Handicapping Theory (3/3) Combined Analysis and Best Bets
By: Dr Bob
Combined Analysis: The key to the 2004 research on my methods was finding a way to combine my situational analysis, fundamental indicators and my math model to give me an overall chance of a team covering at any given number. My performance on my Best Bets the last 5 seasons is an indication that I succeeded in that endeavor and I will continue to refine the accuracy of my methods each year.
As an example, consider a hypothetical game between the New Orleans Saints and Oakland Raiders. New Orleans applies to a 140-60-5 ATS situation that uses 6 parameters, but Oakland also applies to a statistical profile indicator with a record of 86-28-4 ATS. My NFL math model favors New Orleans by 12.4 points when they are only a 7 point favorite in reality. As discussed above, a situation with a record of 140-60-5 and 6 parameters has a 55.8% chance of winning if the line is fair. The fundamental indicator favoring the Raiders has a 56.1% chance of winning given a fair line, and my math model would give the Saints a 56.9% chance of covering at a line of -7 points. The trick is assigning a point value to the situation and the fundamental indicator based on their chance of covering at a fair line. I simply put everything in terms of points based on the relationship between point differentials and the chance of covering of my math model. Each point difference in my math model is worth about 1.3% in chance of covering, so each percentage point is worth about 0.8 points (1/1.3). In this case, the situation favoring New Orleans is worth 4.5 points while the fundamental indicator favoring Oakland is worth 4.8 points. My math model favors New Orleans by 12.4 points, so adding the value of the situation and the indicator would result in an overall prediction of the Saints by 12.1 points (+4.5 - 4.8 + 12.4 = 12.1), which would give them a 56.5% chance of covering at the line of -7 points. Obviously, things can become a lot more complicated when there are multiple situations and indicators applying to a particular game - which is most often the case, but my years of studying probability theory have given me the tools to sort through it all and come up with an accurate measure of the overall affect of the situations and indicators.
Best Bets: Over the last 5 seasons I have gone 572-414-28 (58%) on a Star basis on my NCAA Football Best Bets and 210-168-2 (56%) on my Strong Opinions, and I have gone 1131-853-42 (57%) on a Star basis on my NCAA Football Best Bets and 327-266-8 (55%) on my Strong Opinions over the last 10 seasons. On NFL sides, I have gone 845-754-52 (53%) on a Star basis on my NFL Best Bets, and 159-132-5 (55%) on my Strong Opinions. In Basketball, I have gone 2168-1843 (54%) over the last 10 years.
Going forward, I expect to continue my success, and to win around 59% of my 4-Star bets, 56% of my 3-Star bets, 55% of my 2-Star bets, and 54% of my Strong Opinions in Football. In Basketball, I expect to win around 57% of my 4-Star bets, 55% of my 3- Star bet, and 54% of my 2-Star bets. A lot of handicappers use situational analysis and math models in their handicapping but few, if any, of them have studied the predictability of their methods, as I have, or found a realistic way of combining their methods for an overall measure of predicted success on every game.
Bob Stoll of Dr. Bob Sports can be reached at DrBobSports.com.
Handicapping Theory (2/3) Technical Analysis and Team Trends
By: Dr. Bob
Fundamental Analysis: Fundamental analysis is the old fashioned way of handicapping. Fundamentalists look at matchups or try to envision how a game will play out. They study the strengths and the weaknesses of teams and try to determine if one team has a significant advantage over another based on their ability to exploit a team's weakness with their strength. For example, a fundamental analyst might claim that, 'Denver is a great running team and they should have no problem running against a team like Cincinnati that has trouble stopping the run.' Or that, 'Carolina's All-Pro defensive end will take advantage of their opponents rookie left tackle and be in the quarterback's face all day.'
Such a style of handicapping depends on a very keen knowledge of each team and their personnel. The problem with this sort of fundamental analysis is that most mis-matches in a game are already reflected in mathematical models, as well as in the point spread.
The key to fundamental analysis is finding statistical indicators that have led to point spread success, beyond the most obvious observations which everyone already knows, and odds makers have already taken into account. Handicapping based solely on fundamental analysis (which is what you'll see from 99% of touts and analysts) is lazy, inaccurate and unlikely to gain an edge over odds makers and/or other bettors. However, more nuanced fundamental analysis can be a useful addition to more comprehensive and predictive analytical models.
Technical Analysis: Technical analysis is the study of patterns and is based on the psychological ups and downs of teams as well as the psychological patterns of those that bet sports. Obviously teams have their ups and downs, and I use trends and situations to identify when teams are likely to play well and when they are not based on patterns that have lead to point spread success and failure in the past.
Team Trends: The most widely used and understood type of technical analysis is the study of team specific patterns which I simply call team trends. When I started handicapping back in the mid-80's team trends were an important part of selecting which teams I bet on. I would go back into my logs of results and study how teams performed at home and on the road, as a favorite or as an underdog, after a win or after a loss, and under other circumstances. I found that the most insightful team trends were the ones that were based on recent performances and explained how teams performed after good and bad performances. For instance, the 49ers were 48-17 ATS (Against The Spread) from 1981 through 1997 when they lost straight up and failed to cover the point spread in their previous game. What this told me is that the 49ers had a strong tendency to be more focused after a poor outing. This is a trend that worked for many years despite coaching and player changes over the years. Of course, the 49ers have basically had the same type of team over all of those years, and the 49ers' tradition started under Bill Walsh in the early 80's has been handed down through the generations of players. It also helped that San Francisco had only two quarterbacks during those 17 years, and that both Joe Montana and Steve Young had the personality types that made them perform better after poor outings.
Most of the personality of a team comes from the head coach, and I have noticed that patterns follow coaches from team to team. For instance, Jon Gruden's teams have a tendency to play well as a favorite after a loss (10-5 ATS from 1998 through 2001 in Oakland and 17-10-2 ATS from 2002 through 2008 with Tampa Bay) and poorly when favored by three or more the week following a victory (5-15 ATS with Oakland and 9-17 ATS with Tampa Bay).
On the other hand, there are some types of team trends that simply do not predict what will happen in the future. Over the past 10 years, I have studied the results of all statistically significant team trends that I have used in my game notes and tallied to results, broken down by the type of trend it is. The type of team trends that were the best indicators are what I call personality trends, which are the trends that explain how teams react to recent performances, such as how a team performs after a win or a loss, or after two straight spread wins, or after allowing 28 points or more (like the ones I used in the examples above). Certain types of team trends don't work at all, such as series history trends or trends that deal with a specific game number. A series history trend is a trend that states that Team A has covered 10 straight times against Team B. I have found that regardless of how many times in a row a team has covered against another team, the chance that they cover in the next meeting remains unaffected. A trend that says that Team A has gone 13-1 in their second road game of the season or is 8-0 in week number 5 doesn't make any sense and does not have any value in predicting the future. I know what types of trends tend to work and to what degree they work.
While the use of team trends worked very well during the 80's and through the mid-90's, the advent of free agency and the constant changing of head coaches in the league changed the personalities of teams every few years, making previous patterns of these teams meaningless in most cases. I tend to shy away from most team oriented trends unless the head coach or core of star players has been intact over the term of the trend. I certainly wouldn't pay much attention to a Carolina Panthers trend that included games prior to the arrival of head coach John Fox, who changed the personality of that team. On the other hand, longer term trends of the Philadelphia Eagles do have some validity due to the long tenure of head coach Andy Reid - even though the personnel have changed over the years.
Team trends can still be a very effective handicapping tool, but I do not use` team trends that no longer explain the personality of a team. From a statisticians' point of view, a trend is basically a sample of games taken from a pool of results. When the pool from which the sample was taken changes, the sample of games is no longer representative of that pool and should thus not be used as a forecasting tool. Thus, team trends work best with teams that have had the same coach or core group of players for at least as far as the trend goes back.
Situational Trends: As the use of team trends became more limited because of free agency and coaching changes, I began looking for patterns that explained the results of all teams that were in the same set of circumstances. For instance, how did all teams perform following consecutive games in which they allowed less than 10 points? Or, what is the record of Monday night home underdogs? These league-wide patterns are referred to as situational trends. I have found that situational trends are better indicators of future point spread results than team trends, because team specific changes (such as coaching changes and free agency) have little effect on league wide patterns. The patterns that exist in the NFL and in college football have existed for years and are based on the psychological ups and downs that exist in all teams and in the wagering habits of the betting public.
While all of my situations deal with the patterns that exist in team performance, some of them also are enhanced by the betting patterns of the public, who are influenced by more recent performances of a team. A lot of the situations that I use deal with playing on teams that have been playing below expectation (bounce-back situations) and playing against those that are playing well in recent weeks (letdown situations). Bounce-back and letdown situations work partly because the betting public overreacts to a string of good and bad performances and bets accordingly. A team may become out of favor after a couple of terrible performances while other teams may get more support than warranted from a couple of very good performances. Since the point spread is as much a measure of public perception as it is a projection of a median outcome, the point spread gets over-adjusted in conjunction with the public's fear of betting on a team on the slide or with their eagerness to bet on a team playing especially well in recent weeks.
This sort of betting behavior based on short-term results gives the smart player line value and that is why these sort of bounce-back or letdown situations produce good results. For instance, NFL teams that win back-to-back games straight up as an underdog are just 52-72-3 ATS (Against The Spread) in their next game if they are on the road and not getting more than 7 points (since 1981), including 25-51-2 ATS if visiting a non-divisional opponent.
There are a couple of reasons for this. First, teams that win back-to-back games as an underdog tend to get more support from the betting public and as a result the point spread deviates from a realistic point spread to a line that represents current public perception of the "hot" team. At the same time, the team that has just won back-to-back upsets generally is not quite as hungry, as the coach has less ammunition to motivate them with. Also, the coaching staff is generally reluctant to change the offensive and defensive schemes that produced the two victories. Their next opponent, however, has two weeks of films to figure out how to find a weakness in the schemes that have been so successful and these opposing players and coaches will prepare for the game against the "hot" team with more focus because of their recent success. So, not only is the "hot" team in this situation due to play at a lower level but the point spread has also moved in our favor to create value because the betting public is generally afraid to bet against a "hot" team. Playing against a non-divisional opponent gives the team off two upset wins even less reason to get fired up. This situation does not work nearly as well when playing against home teams off back-to-back upset wins because it is easier to maintain a high level of intensity in front of the home fans.
Not all situations are in the bounce-back or letdown mode and take advantage of misguided public perception and the natural fluctuations of team performance. There are also what I call momentum situations and these deal with playing on teams that are playing well and playing against teams that are playing poorly. For instance, in the NFL home underdogs are 193-136-10 ATS if they won straight up as an underdog the previous week. Bad teams in the NFL generally lack the confidence to beat a good team and there is nothing like an upset win to boost confidence. The confidence of winning as an underdog is enhanced by playing in front of the home fans and thus creates a good momentum situation. In general, the NFL is a contrary league, meaning that most of the situations involve going against teams that have been playing well and going with teams that have been struggling. College football, meanwhile, is more of a momentum sport and many more of the good technical situations in college football involve playing on a team that has been playing above expectations.
Does Technical Analysis Work? Technical analysis has come under scrutiny by fundamental handicappers and some sports bettors due to the fact that anybody searching a database randomly for patterns will find situations that have produced very good results. However, the key is to look for situations that make sense. I don't use trends such as "The Steelers are 13-2 in week number 7" (Do they actually know that week 7 is their week and gain confidence from it?) or "bet on home dogs from +2 to +4 if it's a weeknight MAC game" (the more narrow the point spread range is the more likely it is a random occurrence and not a true indicator of a real pattern).
So how can I be sure that technical analysis works? At the beginning of each year, I make a list of the situational angles that I think are meaningful (they are all easily statistically significant). At the end of the year, I tally the results of these angles. In the last 10 years of doing this, I have found that the situational angles that I use (remember, if your angles don't make sense they are not going to hold up as well) have won at a profitable rate of 55%, and that the situations with a higher statistical significance (i.e. a higher t-value) have proven to be even more predictive.
Many handicappers tend to back-fit past data by adding more and more factors (parameters) to a situation until they have a very high percentage angle (but also a much smaller sample size). However, my research has shown that a situation's predictability is sacrificed with each parameter added to derive that situation. For instance, a situation with a record of 50-20 (71%) that is derived using 10 factors isn't as predictive as the 59% home underdog situation that I presented above, which has just 4 parameters (this game home, this game dog, won last game, dog last game) and a much larger sample size. It's easy to find a very high-percentage situation if you use an unlimited number of parameters to get to that situation, but all that will result in is a situation that explains what has happened rather than something that helps predict what will happen.
My research, and the theories of statistics, shows that more predictive angles have fewer factors and a larger sample size, rather than a smaller sample situation with a high winning percentage that was derived by using too many parameters. Further research I did in the Summer of 2004 (which I update each summer) enables me to accurately assess a situation's future performance based on the win percentage, sample size, number of parameters and more recent performance (i.e. record of the angle over the past 3 seasons). That research led to a more realistic use of situational analysis than I've employed in the past. For instance, I can now tell you that a situation with a record of 140-60-5 ATS that uses 6 parameters has a 56.8% chance of winning the next time it applies if the line is otherwise fair according to my metrics. Having a realistic expectation of a situation's value has helped my overall analysis immensely, and I will continue to devote time each summer to update the research on the predictability of my situational analysis.
Remember, just because a situation is 70% over 200 games in the past does not mean that it will win 70% of the time in the future. A 140-60 situational trend is simply a sample of 200 games selected from a population consisting of all NFL games. Since the NFL is constantly changing (although the league as a whole doesn't change nearly as quickly as most individual teams do), the results of the same situation in the future will not fully reflect the past. Also, by definition, a statistically significant trend has a 5% probability of being caused by no more than chance variation, and the record of those trends can be expected to be 50% as a whole, bringing down the overall percentage of all significant trends. There is also going to be a certain level of back-fitting involved in finding a situation, which also lowers the future percentage of the situation. Of course, the better the record, the greater number of games in the sample, and the fewer parameters there are in an angle the more likely that the situation is real and not just random.
Bob Stoll of Dr. Bob Sports can be reached at DrBobSports.com.
Handicapping Theory (1/3) Model Handicapping
By: Dr. Bob
Handicapping Theory There are three general theories of how a bettor can gain an edge handicapping sports: Model Handicapping, Fundamental Analysis and Technical Analysis. In this three-part article, I explain each of these theories independently, and how I combine them to produce my Best Bets.
Model Handicapping The core of my handicapping comes from the mathematical model I have built which predicts the results of games more accurately than the public or Las Vegas odds makers. Less sophisticated simulators that try to come up with a formula to predict future games tend to make the same mistake; they use regression analysis to find the correlation between different statistics and point differential. While that exercise is very useful for explaining which statistics impact a game's result, regression is not necessarily useful in using past statistical averages to predict future results since some important statistics simply don't correlate very highly to the future. For example, turnovers are the number one factor in point differential in football, but turnovers are also the least predictable statistic. A model that is based on regression analysis will weigh turnovers very highly, but since past turnovers do not correlate highly with future turnovers such models will over-weigh the affect of past turnovers - creating a model that is good at explaining what has happened but not very good at predicting what will happen.
Fumbles in particular are almost 90% due to variance - that is to say that historically, if you took all of the teams that fumbled a lot over the first 8 games of a season, and all of the teams that fumbled very little over the first 8 games, those two groups of teams fumbled at a similar rate over their last 4 games. In other words, fumbles are almost completely random, and if 119 teams play 8 games each, you'll have some team with lots of fumbles and some teams with almost no fumbles, but going forward, those statistics are not predictive of future performance. In other words, when the talking heads on ESPN praise teams that 'hold on to the football' and criticize teams with 'fumbilitis,' one must realize that such labels are just fooling you with randomness, and that in future games, the 'hold on to the football' teams will not necessarily fumble less than the 'fumbilitis' teams. (Of course, when this happens, the talking heads then say, "Iowa fumbled 10 times in the first 5 games, but has only fumbled once in the 5 games since then. They have learned how to take care of the football!") This is just the most obvious of literally hundreds of different metrics which are factored into my mathematical model, and is one of the reasons that my model is much better than regressive models and has a consistent, winning track record to prove it.
My math model incorporates the predictability of past statistics to future games and uses each team's compensated statistics rather than their raw stats, which adds to the accuracy of my prediction. Compensated statistics are derived by comparing a team's statistics to the statistics of the opponents that they have faced.
For instance, if Oregon is averaging 3.6 yards per carry, and Rutgers is averaging 4.0 yards per carry, but Oregon's opponents (when adjusted for schedule strength) only project to allow a combined 3.4 ypc against an average opponent, and Rutgers' opponents project to allow a combined 4.2 ypc against an average opponent, then compensated statistical analysis (which I have tested over a sample size of tens of thousands of games) predicts that Oregon is actually likely to fare better rushing against an average run defense than would Rutgers, despite the fact that Rutgers is running at a rate of 4.0 ypc to Oregon's 3.6. Using compensated statistics in combination with the predictive nature of each statistic used in my model produces an accurate measure of the true differences between two teams future performances - not the difference between their past performances.
I also adjust my projected numbers based on current personnel for each team and those extra hours of statistical work have paid off handsomely over the years (and I get better each year at making those adjustments). A lot of my edge comes from some complex defensive player analysis models which I have built to evaluate the effects of defensive injuries. Most other handicappers - even the most sophisticated ones - all but ignore defense, and their lines are often off as a result. I also remove meaningless plays from my data set such as kneel downs at the end of a half or game and quarterback spikes, so the game statistics that I use are more representative of a team's performance than the statistics used by other handicappers who take a lazier approach and just plug in box scores.
I have been using my current NFL model for 8 years, and my NCAA math model since 2001 (with major upgrades in 2005). As you can see in my Past Performance (link to Past Performance Page) section, the results have been fantastic. The NFL model begins in Week 1 of each year (since I have so much reliable data from previous years), while the NCAA model kicks in Week 5.
Determining Value One of the critical advantages of Model Handicapping is that it allows me to quantify my edge. That is to say, that over many years my model not only identifies advantageous lines, but also can give me a rough percentage estimate of how likely a given team is to cover. We delve much further into the significance of this in the Investment Strategies (link to MM Summary Essay) article, but briefly, quantifying my edge allows bettors to adjust my bet sizes for optimal bankroll growth, which allows my customers to make more money.
It takes years of careful tweaking and analysis to really determine how much value each point of difference between a bettor's own lines and Vegas' lines is worth. I have used statistical software to create a regression equation predicting home team spread result as a function of the line differential of my power ratings/math model from the actual line. For instance, I have 10 years using my NFL math model and the equation to predict the chance that the home team covers the spread is .505 + 0.0128xLD, where LD is the line differential between my math model prediction and the line. So, for every point differential, I can add 1.28% to my chance of winning. Each NFL game has an unknown hypothetical 'perfect' line where each side would cover exactly 50% of the time, and I try my best to arrive at that line. If I had 'perfect' lines, then I would have about a 3% advantage per point differential between my lines and the Vegas' lines. Over the last two decades, my advantage has been 1.28% per point, meaning that I have clearly demonstrated that my lines are superior to Vegas' lines, but that they aren't 'perfect' yet. I spend the majority of each summer researching my methods and fine tuning my analysis, and my lines have become more and more accurate each year.
Remember, it doesn't matter how much of a differential there is between your ratings/math model if your line is not proven to be better than the actual point spread, as my lines have!
Power Ratings Many handicappers have a set of ratings, most often referred to as power ratings, that gauge the overall strength of each team in comparison to every other team. They then take the difference in ratings between two teams as the predicted point differential between the teams if they met on a neutral field. Of course, teams don't usually meet on a neutral field so points are added to the home team to compensate for the advantage that most teams have playing at home. The home field advantage can be a set amount for all teams (such as 2.5 or 3 points in the NFL) or can vary from team to team depending on their individual variance in their level of play at home and on the road.
While the concept of power ratings is rather simple, it is very difficult to come up with a set of accurate ratings. The problem with most power ratings methods is that the ratings are generated using some sort of mathematical process based on the past performance of each team and the level of opposition that they have faced. An example of this is the Sagarin Ratings seen in USA Today each week. I've talked to many amateur handicappers that use the Sagarin Ratings to figure out if the point spread is too high or low on a particular game. What is important to remember is that the Sagarin Ratings, and any other mathematically produced set of ratings, explain what has already happened rather than what will happen. In other words, while it is true that these ratings accurately reflect the difference in the performance of each team up to that point of the season they are not a predictive tool to be used to forecast the future performance level of teams, which is what we are truly interested in as handicappers.
If beating the point spread were as easy as picking up the Tuesday USA Today, checking the Sagarin Ratings and making your wagers based on that, then everyone would be winning and sports books would all be out of business. Obviously, that is not the case. So, while the Sagarin Ratings can be used to see how teams have performed up to that point of the season, do not depend on them to forecast how teams will perform in their next game.
Power ratings are typically based off of the final scores of games - in football, there is a lot of 'noise' and 'variance' in scoring, and points are not nearly as useful for predicting the outcomes of games. Furthermore, power ratings which reduce every team to a single number ignore the enormous importance of matchups. If Texas Tech and Georgia Tech have similarly rated offenses, then you would expect them to fair similarly against a defense that had an average rating across the board in all defensive metrics. However, against a defense with an average overall rating, but on a more specific level, with very high run-defense ratings (allowing 3.1 ypc against opponents who combine for an adjusted 4.5 ypc) and very bad pass-defense ratings (allowing 9.8 ypa against opponents who combine for an adjusted 6.4 ypa), you would expect Texas Tech's pass-heavy offense to fair comparably better than Georgia Tech's run-heavy offense, even though the two offenses are rated similarly overall. Obviously analyzing matchups is much deeper and more complex than this, and often gets into very technical data concerning advantages at individual positions, but this simple example illustrates the overall concept of how power ratings do not factor matchups.
Bob Stoll of Dr. Bob Sports can be reached at DrBobSports.com.