minutes to go and lost in regulation. Certain team have great comebacks and others every once in a while, great collapses.
Then of course on the 2nd of the home and home, they beat Pittsburgh 1-0 the following night. Gotta live with those, they will come back in your favor eventually.
Every team has their "identity" long term. In MLB, Philadelphia is known for playing average until the last few innings. They lead the league in comeback wins over the past 2-3 seasons.
I have been apart of quite a few. Every time I bet on them, they take me for a ride. It usually ends well so not a bad ride.
@Mike
Did not mean to take over your post.
Jonathan-
1- It varies from sport to sport. The college sports have major differences to account for the large disparity in talent. I usually run correlation data with points without attempting to factor actual points. Essentially combining highly correlated statistics with points together to form one number.
2. I can't reveal most my sources, that would be taken this to far. But converting any set of rankings, even if it's teams ranked 1-30 is possible. For example one I use for NFL is from profootballoutsiders. They don't convert the winning % for you, but based on the rankings and historical rankings they show, you could create a formula to assign a winning% with good accuracy.
3. Just for simplicity sake I only convert favorites, two separate formulas to convert favorites and dogs. For favorites the formula is (-100*win %)/(1-win%). I throw in -10 at the end to account for juice, but as moneylines grow that number grows as well.
As for your comments about the injury post I think your on the right track. Any work involving travel distances, time zone changes, fatigue resulting from scheduling are all very much valid. Oddsmaker's certainly adjust to fatigue based on scheduling, probably still not enough, but the data is indisputable that it does adversely effect performance so once it deemed definitive they will factor it in.
Your work with injuries is smart. It's not exactly the angle I'm looking at, but smart none the less. I think the historical context is what will provide the exceptional value I'm seeking (also provide the ungodly amount of time to create). Certainly for the database to be effective you would have to find a way to establish some lose value for a player or player category. For example, starter, all star etc. One thing I have noticed with injuries over the years is if it's high profile, fading that team is generally 50/50. This is very common when a QB goes down. People always fade teams without the starting QB but that is to high profile to get value out of in most cases (certainly not Peyton this year).
Thanked by
Just as an example here is Wednesdays card. 2 plays qualify.
I project Tennessee at -295 the line indicates -350. This is an example in which I show positive value on South Carolina and the majority of wagers will be on Tennessee.
I project Buffalo at -386 the line indicates -263. This is an example in which I show negative value on Central Michigan, but the majority of wagers are on Buffalo. So it becomes a fade Buffalo play.
Thanked by
Thanked by